Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1808.10582	2	1808	2018-11-21T09:36:47	Effect of Polarization Field on the Photocurrent Characteristics of GaN/AlN Multi-Quantum-Well Avalanche Photodiode	[ "physics.app-ph" ]	Polarization is an important property of GaN/AlN multi-quantum-well (MQW) avalanche diode (MAPD) but has been ignored in recent analyses of MAPD to simplify the Monte Carlo simulation. Here, the photocurrent characteristics of GaN/AlN MAPD are investigated to understand the role of polarization field in the MQW structure. Carrier multiplication in AlN/GaN MAPD is found as a result of interfacial impact ionization not much helped from external field but instead considerably contributed by the polarization field. In addition, the movement of ionized electrons out of quantum well is proved as Fowler-Nordheim tunneling process helped by the polarization field in AlN barrier. Furthermore, the transport of ionized electrons through MQW structure is found influenced by the reverse polarization field in GaN layer, which could be suppressed by the external electric field. With all the three effects, the quick photocurrent increase of GaN/AlN MAPD is ascribed to the efficient transport of interfacial-ionized electrons step by step out of MQW at high voltages, which is in big difference from conventional APD due to avalanche breakdown.	physics.app-ph	physics	Effect of Polarization Field on the Photocurrent Characteristics of GaN/AlN Multi-Quantum-Well Avalanche Photodiode Wangping Wang,1,a) Qian Li,1 Xingzhao Wu,2 Jianbin Kang,1 Yi Luo,2 Mo Li,1 and Lai Wang2 1Microsystem and Terahertz Research Center, China Academy of Engineering Physics, Chengdu 610299, People's Republic of China 2Tsinghua National Laboratory for Information Science and Technology, Department of Electronic Engineering, Tsinghua University, Beijing 100084, People's Republic of China Polarization is an important property of GaN/AlN multi-quantum-well (MQW) avalanche diode (MAPD) but has been ignored in recent analyses of MAPD to simplify the Monte Carlo simulation. Here, the photocurrent characteristics of GaN/AlN MAPD are investigated to understand the role of polarization field in the MQW structure. Carrier multiplication in AlN/GaN MAPD is found as a result of interfacial impact ionization not much helped from external field but instead considerably contributed by the polarization field. In addition, the movement of ionized electrons out of quantum well is proved as Fowler-Nordheim tunneling process helped by the polarization field in AlN barrier. Furthermore, the transport of ionized electrons through MQW structure is found influenced by the reverse polarization field in GaN layer, which could be suppressed by the external electric field. With all the three effects, the quick photocurrent increase of GaN/AlN MAPD is ascribed to the efficient transport of interfacial-ionized electrons step by step out of MQW at high voltages, which is in big difference from conventional APD due to avalanche breakdown. _____________________________ a) Electronic mail: [email protected]. b) Electronic mail: [email protected]. 1 Avalanche photodiode has been widely used for weak light detection.1-3 Geiger mode avalanche photodiode is capable of single photon detection but suffered from after pulse in avalanche-quenching operation.1-3 The avalanche-quenching circuit is not needed for avalanche photodiode with photomultiplier tube (PMT) like multiplication process, which could be realized when the electron and hole ionization rate are in large difference.4 Greatly improved ionization rate ratio has been demonstrated in earlier AlGaAs/GaAs muti-quantum-well (MQW) avalanche photodiode (MAPD) and is attributed to the enhanced electron impact ionization by the conduction band edge discontinuity at heterointerface (△Ec).5-6 The △Ec is greatly increased in GaN/AlN MQW structure and PMT-like high avalanche gain has been demonstrated for a 20 periods GaN/AlN MAPD,7 which implies the advantage of III-nitride heterostructure for solid state PMT application. Ensemble Monte Carlo simulation has been carried out to obtain the ionization coefficient for GaN/AlN MAPD.7-10 However, the polarization fields in AlN and GaN layer were not considered and the interface scattering was ignored to simplify the simulation model of GaN/AlN MAPD.7-10 It was predicted that besides the conduction band edge discontinuity, the electron ionization rate could also be enhanced by the polarization fields in III-nitride MAPD.11 Thus the impact ionization process could not be well modeled if the polarization field in GaN/AlN MQW structure is not considered in the Monte Carlo simulation. Moreover, the directions of polarization fields in AlN and GaN layer are contrary to each other.11 Therefore the electrons transport through the MQW structure are expected influenced by the reverse polarization field in AlN and GaN layer. To our knowledge, the carrier transport through polarized MQW structure is not well discussed in previous analyses of III-nitride MAPD.7-11 In this work, the photocurrent characteristic of GaN/AlN MAPD is investigated to understand the effects of polarization field on electron ionization and electron transport. At first, the GaN/AlN MAPD is theoretically investigated to get the polarization field modified energy band profile. Then a conventional APD (CAPD) is designed with similar energy band profile as MAPD but without the MQW structure. The photocurrent of MAPD is experimentally compared to that of CAPD for illustrating the effect of polarized MQW on electron ionization. Finally, the carrier transport through polarized MQW is analyzed with the rate of increase and noise feature in MAPD photocurrent. The structure of GaN/AlN MAPD is illustrated in Fig. 1(a) and could be simply depicted as 20 AlN (10 nm)/GaN (10 nm) periodic layers inserted into the impact ionization region of p-i-p-i-n structure. The structure of 2 homojunction CAPD is shown in Fig. 1(b) and is further described later. The MAPD and CAPD are both made of the same double-mesa device structure and device fabricating parameters for better comparison. The role of polarization field on conduction band structure of GaN (10nm)/AlN (10 nm) MAPD is presented in Fig. 2(a). The calculation of band structure is based on self-consistent solution of Schrödinger-Poisson equations. The polarization field strength in AlN and GaN layer is obtained as 3.2 MV/cm and -3.2 MV/cm respectively, based on the software package of Crosslight APSYS with Coulomb screening ratio set as 0.5. For the MAPD intentionally ignoring the polarization field, the energy band diagram is shown as tilted profile dropped down by the built-in electric field and external electric field. In contrast, for the MAPD considering the polarization field, uplifted energy band profile is found for the MAPD structure, with nearly-flat energy band profile in the light-absorption layer. Note that the polarization field in GaN layer is in reverse direction contrary to the external electric field and built-in electric field of MAPD structure. As shown in Fig. 2(b), the electric field strength in the light-absorption layer is changed from positive to negative by the internal polarization field, which explains the nearly-flat band profile in the absorption layer. In the latter part of the paper, a comparative CAPD is designed with similar band profile as polarized MAPD but without the MQW structure to fully illustrate the effect of polarized MQW. In consideration of the nearly-flat band profile in the MAPD absorption layer, the CAPD should be designed as p-i-n structure but not as the same p-i-p-i-n structure of MAPD. The comparison could be made since the nearly-flat band profile in absorption layer of MAPD is found very stable regardless of the bias voltage. In Fig. 3(a), the photocurrent of GaN/AlN MAPD is compared with that of GaN CAPD to obtain the role of polarization field on ionization process. The CAPD is designed with intrinsic layer thickness as 20 GaN periodic layers, or half of the MQW thickness in MAPD. At same bias voltage, the external field strength in MAPD is about half of that in CAPD, but the photocurrent of MAPD is orders larger than that of CAPD. This is an indication of the structure-induced carrier multiplication in MAPD. Furthermore, increase of more than one order is observed in the photocurrent of MAPD applied to 10 V, while negligible increase is shown in the photocurrent of CAPD applied to 60 V. Hence the carrier multiplication could be started without much help from the external electric field on MAPD. In GaN/AlN MAPD, aside from external field, the electrons are also heated by the 3.2 MV/cm polarization field in AlN layer, 2.0 eV △ Ec at GaN/AlN interface, less than 0.1 MV/cm build-in electric field of p-i-p-i-n doping, but are cooled down by the -3.2 MV/cm polarization field in GaN layer. The build-in 3 electric field is negligible compare to the large polarization field in GaN/AlN MAPD. Therefore, at low voltages the enhanced impact ionization in MAPD is mainly resulted from the combined action of polarization field in AlN layer and △ Ec at GaN/AlN interface. Accordingly, the impact ionization process in MAPD could be described as interfacial impact ionization at GaN/AlN interface. Note that the energy for electrons to trigger ionization events is about 3.7 eV in GaN layer.9 For electrons from AlN to GaN at 60 V high voltage, the gained electron energy is 1.5 eV from 1.5 MV/cm external field and 2.0 eV from the GaN/AlN △ Ec, which combined together is still smaller than the 3.7 eV ionization energy. Hence unlike CAPD in which the impact ionization is fully due to external electric field, the interfacial impact ionization in MAPD is considerably contributed by the polarization field. There is no external field induced impact ionization in GaN well layer of MAPD, for the quantum well layer is designed very thin to allow just one impact ionization triggered here. The polarization field is not only beneficial to the electron ionization in MAPD, but is also helpful to the movement of electrons out of quantum well. As shown in Fig. 4, electron trapping at GaN/AlN interface is expected due to the nearly-flat band profile in absorption layer and tilted-up band profile in quantum well layer. Considering that the band profile of AlN layer is greatly dropped down by the polarization field, the electrons may transport through the AlN barrier in Fowler-Nordheim (FN) tunneling process. The FN tunneling current is proportional to ε2exp(−A/ε), where ε is electric field on tunneling barrier.12 Accordingly, the FN tunneling probability is greatly enhanced by the 3.2 MV/cm polarization field in AlN barrier. Interestingly, there is evidence for the proposed tunneling transport in the photocurrent characteristic of Fig. 3(b). Negative differential conductance (NDC) is clearly observed in MAPD photocurrent curve, with peak current about three times of valley current and peak voltage around 11 V, and is confirmed with repeated measurements of many MAPD chips. For the photocurrent of MAPD, the impact ionization is only triggered by the photoelectrons injected into MQW structure. The photoelectron injection efficiency could be greatly enhanced when photoelectrons resonant tunnel through the AlN barriers as indicated by the arrow in Fig. 4. Confined states in each GaN quantum well is obtained by self-consistently solving Schrödinger equation for the whole MQW structure. As shown in the inset figures of Fig. 4, for the 1st GaN quantum well in equilibrium condition, discrete energy levels are found above conduction band edge of light-absorption layer (Ecphoto). These discrete levels drop down and in alignment with Ecphoto around 10V bias, which is close to the observed NDC peak voltage. Hence, the NDC feature is attributed to the triangular double barrier resonant tunneling of photoelectrons from absorption layer into MQW structure. 4 The resonant tunneling is possible because the confined states of GaN quantum well are greatly lifted up by the polarization field. It is expected that the ionized electrons in each GaN well layer is step by step multiplied at following GaN/AlN interface. The transport of ionized electrons is influenced by the reverse polarization field in GaN layer and is investigated with different rates of increase of the MAPD photocurrent in Fig. 3(b). As shown in the linear scale photocurrent curve, there are two rates of increase of the MAPD photocurrent: slow increase labeled as rate 1 and quick increase labeled as rate 2, with transition voltage around 33 V between rates 1, 2. Unlike the sharp increase of CAPD photocurrent due to avalanche breakdown, the quick increase of MAPD photocurrent is observed to be saturated above 65 V, hence rules out the possibility of avalanche breakdown in MAPD. Moreover, it is observed that feature of large noise is appeared in rate 1 part of MAPD photocurrent curve and is gradually reduced in rate 2 part of photocurrent with increasing bias voltage. The noise feature is confirmed with repeated measurements in many chips of GaN/AlN MAPD. In contrast, no such noise feature is observed in the photocurrent curve of CAPD. Thus the noise characteristic of MAPD is correlated with the polarized MQW structure and is not related with device processing parameters or measurement equipment. In view of the noise feature, the different rate of increase of the MAPD photocurrent is explained with arrow- indicated electron steam through the polarized MQW structure in Fig.5. There are two kinds of electrons inside each quantum well layer: interfacial-ionized electrons created in current quantum well and drifting electrons from previous quantum well to current quantum well. As indicated by arrow 1, the interfacial-ionized electrons are partly trapped and scattered by the reverse polarization field in quantum well layer while the drifting electrons are scattered at previous and current AlN barriers. Note that the AlN barriers are uplifted in the band profile also due to the reverse polarization field in GaN layer. Hence the photocurrent is observed both of noisy feature and slow rate of increase below 33 V, because of the large carrier trapping and interface scattering for arrow 1 indicated electron stream. At high voltages, not only the reverse polarization field in GaN well layer is partly offset by the external field, but also the FN tunneling through AlN barrier is enhanced. Furthermore, as indicated by arrow 2, the drifting electrons are easily heated over AlN barrier at high voltage, thereby minimizing the influence of MQW structure on carrier transport. Thus quick increase of photocurrent is observed with gradually reduced noise above 33 V, thanks to the greatly reduced carrier trapping and interface scattering in MQW structure. As indicated by the dash arrow in Fig. 5, the AlN barrier is dropped just below the previous quantum well at 33 V, 5 which sets the transition voltage for arrows 1, 2 indicated electron stream and rates 1, 2 indicated photocurrent. From the above analysis, although the reverse polarization field in GaN layer enhances the carrier trapping and interface scattering in MQW structure, the multiplication gain of GaN/AlN MAPD is influenced negligibly by the polarization field in GaN layer. In consideration of the photocurrent characteristics in Fig. 3(a), the best detection voltage of GaN/AlN MAPD is set around 60 V, when the reverse polarization field is greatly suppressed and the interfacial-ionized electrons are efficiently transported step by step out of MQW to collector contact. Moreover, the polarization field in GaN layer is found inverse proportional to its layer thickness.11 Accordingly, the GaN/AlN MAPD could be further optimized to reduce the reverse polarization field by adopting thick GaN well layer. The carrier trapping and interface scattering noise of MAPD are also reduced with this optimization. In summary, the photocurrent characteristic of GaN/AlN MAPD has been theoretically and experimentally investigated. The polarization field in AlN layer is not only confirmed of large contribution to the enhanced ionization rate in MAPD but is also helpful to the movement of electrons out of quantum well layer. The polarization field in GaN layer has influence on the transport of ionized electrons through MQW structure but could be suppressed by the external electric field. Although the avalanche gain is currently smaller than CAPD, the GaN/AlN MAPD has unique advantages over CAPD. Besides the demonstrated PMT-like detection mode of MAPD without avalanche-quenching circuits, the GaN/AlN MAPD is expected of better responsivity spectral cutoff over CAPD. Due to the reverse polarization field in GaN layer, as shown in Fig. 2(b), the light absorption layer in GaN/AlN MAPD is found with electric field strength lower than 5×104 V/cm, which is orders smaller than the typical 106 V/cm field strength in the absorption layer of CAPD. Therefore, the Franz-keldysh effect is minimized in the absorption layer of III-nitride MAPD, which should improve the responsivity spectral performance of III-nitride MAPD at high detection bias. ACKNOWLEDGMENTS This work is supported by the National Natural Science Foundation of China (Grant Nos. 61574082). REFERENCES 1R. H. Hadfield, Nature Photon. 3, 696 (2009). 2S. Nikzad, M. Hoenk, A. D. Jewell, J. J. Hennessy, A. G. Carver, T. J. Jones, T. M. Goodsall, E. T. Hamden, P. Suvarna, J. Bulmer, F. Shahedipour-Sandvik, E. Charbon, P. Padmanabhan, B. Hancock, and L. D. Bell, Sensors 6 16, 927 (2016). 3D. M. Theodore and P. Roberto, Rep. Prog. Phys. 80, 106501 (2017). 4J. David, Nature Photon. 10, 364 (2016). 5F. Capasso, W. T. Tsang, A. L. Hutchinson, and G. F. Williams, Appl. Phys. Lett. 40, 38 (1982). 6K. F. Brennan and I. I. J. Haralson, Superlattices Microstruct. 28, 77 (2000). 7J. Zheng, L. Wang, X. Wu, Z. Hao, C. Sun, B. Xiong, Y. Luo, Y. Han, J. Wang, H. Li, J. Brault, S. Matta, M. A. Khalfioui, J. Yan, T. Wei, Y. Zhang, and J. Wang, Appl. Phys. Lett. 109, 241105 (2016). 8J. Zheng, L. Wang, D. Yang, J. Yu, X. Meng, Z. Hao, C. Sun, B. Xiong, Y. Luo, Y. Han, J. Wang, H. Li, M. Li, and Q. Li, Sci. Rep. 6, 35978 (2016). 9J. Zheng, L. Wang, X. Wu, Z. Hao, C. Sun, B. Xiong, Y. Luo, Y. Han, J. Wang, H. Li, M. Li, J. Kang, and Q. Li, Appl. Phys. Express 10, 071002 (2017). 10J. Zheng, L. Wang, X. Wu, Z. Hao, C. Sun, B. Xiong, Y. Luo, Y. Han, J. Wang, H. Li, M. Li, J. Kang, and Q. Li, IEEE Photon. Technol. Lett. 29, 2187 (2017). 11B. Doshi, K. F. Brennan, R. Bicknell-Tassius, and F. Grunthaner, Appl. Phys. Lett. 73, 2784 (1998). 12S. M. Sze and K. K. Ng, Physics of Semiconductor Devices, 3rd ed. ( John Wiley & Sons, Inc., Hoboken, New Jersey, 2007) p. 437. 7 FIG.1 Material structure of (a) GaN/AlN MAPD and (b) GaN CAPD. FIG.2 (a) Conduction band profile of GaN/AlN MAPD at 60 V voltage with or without considering polarization field. (b) Electric field distribution in 60 V biased MAPD considering polarization field. 8 FIG.3 Photocurrent characteristics of MAPD and CAPD (a) in log-scale (b) in linear scale. The photocurrent is obtained by subtracting dark current from light current. FIG.4 Calculated conduction band profile for MAPD in 10 V voltage bias. Inset figures show the confined energy levels of the GaN quantum well (circle part of main figure) in equilibrium condition and in energy alignment condition. 9 FIG.5 Calculated conduction band profile for the GaN/AlN MAPD in 10 V, 33 V, and 60 V voltage bias. The arrows indicate electron stream through MQW structure. 10
1903.01985	1	1903	2019-02-26T14:50:00	Detailed Performance Loss Analysis of Silicon Solar Cells using High-Throughput Metrology Methods	[ "physics.app-ph", "physics.data-an" ]	In this work, novel, high-throughput metrology methods are used to perform a detailed performance loss analysis of approximately 400 industrial crystalline silicon solar cells, all coming from the same production line. The characterization sequence includes a non-destructive transfer length method (TLM) measurement technique featuring circular TLM structures hidden within the busbar region of the cells. It also includes a very fast external quantum efficiency and reflectance measurement technique. More traditional measurements, like illuminated current-voltage, Suns-VOC, and photoluminescence imaging are also used to carry out the loss analysis. The variance of the individual loss parameters and their impact on cell performance are investigated and quantified for this large group of industrial solar cells. Some important correlations between the measured loss parameters are found. The nature of these distributions and correlations provide important insights about loss mechanisms in a cell and help prioritize efforts to optimize the performance of the production line.	physics.app-ph	physics	Presented at the 4th World Conference on Photovoltaic Energy Conversion, WCPEC-4, June 14 2018, Waikoloa Village, Hawaii USA Detailed Performance Loss Analysis of Silicon Solar Cells using High- Throughput Metrology Methods Mohammad Jobayer Hossain1, Geoffrey Gregory2, Hardik Patel2, Siyu Guo3, Eric J. Schneller2,3, Andrew M. Gabor4, Zhihao Yang5, Adrienne L. Blum6, Kristopher O. Davis1,2,3 1CREOL, the College of Optics and Photonics, University of Central Florida, Orlando, FL, USA 2Department of Materials Science and Engineering, University of Central Florida, Orlando, FL, USA 3Florida Solar Energy Center, University of Central Florida, Cocoa, FL, USA 5School of Materials Science and Energy Engineering, Foshan University, Guangdong, China 4BrightSpot Automation, Westford, MA, USA 6 Sinton Instruments, Boulder, CO, USA Abstract -- In this work, novel, high-throughput metrology methods are used to perform a detailed performance loss analysis of ≈400 industrial crystalline silicon solar cells, all coming from the same production line. The characterization sequence includes a non-destructive transfer length method (TLM) measurement technique featuring circular TLM structures hidden within the busbar region of the cells. It also includes a very fast external quantum efficiency and reflectance measurement technique. More traditional measurements, like illuminated current-voltage, Suns- VOC, and photoluminescence imaging are also used to carry out the loss analysis. The variance of the individual loss parameters and their impact on cell performance are investigated and quantified for this large group of industrial solar cells. Some important correlations between the measured loss parameters are found. The nature of these distributions and correlations provide important insights about loss mechanisms in a cell and help prioritize efforts to optimize the performance of the production line. Index Terms -- performance analysis, performance loss, device manufacture, photovoltaic semiconductor device measurement, silicon, silicon devices. semiconductor cell, I. INTRODUCTION Analyzing and quantifying the various energy conversion losses occurring in photovoltaic (PV) cells and modules is fundamental to better understanding how these devices behave and engineering them to be better (e.g., more efficient, less expensive, more durable). Performance loss analysis normally revolves around decoupling loss mechanisms (e.g., optical, recombination, resistive) and quantifying their influence on cell performance. Due to the time consuming nature of the measurements involved, a detailed performance loss analysis is typically only carried out on small groups of solar cells, normally just in R&D settings [1]. However, with the advent of new high-throughput metrology techniques, it has become more convenient to perform these types of studies on larger groups of cells. In this work, we've carried out five different measurement techniques on ≈400 industrial crystalline silicon (c-Si) solar cells, all from the same production line, and will present a detailed performance loss analysis on this statistically relevant group of cells. The five measurement techniques include: (1) illuminated I-V at standard test conditions, a common method used to test and bin cells following their fabrication [2]; (2) Suns-VOC [3]; (3) photoluminescence (PL) imaging [4]; (4) high-speed quantum efficiency and reflectance spectroscopy [5]; and (5) non-destructive transfer length method (TLM) measurements [6]. II. EXPERIMENTAL DETAILS AND ANALYSIS TECHNIQUES The c-Si solar cells used in this work are industrial-sized, multicrystalline silicon cells featuring five busbars. These are industry standard Al-BSF (aluminum back surface field) cells, featuring isotropic texturing, a SiNx anti-reflection coating (ARC), phosphorus-doped n+ emitter on the front, screen- printed Ag front contacts, and a full area screen-printed Al rear contact that forms the p+ BSF upon firing. The only difference in these cells are the cTLM structures hidden within the busbars, which don't the cell processing or performance. influence Fig. 1. (Left) Illustration of the cell architecture used in this study and (right) image of the cTLM structures hidden within the busbars of the cells. Illuminated I-V curve measurements under standard test conditions and Suns-VOC measurements were performed using Sinton Instruments FCT-750 in-line cell tester. The FCT-750 is a production cell tester capable of measuring cells at line speeds of 3600 units per hour. This cell tester is unique in that it reports conventional cell test parameters (ISC, VOC, RS, RSH, power, efficiency, and fill factor) and supplements these parameters with an advanced Suns-VOC analysis and substrate doping measurement. The Suns-VOC analysis allows for a true RS measurement, pseudo I-V parameter measurement without the effects of RS, and carrier lifetime data. Open-circuit PL images were obtained with a BT Imaging LIS-R1. The imaging was carried out at two injection levels, 0.1 suns and 1sun, to obtain the spatial distribution of VOC over the cells. The laser in LIS- R1 shines the sample with 808 nm light. PL emission occurs due to the radiative recombination in the sample. The broad emission spectrum is filtered by a long wavelength pass filter having a cutoff wavelength of 920 nm before it reaches the detector. It makes sure that the part of the laser excitation reflected from the sample does not reach the detector. 3.01•1017 cm-2s-1 and 3.01•1016 cm-2s-1 photon flux were assumed for 1 sun and 0.1 sun, respectively. The experiment was carried out at room temperature. A spatial open-circuit voltage image Vxy can be obtained from the PL intensity image (IHxy) using the following equation [4]: 𝑉𝑉𝑥𝑥𝑥𝑥=𝑉𝑉𝑇𝑇 .log�𝐼𝐼𝐻𝐻𝐻𝐻𝐻𝐻−𝐵𝐵𝐻𝐻𝐻𝐻⋅𝐼𝐼𝐻𝐻 𝐶𝐶𝐻𝐻𝐻𝐻 � (1) Here VT is the thermal voltage. IH is the illumination intensity at which the PL image IHxy was taken. Bxy is a called a background calibration constant which accounts for diffusion limited carriers. It is generally derived from an additional short- circuit current image. However, since Bxy is negligible in comparison to IHxy, it has been ignored for simplicity. Cxy is a calibration constant independent of electrical bias and illumination conditions. It can be expressed as, 𝐶𝐶𝑥𝑥𝑥𝑥=𝐼𝐼𝐿𝐿𝑥𝑥𝑥𝑥⋅exp�𝑉𝑉𝑂𝑂𝑂𝑂𝑉𝑉𝑇𝑇� (2) Here ILxy is another open-circuit PL image at a lower illumination intensity (0.1 sun for this work). VOC is the measured open-circuit voltage. The high-speed EQE and reflectance measurements were performed with a customized Tau Science FlashQE system featuring an xy-gantry and integrating sphere. The FlashQE syatem has 41 LEDs whose emission ranges from 365 nm to 1280 nm with nonuniform spacing among them. The integrating sphere diffuses the light from all the LEDs and combines them into a spot size of 4 mm diameter. The LEDs turn ON and OFF at different frequencies (KHz range). Their individual contribution to the total photogenerated current is decoupled by taking its Fourier tansform and correlating with the corresponding ON-OFF frequency. In this case, the EQE and reflectance were measured on a 10•10 grid for each cell (i.e., 100 points on each cell). Measurement at each of these points takes approximately 1 second. The measurement locations were set to avoid the busbars and had the same amount of physical shading fraction for all the measurements in a cell. Different loss components of the short-circuit current density (JSC) [7] were extracted using the method outlined in [5], including: (a) front surface reflectance (JR-f); (b) escape reflectance (JR-esc) ; (c) absorption in the anti-reflection coating and loss in the emitter (Jloss-e); and (d) loss in the bulk and rear, or base (Jloss-b). and Fig. 2. EQE highlighting various JSC losses (a) front surface reflectance, (b) escape reflectance (c) emitter loss (d) bulk and rear surface loss. reflectance spectrum In the FlashQE measurement, light from the integrating sphere shines a spot on the solar cell. Part of it gets reflected from the front surface and gridlines. Busbars are avoided in the measurement process; therefor there is no reflection from them. The rest of the light enters into the cell and starts getting absorbed there. However, longer wavelength lights find it difficult to get absorbed. They are reflected from the rear side metallization and reach the detector escaping the cell. This kind of reflectance is called escape reflectance. Since escape reflectance occurs only in the longer wavelengths, the total reflection (sum of front and escape reflectance) starts increasing rapidly after certain point in longer wavelength region. Therefore a linear trendline beyond that point (≈1020 nm in this study) represents the front reflection in addition to all the reflections before that point. The difference between total and front reflection provides escape reflectance. The wavelength with minimum reflection tells about the thickness of the SiNx ARC as in equation 3. Considering first order reflection the thickness of ARC, 𝑡𝑡=𝜆𝜆min4𝑛𝑛 𝐼𝐼𝐼𝐼𝐼𝐼(𝜆𝜆)= 𝐸𝐸𝐸𝐸𝐸𝐸(𝜆𝜆) 1−𝑅𝑅𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 (3) (4) Here, λmin is the wavelength corresponding to minimum reflection, n is the refractive index of SiNx. Since the absorption in the antireflection coating is negligible, the internal quantum efficiency (IQE) can be calculated from the measured value of EQE and reflectance, Apart from the losses due to reflection and shading, parasitic absorptions also occur in the bulk, emitter and the rear surface of the solar cell. Emitter losses occurs in the short wavelengths when bulk and rear losses occur mostly in the long wavelengths. The losses in the emitter can be modelled as a hypothetical dead layer thickness Wde. With this assumption, IQE can be expressed as, 𝐼𝐼𝐼𝐼𝐼𝐼(𝜆𝜆)=1𝑘𝑘exp�−𝑊𝑊𝑑𝑑𝑑𝑑 𝐿𝐿𝑡𝑡(𝜆𝜆)� 𝐿𝐿𝑑𝑑𝑒𝑒𝑒𝑒 11+𝐿𝐿𝑡𝑡(𝜆𝜆) (5) Here, Leff and La are the effective diffusion length in the base and absorption length, respectively. k is a scaling factor. In this equation, Wde, Leff and k are the unknowns, which can be determined by a simple iterative process. It begins with a reasonable guess of the diffusion length. With this the slope and intercept of the ln(IQE(1+La/Leff)) vs. 1/La plot provides the value of Wde and k respectively. Based on the error present, the value of Leff is adjusted and another set of Wde and k are calculated. This iterative processes continues until it converges on the solution of the equation. Once all the parameters are obtained, they can be used for calculation of the individual losses. The dead layer thickness is valid for the shorter wavelengths. Thus, the total loss in the emitter can be divided into two parts, 𝐴𝐴𝑒𝑒,𝐼𝐼(𝜆𝜆)=1−𝐼𝐼𝐼𝐼𝐼𝐼(𝜆𝜆)⋅�1−𝐿𝐿𝑡𝑡(𝜆𝜆) 𝐿𝐿𝑑𝑑𝑒𝑒𝑒𝑒� 𝐴𝐴𝑒𝑒, 𝐼𝐼𝐼𝐼(𝜆𝜆)=1−exp�−𝑊𝑊𝑑𝑑𝑑𝑑 𝐿𝐿𝑡𝑡(𝜆𝜆)� 𝐼𝐼𝐼𝐼(𝜆𝜆)𝜙𝜙𝑖𝑖𝑛𝑛(𝜆𝜆) 𝑑𝑑𝜆𝜆 𝐽𝐽𝑠𝑠𝑠𝑠=𝑒𝑒∫ 1280 𝑛𝑛𝑛𝑛 365 𝑛𝑛𝑛𝑛 𝐼𝐼 The remaining parasitic losses can be attributed to the base and rear surface loss. JSC is an important parameter in determining cell performance. It can be determined from the following equation. (6a) (6b) (7) Here, e is the charge of an electron. The limits of the integrals are taken from 365 nm to 1280 nm, because this is the range of light the LEDs in the FlashQE emit. The non-destructive cTLM measurements were performed using a BrightSpot Automation ContactSpot-PRO system. The contact resistivity (ρC) and emitter sheet resistance (Rsheet) were extracted from the cTLM measurements using the technique in [6]. The ρC and Rsheet of c-Si solar cells is traditionally measured using TLM measurements, wherein the test structures are created by isolating strips of a cell (e.g., laser scribe then cleave) or by fabricating special test structures within a wafer. The former can be applied to industrial cells, but is destructive, and the latter is non-destructive, but cannot be used on industrial cells. The non-destructive technique used in this work relies on hiding circular TLM (cTLM) structures within the busbars of the cells, and therefore does not result in any additional shading of the cell, so it can be used on finished solar cells. III. RESULTS AND DISCUSSION Illuminated I-V, Suns-VOC, PL, high speed EQE + R, and cTLM measurements were all carried out for the ≈400 finished solar cells coming from the same production line. Histograms of some of this combination of measurements are shown in Figure 3. The distributions of the parameters can help quantify variance due to materials and processes. It can also provide insight into what the various parameters determined from loss mechanisms are limiting cell performance and help prioritize efforts to optimize performance of the production line. For example, histograms of the loss parameters determined from the EQE + R data show that loss in the bulk/rear of the cell (Jloss-b) is the primary limiting factor for JSC, followed by front surface reflectance (JR-f). Fig. 3. Histograms of the various parameters calculated from the illuminated I-V, Suns-VOC, PL, high speed EQE + R, and cTLM measurements. Note, RS is determined from the difference between the I-V and Suns-VOC curves. Doing all these measurements for a high volume cell group has provided an unique opportunity to investigate the proper causes behind losses in solar cells and the relationships among different performance parameters. The 1 sun open-circuit PL and JSC image of three different cells are illustrated in Figure 4, a good cell, an average cell, and a poorly performing cell, in terms of efficiency. For each cell, the positions and patterns of these localized areas of poor performance are virtually identical in both images due to the known reciprocity EQE and PL. of the device is limiting the overall τeff of the cells. Here, the current loss in the bulk and rear of the cell (Jloss-b) shows a correlation to τeff, whereas the current loss due to parasitic absorption in the SiNx ARC and recombination in the emitter (Jloss-e) does not. Additionally, and the effective base diffusion length (Leff) also has a strong correlation to τeff, further indicating that that recombination in the bulk and rear of the cell is limiting τeff and not recombination in the emitter. This makes sense considering these are all multi Al-BSF cells. Another example uses the Suns-VOC and cTLM measurements to separate three components of RS. In Figure 6(b), the lumped cell RS is correlated to Rsheet, (cTLM), ρb (Suns-VOC), and ρc (cTLM). The strength of the correlations indicates that the Rsheet of the emitter has the strongest influence, followed by the ρb of the wafer next. The ρc of these specific cells is too low to significantly affect the overall RS. 6(c). Another A correlation between Jloss-e and the emitter Rsheet is noticed in is noticed between Rsheet and dead layer thickness. Both Jloss-e and dead layer thickness was derived from the FlashQE data and account for losses in the emitter, while Rsheet was measured using cTLM method. correlation Figure similar PL images and spatially-resolved JSC of three different Fig. 4. cells, a good, average, and a poorly performing cell - the efficiency values for each are given on the left hand side of the figures. Note, the poorly performing cell has a different JSC scale. Efficiency is the most important parameter in solar cell production. The big data set coming from the I-V, Suns-VOC, cTLM, PL and FlashQE measurements of the ≈400 cells was investigated to find out a comprehensive picture of how different parameters affect the cell efficiency. The efficiency from the I-V measurements was compared with all other measured data. As depicted in the figure 5, a strong linear relationship is observed between the efficiency and J01 loss. A similar, but relatively weaker relationship is noticed between efficiency and J02. As expected, efficiency showed a linear dependence on the spatially-averaged JSC. The strong linear dependence on the spatial standard deviation of JSC shows a quantitative relationship between efficiency lost and within-wafer EQE variation, which in this case is driven by bulk and rear recombination (see the correlation between efficiency and Leff). The mean PL intensities at 0.1 sun and 1 sun and the mean Vxy derived from the PL images all show a strong correlation to efficiency. Interestingly, no correlation between efficiency and standard deviation of Vxy is observed. Another method of analyzing the data is to evaluate correlations between the numerous parameters measured with these different techniques, with a focus on relationships between parameters that can actually provide useful insight into manufacturing. For example, Figure 6(a) shows which region Fig. 5. Correlations between the cell efficiency and various parameters, including: (a) RS, J01, and J02 measured using I-V and Suns-VOC; (b) the mean JSC, within-cell standard deviation of JSC, and Leff measured with EQE + R; and (c) the mean PL intensity at 0.1 sun (yellow) and 1 sun (orange), mean Vxy, and within-wafer standard deviation of Vxy measured using PL imaging. REFERENCES [1] R. Evans, J. Dore, E. V. Voorthuysen, J. Zhu, and M. A. Green, "Data mining photovoltaic cell manufacturing data", in IEEE 40th Photovoltaic Specialist Conference (PVSC), 2014. [2] H. Field and A. M. Gabor, "Cell binning method analysis to minimize mismatch losses and performance variation in Si-based modules", in IEEE 29th Photovoltaic Specialists Conference (PVSC), 2002. [3] M. J. Kerr, A. Cuevas, and R. A. Sinton, "Generalized analysis of quasi-steady-state and transient decay open circuit voltage measurements", Journal of Applied Physics, vol. 91, no. 1, p. 399, 2002. [4] M. Glatthaar, J. Haunschild, R. Zeidler, M. Demant, J. Greulich, B. Michl, W. Warta, S. Rein, and R. Preu, "Evaluating luminescence based voltage images of silicon solar cells", Journal of Applied Physics, vol. 108, no. 1, p. 014501, 2010. [5] E. J. Schneller, K. Ogutman, S. Guo, W. V. Schoenfeld, and K. O. Davis, "Crystalline Silicon Device Loss Analysis Through Spatially Resolved Quantum Efficiency Measurements", IEEE Journal of Photovoltaics, vol. 7, no. 4, pp. 957 -- 965, 2017. [6] G. Gregory, A. Gabor, A. Anselmo, R. Janoch, Z. Yang, and K. O. Davis, "Non-Destructive Contact Resistivity Measurements on Solar Cells Using the Circular Transmission Line Method", in 44th IEEE Photovoltaic Specialist Conference, 2017. [7] M.J. Hossain, B. Tiwari, I. Bhattacharya, "Novel High Efficiency Quadruple Junction Solar Cell with Current Matching and Quantum Efficiency Simulations", Solar Energy, vol. 139, pp. 100-107, 2016. (a) Correlations between τeff at VMP, determined from Suns- Fig. 6. VOC, and Jloss-b, Jloss-e, and Leff, all determined from the EQE data. (b) Correlations between RS, determined from the difference in the I-V and Suns-VOC curves, and Rsheet (cTLM), ρb (Suns-VOC), and ρc (cTLM). (c) Correlations between Rsheet, determined from cTLM, and both Jloss- e and the dead layer thickness, both determined from the EQE data. IV. CONCLUSIONS Access to larger sets of device parameters can provide useful insight into specific loss mechanisms and be used for process control and specific defects can be identified, classified, and their impact quantified. In this work, we observed a variety of notable trends. Spatial variation of JSC within a cell results in lower efficiency, whereas spatial variation in VOC did not. J01 recombination had a stronger impact on cell efficiency than RS. Carrier lifetime has strong correlation with base-loss and diffusion length, but no correlation with emitter-loss. This shows that this group is limited by recombination in the bulk and rear. RS is primarily influenced by emitter Rsheet, followed by the wafer ρbulk. No correlation was found between RS and ρc. Rsheet has strong correlation with current loss in the emitter and the dead layer thickness, calculated from the QE at short wavelengths. ACKNOWLEDGEMENTS This work is supported by the U.S. Department of Energy SunShot Initiative under grant number DE-EE-0008155. View publication stats View publication stats
1905.09984	1	1905	2019-05-24T01:05:35	Experimental Realization of Acoustic Bianisotropic Gratings	[ "physics.app-ph" ]	Acoustic bianisotropic materials couple pressure and local particle velocity fields to simultaneously excite monopole and dipole scattering, which results in asymmetric wave transmission and reflection of airborne sound. In this work, we systematically realize an arbitrarily given bianisotropic coupling between the pressure and velocity fields for asymmetric wave propagation by an acoustic grating with inversion symmetry breaking. This acoustic bianisotropic grating is designed by optimizing the unit cells with a finite element method to achieve the desired scattering wavevectors determined by the bianisotropic induced asymmetric wave propagation. The symmetry and Bloch wavevectors in the reciprocal space resulted from the grating are analyzed, which match with the desired scattering wavevectors. The designed structures are fabricated for the experimental demonstration of the bianisotropic properties. The measured results match with the desired asymmetric wave scattering fields.	physics.app-ph	physics	Experimental Realization of Acoustic Bianisotropic Gratings Steven R. Craig1,†, Xiaoshi Su2,†, Andrew Norris2, and Chengzhi Shi1,3,* 1 Meta Acoustics Lab, GWW School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA 2 Department of Mechanical and Aerospace Engineering, Rutgers University, Piscataway, NJ 08854, USA 3 Parker H. Petit Institute for Bioengineering and Biosciences, Georgia Institute of Technology, Atlanta, GA, 30332, USA Acoustic bianisotropic materials couple pressure and local particle velocity fields to simultaneously excite monopole and dipole scattering, which results in asymmetric wave transmission and reflection of airborne sound. In this work, we systematically realize an arbitrarily given bianisotropic coupling between the pressure and velocity fields for asymmetric wave propagation by an acoustic grating with inversion symmetry breaking. This acoustic bianisotropic grating is designed by optimizing the unit cells with a finite element method to achieve the desired scattering wavevectors determined by the bianisotropic induced asymmetric wave propagation. The symmetry and Bloch wavevectors in the reciprocal space resulted from the grating are analyzed, which match with the desired scattering wavevectors. The designed structures are fabricated for the experimental demonstration of the bianisotropic properties. The measured results match with the desired asymmetric wave scattering fields. Manipulating acoustic wave propagation facilitates the development of novel applications in numerous fields, including lensing [1], noise control [2], and high-intensity focused ultrasound therapies [3, 4]. Due to the reversible nature of wave functions, classical acoustic waves propagate with symmetries in both time and space. Asymmetric transmission and reflection stem from breaking these symmetries for the realization of nonreciprocal acoustic propagation, inversion symmetry violation, or parity-time symmetry. A common way to achieve nonreciprocity in acoustics is by altering the frequency of incident waves using nonlinear materials and filtering unwanted frequencies with sonic crystals, enabling one-way acoustic propagation [5]. Using nonlinearities to achieve asymmetric wave propagation has also been demonstrated with active elements by coupling Helmholtz resonators with a piezoelectric material and a nonlinear circuit [6]. Alternative techniques to realize nonreciprocity have been accomplished with circular flow in a resonant ring cavity to introduce a momentum bias on the propagating wave [7]. This is an acoustic analog in electromagnetics, achieved by a magnetic field. Appling this approach of nonreciprocal circular flow to an array of cylindrical sonic crystals leads to the creation of a topologically protected edge state with one-way acoustic wave propagation, which is an acoustic analogy of the quantum Hall effect [8-10]. Besides breaking the time reversal symmetry with circular flow or synthetic angular momentum, topologically protected one-way edge states were also realized by breaking the inversion symmetry of phononic crystals that induces acoustic pseudospin, analogous to quantum spin Hall effect [11-13]. Generalizing higher dimensions leads to the exploration of acoustic Weyl points and Fermi arcs, resulting in to asymmetric wave the Zeeman Effect topological acoustics in to propagation in 3D [14-16]. In addition, asymmetric wave transmission has been observed in non-Hermitian structures with pure loss effects [17]. Non-Hermitian acoustic structures with parity-time symmetry, where the loss and gain materials were exactly balanced, was demonstrated to exhibit asymmetric wave reflection while maintaining total transmission on both sides of the structure at their exceptional points [18-20]. Bianisotropic materials enable asymmetric wave scattering with unitary efficiency in the bulk state compared to the existing approaches [21]. Bianisotropic properties were observed in electromagnetism where a coupling tensor that relates electric and magnetic fields with monopolar and dipolar moments in the scattered waves [22]. Acoustic bianisotropic materials couple the pressure and local particle velocity fields with monopole and dipole scattering resulting in asymmetric wave propagations [23-30]. The bianisotropic material was used as an effective sound barrier [31]. The coupling between the pressure and local particle velocity fields along with the scattered acoustic monopole and dipole is characterized by a polarizability tensor. In a 2D scenario, the polarizability tensor is a second order tensor given by �𝑀𝑀𝐷𝐷𝑥𝑥𝐷𝐷𝑦𝑦�= �𝛼𝛼�𝑝𝑝 𝛼𝛼�𝑥𝑥𝑝𝑝𝑝𝑝 𝛼𝛼�𝑦𝑦𝑝𝑝𝑝𝑝 𝛼𝛼�𝑦𝑦𝑝𝑝𝑝𝑝 𝛼𝛼�𝑦𝑦𝑥𝑥𝑝𝑝𝑝𝑝 𝛼𝛼�𝑦𝑦𝑦𝑦𝑝𝑝𝑝𝑝� �𝑝𝑝𝑣𝑣𝑥𝑥𝑣𝑣𝑦𝑦�, (1) 𝛼𝛼�𝑥𝑥𝑝𝑝𝑝𝑝 𝛼𝛼�𝑥𝑥𝑥𝑥𝑝𝑝𝑝𝑝 𝛼𝛼�𝑥𝑥𝑦𝑦𝑝𝑝𝑝𝑝 where 𝑀𝑀 is the monopole scattering, 𝐷𝐷𝑥𝑥 and 𝐷𝐷𝑦𝑦 are the dipole scattering along the x and y axis, 𝑝𝑝 is the pressure field, 𝑣𝑣𝑥𝑥 and 𝑣𝑣𝑦𝑦 are the local particle velocity fields in the x and y directions, and 𝛼𝛼�𝑗𝑗𝑖𝑖 are the elements of the polarizability tensor with i denoting the coupling between the pressure and local particle velocity fields Page 1 is completely reflected by and j denoting the coupling axes, respectively. Given an arbitrary polarizability tensor, one can obtain the scattering field of a bianisotropic particle [32]. For example, the scattering of a bianisotropic particle allowing for both forward and backward scattering, whose polarizability tensor is a function of frequency given by Fig. 1a, is obtained in Fig. 1b. Similarly, the scattering of another bianisotropic particle with a polarizability tensor given by Fig. 1c only permitting backwards scattering is shown in Fig. 1d. In this work, we aim to realize an acoustic bianisotropic grating with its polarizability tensor given by Fig. 1a at 6 kHz (wavelength equal to 5.7 cm). Because the bianisotropic grating we intend to fabricate is linear, the wavelength and frequency remain unchanged during the scattering process, and the incident and scattered wavevectors retain on a circle with its radius given by k = 2π/λ = 109.9 m-1 (Figs. 2a and 2b). The Bloch wavevectors of the grating required to achieve this asymmetric wave propagation in Fig. 1b is given by the difference between the incident and scattered wavevectors shown by the black arrows in Figs. 2a and 2b. A -45-degree incident wave the bianisotropic grating back towards the source that structure (Fig. 2a). The time reversal of the scattering process for this incidence is identical to itself (Fig. 2a). On the other hand, a 45-degree incident wave is refracted with a total transmission in the -45-degree direction, which requires a Bloch wavevector (0, requires a Bloch wavevector (−√2𝑘𝑘,√2𝑘𝑘) given by the −√2𝑘𝑘) from the grating shown in Fig. 2b. The time reversal of this incident case exhibits the same refraction, confirming the preservation of time reversal symmetry in the bianisotropic grating. The asymmetric wave scattering of the -45- and 45-degree incidents reveal the violation of parity symmetry in the y direction, indicating the inversion symmetry breaking of the bianisotropic grating about the x-axis. Thus, the mirrored Bloch wavevectors (√2𝑘𝑘,√2𝑘𝑘) and (0, √2𝑘𝑘) about the x-axis are not provided by the grating. To realize this bianisotropic grating, we determine the grating periodicity from the scattered wave directions from the aforementioned polarizability tensor. The geometry of the unit cells are optimized based on a finite element method, which maximizes the scattering efficiency of the grating (Supplementary Information). The resulting bianisotropic grating consists of a periodic array of scatterers along the y-axis with 40.4 mm spacing. Each scatterer consists of a rectangle having a base of 39.5 mm and height of 14.8 mm adjacent to a second rectangle with a 19.5 mm base and 21.2 mm height centered above the first (Fig. 2c). In order to confirm that the grating elements provide the Figure 1: (a) The elements of the polarizability tensor of a bianisotropic particle that allows both forward and backward scattering as functions of frequency. (b) Asymmetric wave scattering for different incidences from the bianisotropic particle in (a). (c) The elements of the polarizability tensor of a bianisotropic particle with only backward scattering. (d) Asymmetric wave scattering for different incidences from the bianisotropic particle in (c). For both (b) and (d), the blue arrows represent incident and red arrows represent scattered waves, respectively. desired scattering in Fig. 1b, Fourier analysis of the symmetry properties of the scattered wavevectors in the reciprocal space is performed (Fig. 2d), which results in the wavevectors required by the scattering in Figs. 2a and 2b. To verify the bianisotropic properties of the designed grating experimentally, 25 aluminum grating elements were machined by CNC milling (Fig. 2e). In our experiment, the grating elements were arranged in a two-dimensional waveguide with a height of 2 cm bounded by two 1000 mm by 1500 mm acrylic sheets. Plane waves were generated by a speaker array consisting of twelve 17 mm diameter speakers spaced 𝜆𝜆/2 apart, positioned at a 45-degree angle relative to the grating with the closest edge of the array being 500 mm away from the grating (Fig. 2e). In order to create a consistent signal, the pressure amplitude of the speakers was controlled by a digital, multichannel recorder while an omnidirectional microphone attached to a motorized positioner measured the pressure fields for both -45- and 45-degree incidence cases. The motorized positioner was controlled by a MATLAB script to scan two 450 mm by 800 mm areas, which are 10 mm away from the grating with a scan resolution of 𝜆𝜆/10. A lock- in amplifier records the amplitude and phase of the acoustic waves in the scan area (Fig. 2e). To distinguish the incident and scattered waves, spatial 2D Fourier Page 2 Figure 2: (a) Wave-grating interaction for the transmission case for -45-degree incidence in the reciprocal space. (b) Wave-grating interaction for +45-degree incidence as well as the time reversed case represented by the dashed lines. For both (a) and (b), the incident, scattered, and Bloch determined by inverse Fourier analysis of the Bloch wavevectors. (d) 2D spatial Fourier transform of the grating geometry with the expected scattering wavevectors having the highest intensity. (e) Experimental set up for the bianisotropic transmission grating. The zoom out shows the detailed structure of the grating. The outlined rectangles with red dashed lines represent the scanned areas. (f) Measured pressure field of acoustic waves propagating wavevectors are represented by 𝑘𝑘�⃗𝑖𝑖, 𝑘𝑘�⃗𝑠𝑠 and 𝛥𝛥𝑘𝑘, respectively. (c) Bianisotropic grating structure with asymmetric wave scattering in (a) and (b) with positive 𝑘𝑘𝑥𝑥 component for -45-degree incidence. (g) Measured pressure field with negative 𝑘𝑘𝑥𝑥 component for -45-degree. (h) Measured pressure field of acoustic waves propagating with positive 𝑘𝑘𝑥𝑥 component for +45-degree incidence. (i) Measured pressure field with negative 𝑘𝑘𝑥𝑥 component differentiating the waves with positive 𝑘𝑘𝑥𝑥 (Figs. 2f and 2h) and negative 𝑘𝑘𝑥𝑥 components (Figs. 2g and 2i). For for +45-degree incidence. All the pressure fields in (f), (g), (h), and (i) are normalized by their corresponding maximum amplitude in each measurement. analysis is performed on the measured pressure fields, in Fig. 1b with finite dimensions of the waveguide possess open boundaries at the edges of the acrylic sheets resulting in unwanted reflections (as shown by the Fourier analysis in Supplementary Information). Further error arises from the fabrication and imperfect alignment of the grating. These experimental results confirm the feasibility for the realization of bianisotropic properties by breaking the inversion symmetry of the structure. The realization of a bianisotropic grating with polarizability defined by Fig. 1d at 6kHz is similar to the procedure previously executed. By analyzing the wave-grating interaction in reciprocal space, the Bloch wavevectors can be determined for each incidence case. We continue to use a bianisotropic grating with linear elements, the incident and scattered wavevectors are known to remain on a circle with radius 𝑘𝑘 = 109.9 m-1 Bloch wavevector of (−√2𝑘𝑘,−√2𝑘𝑘) (Fig. 3a). This Bloch (Fig. 3a and 3b). A wave with 45-degree incidence interacting with the bianisotropic grating results in a total reflection back towards the source requiring a wavevector is identical to that of the time reversal -45-degree incidence, the acoustic wave is reflected back towards the source by the bianisotropic grating with pressure reflection coefficient R = 0.72 and no transmission through the grating (Fig. 2f and 2g). This experimental result matches the desired scattering of the bianisotropic structure its polarizability tensor given by Fig. 1a. On the other hand, a 45-degree incidence results in a transmission at a -45-degree angle with pressure transmission coefficient T = 0.78 and no reflection from the grating (Figs. 2h and 2i), which matches the scattering of the bianisotropic material in Fig. 1b. The full wave simulations reflection and transmission for the two incident cases, respectively (Supplementary Information), which indicates near perfect scattering efficiencies. The experimental error arises from the speaker array being an imperfect source that excites unwanted wavevectors. In addition, the show near perfect Page 3 Figure 3: (a) Wave-grating interaction for the reflection case for 45-degree incidence in the reciprocal space. (b) Wave-grating interaction for normal incidence as well as the time reversed case represented by the dashed lines. For both (a) and (b), the incident, scattered, and Bloch wavevectors are Fourier analysis of the Bloch wavevectors. (d) 2D spatial Fourier transform of the grating geometry with the expected scattering wavevectors having the highest intensity to the left of the inversion symmetry about the y-axis (e) Experimental set up for the bianisotropic reflection grating. The zoomed in image shows the detailed structure of the grating. The outlined rectangle with red dashed lines represents the scanned areas. (f) Measured represented by 𝑘𝑘�⃗𝑖𝑖, 𝑘𝑘�⃗𝑠𝑠 and 𝛥𝛥𝑘𝑘, respectively. (c) Bianisotropic grating structure with asymmetric wave scattering in (a) and (b) determined by inverse pressure field of acoustic waves propagating with a positive 𝑘𝑘𝑥𝑥 component for normal incidence. (g) Measured pressure field with negative 𝑘𝑘𝑥𝑥 component for normal incidence. (h) Measured pressure field of acoustic waves propagating with positive 𝑘𝑘𝑥𝑥 component for +45-degree incidence. (i) Measured pressure field with negative 𝑘𝑘𝑥𝑥 component for +45-degree incidence. (j) Measured pressure field of acoustic waves propagating with positive 𝑘𝑘𝑥𝑥 component for -45-degree incidence. (k) Measured pressure field with negative 𝑘𝑘𝑥𝑥 component for -45-degree incidence. All the pressure requires a Bloch wavevector of (−�1+√22�𝑘𝑘,√22𝑘𝑘) (Fig. fields in (f), (g), (h), (i), (j), and (k) are normalized by their corresponding maximum amplitude in each measurement. propagation. For normal incidence, the acoustic wave is fully reflected towards 135-degree direction, which 3b). The time reversal case for this incidence requires an identical Bloch wavevector, confirming the time reversal symmetry for both incidence cases. The asymmetric wave propagation for different incidence reveals the violation of parity symmetry about the x- axis. Similar to the bianisotropic grating above, the scattered wave directions determine the grating periodicity while a finite element method is used to optimize the unit cell geometry and maximize the scattering efficiency (Supplementary Information). The resulting bianisotropic grating consists of an array of scatterers having an 80.8 mm periodicity. Each scatterer consists of two rectangles: the first having a base of 18.5 mm and height of 26.5 mm directly underneath the second rectangle with a base of 9.4 mm and height of 21.2 mm (Fig. 3c). Analyzing the grating geometries with spatial Fourier transform reveals the resulted Bloch wavevectors in Fig. 3d matching the scattering wavevectors from the symmetry analysis in Figs. 3a and 3b. The designed bianisotropic grating with thirteen elements are fabricated for the verification of the asymmetric wave scattering for the reflection only case (Fig. 3e). The grating elements were arranged linearly in the same waveguide described above. The speaker array was positioned in the same location for angled incidence, but was relocated to the end of the waveguide, 660 mm away from the grating for the normal incidence case. The motorized positioner scanned an identical rectangular area to measure the total pressure field with both incident and scattered waves. The 2D spatial Fourier analysis mentioned above is used to separate the incident and scattered waves. Experimental measurement shows a normal incidence interacting with the bianisotropic grating results in a reflection at a 135-degree angle with reflection coefficient R = 0.76 (Figs. 3f and 3g). The Page 4 an realize time reversal of this incidence is illustrated in Figs. 3j and 3k with a reflection coefficient R = 0.88. The 45- degree incident wave is reflected with a reflection coefficient R = 0.84 back towards the source (Figs. 3h and 3i). These experimental results match with the desired scattering of the bianisotropic material in Figs. 1e and 1f, whose polarizability tensor is given by Fig. 1d. Once again, the simulations show a near perfect reflection for each incident case (Supplementary Information). The unwanted wavevectors produced by the imperfect source array as well as the open boundaries at the edges of the waveguide affect the reflections in the experiment as shown in the Fourier analysis of the propagating waves in Supplementary Information. The fabrication and alignment errors of the grating also contribute to the experimental error. In conclusion, we demonstrate a systematic approach to arbitrarily given bianisotropic polarizability by a designed acoustic grating (for other arbitrary angles of in Supplementary Information). This is achieved by optimizing the geometry of the grating element to realize maximum scattering efficiency and confirming the asymmetric wave scattering of the desired bianisotropic properties in the reciprocal space through spatial Fourier analysis. The resulted parity symmetry violation of the acoustic wave propagation indicates the necessity of inversion symmetry breaking for the realization of the bianisotropic properties. The experimentally measured acoustic wave interactions with the designed bianisotropic grating match the desired scattering patterns of the given polarizability tensor. Our method paves the road for systematic realization of bianisotropic properties and asymmetric wave propagation with unitary efficiency, which is important for one-way acoustic filtering, sensing, and lensing. incidence, see details †These authors contributed equally to this work ∗Corresponding author. [email protected] References [1] J. Li, L. Fok, X. Yin, G. Bartal, and X. Zhang, Experimental demonstration of acoustic magnifying hyperlens, Nature materials 8, 931 (2009). [2] S. J. Elliott and P. A. Nelson, Active noise control, an IEEE Signal Process. Mag. 10, 12 (1993). [3] R.-Q. Li, B. Liang, Y. Li, W.-W. Kan, X.-Y. Zou, and J.-C. Cheng, Broadband asymmetric acoustic transmission in a gradient-index structure, Appl. Phys. Lett. 101, 263502 (2012). [4] H.-x. Sun, S.-q. Yuan, and S.-y. Zhang, Asymmetric acoustic transmission in multiple frequency bands, Appl. Phys. Lett. 107, 213505 (2015). [5] B. Liang, X. Guo, J. Tu, D. Zhang, and J. Cheng, An acoustic rectifier, Nat. Mat. 9, 989 (2010). [6] B.-I. Popa and S. A. Cummer, Non-reciprocal and highly nonlinear active acoustic metamaterials, Nat. Commun. 5, 3398 (2014). [7] R. Fleury, D. L. Sounas, C. F. Sieck, M. R. Haberman, and A. Alù, Sound isolation and giant linear nonreciprocity in a compact acoustic circulator, Science 343, 516 (2014). [8] D. L. Sounas, C. Caloz, and A. Alu, Giant non- reciprocity at the subwavelength scale using angular momentum-biased metamaterials, Nat. Commun. 4, 2407 (2013). [9] Z. Yang, F. Gao, X. Shi, X. Lin, Z. Gao, Y. Chong, and B. Zhang, Topological acoustics, Phys. Rev. Lett. 114, 114301 (2015). [10] R. Fleury, A. B. Khanikaev, and A. Alu, Floquet topological insulators for sound, Nat. Commun. 7, 11744 (2016). [11] C. He, X. Ni, H. Ge, X.-C. Sun, Y.-B. Chen, M.-H. Lu, X.-P. Liu, and Y.-F. Chen, Acoustic topological insulator and robust one-way sound transport, Nat. Phys. 12, 1124 (2016). [12] J. Lu, C. Qiu, L. Ye, X. Fan, M. Ke, F. Zhang, and Z. Liu, Observation of topological valley transport of sound in sonic crystals, Nat. Phys. 13, 369 (2017). [13] Z. Zhang, Q. Wei, Y. Cheng, T. Zhang, D. Wu, and X. Liu, Topological creation of acoustic pseudospin multipoles symmetry-broken metamaterial lattice, Phys. Rev. Lett. 118, 084303 (2017). flow-free [14] M. Xiao, W.-J. Chen, W.-Y. He, and C. T. Chan, Synthetic gauge flux and Weyl points in acoustic systems, Nat. Phys. 11, 920 (2015). [15] J. Noh, S. Huang, D. Leykam, Y. D. Chong, K. P. Chen, and M. C. Rechtsman, Experimental observation of optical Weyl points and Fermi arc-like surface states, Nat. Phys. 13, 611 (2017). [16] F. Li, X. Huang, J. Lu, J. Ma, and Z. Liu, Weyl points and Fermi arcs in a chiral phononic crystal, Nat. Phys. 14, 30 (2018). [17] W. Zhu, X. Fang, D. Li, Y. Sun, Y. Li, Y. Jing, and H. Chen, Simultaneous Observation of a Topological Edge State and Exceptional Point in an Open and Non-Hermitian Acoustic System, Phys. Rev. Lett. 121, 124501 (2018). [18] X. Zhu, H. Ramezani, C. Shi, J. Zhu, and X. Zhang, P t-symmetric acoustics, Phys. Rev. X 4, 031042 (2014). [19] R. Fleury, D. Sounas, and A. Alù, An invisible acoustic sensor based on parity-time symmetry, Nat. Commun. 6, 5905 (2015). in a Page 5 [20] C. Shi, M. Dubois, Y. Chen, L. Cheng, H. Ramezani, Y. Wang, and X. Zhang, Accessing the exceptional points of parity-time symmetric acoustics, Nat. Commun. 7, 11110 (2016). [21] G. Castaldi, S. Savoia, V. Galdi, A. Alu, and N. Engheta, in Advanced Electromagnetic Materials in Microwaves and Optics (METAMATERIALS), 2013 7th International Congress on (IEEE, 2013), pp. 7. [22] Y. Ra'di, D. L. Sounas, and A. Alu, Metagratings: beyond the limits of graded metasurfaces for wave front control, Phys. Rev. Lett. 119, 067404 (2017). [23] S. A. Tretyakov and A. A. Sochava, Reflection and transmission of plane electromagnetic waves in uniaxial bianisotropic materials, Int J Infrared Millimeter Waves 15, 829 (1994). [24] C. F. Sieck, A. Alù, and M. R. Haberman, Origins of Willis coupling and acoustic bianisotropy in acoustic metamaterials source-driven homogenization, Phys. Rev. B 96, 104303 (2017). [25] M. B. Muhlestein, C. F. Sieck, P. S. Wilson, and M. R. Haberman, Experimental evidence of Willis coupling in a one-dimensional effective material element, Nat. Commun. 8, 15625 (2017). through [26] J. Li, C. Shen, A. Díaz-Rubio, S. A. Tretyakov, and S. A. Cummer, Systematic design and experimental demonstration of bianisotropic metasurfaces for scattering-free manipulation of acoustic wavefronts, Nat. Commun. 9, 1342 (2018). [27] L. Quan, Y. Ra'di, D. L. Sounas, and A. Alù, Maximum Willis Coupling in Acoustic Scatterers, Phys. Rev. Lett. 120, 254301 (2018). [28] L. Quan, D. L. Sounas, and A. Alu, Non-Reciprocal Willis Coupling in Zero-Index Moving Media, arXiv preprint arXiv:1809.09641 (2018). [29] D. Torrent, Acoustic anomalous reflectors based on diffraction grating engineering, Physical Review B 98, 060101 (2018). [30] S. Koo, C. Cho, J.-h. Jeong, and N. Park, Acoustic omni meta-atom for decoupled access to all octants of a wave parameter space, Nature communications 7, 13012 (2016). [31] B.-I. Popa, Y. Zhai, and H.-S. Kwon, Broadband sound barriers with bianisotropic metasurfaces, Nature communications 9, 5299 (2018). [32] X. Su and A. N. Norris, Retrieval method for the bianisotropic polarizability tensor of Willis acoustic scatterers, Phys. Rev. B 98, 174305 (2018). Page 6
1707.06882	1	1707	2017-07-21T13:03:22	Tuning of the Quantum-Confined Stark Effect in Wurtzite $[000\bar{1}]$ Group-III-Nitride Nanostructures by the Internal-Field-Guarded-Active-Region Design	[ "physics.app-ph", "cond-mat.mes-hall" ]	Recently, we suggested an unconventional approach [the so-called Internal-Field-Guarded-Active-Region Design (IFGARD)] for the elimination of the crystal polarization field induced quantum confined Stark effect (QCSE) in polar semiconductor heterostructures. And in this work, we demonstrate by means of micro-photoluminescence techniques the successful tuning as well as the elimination of the QCSE in strongly polar $[000\bar{1}]$ wurtzite GaN/AlN nanodiscs while reducing the exciton life times by more than two orders of magnitude. The IFGARD based elimination of the QCSE is independent of any specific crystal growth procedures. Furthermore, the cone-shaped geometry of the utilized nanowires (which embeds the investigated IFGARD nanodiscs) facilitates the experimental differentiation between quantum confinement- and QCSE-induced emission energy shifts. Due to the IFGARD, both effects become independently adaptable.	physics.app-ph	physics	Tuning of the Quantum-Confined Stark Effect in Wurtzite [0001(cid:3)] Group-III-Nitride Nanostructures by the Internal-Field-Guarded-Active-Region Design S. Schlichting1,, G. M. O. Hönig2, J. Müssener3, P. Hille3, T. Grieb3, J. Teubert4, J. Schörmann4, M. R. Wagner1, A. Rosenauer3, M. Eickhoff3, A. Hoffmann1, and G. Callsen1,† 1: Institut für Festkörperphysik, Technische Universität Berlin, Hardenbergstr. 36, 10623 Berlin, Germany 2: Bundesanstalt für Materialforschung und -prüfung (BAM), 12200 Berlin, Germany 3: Institut für Festkörperphysik, Universität Bremen, Otto-Hahn-Allee 1, 28359 Bremen, Germany 4: I. Physikalisches Institut, Justus-Liebig Universität Giessen, Heinrich-Buff-Ring 16, 35392 Giessen, Germany Recently, we suggested an unconventional approach [the so-called Internal-Field-Guarded-Active-Region Design (IFGARD)] for the elimination of the crystal polarization field induced quantum confined Stark effect (QCSE) in polar semiconductor heterostructures. And in this work, we demonstrate by means of micro-photoluminescence techniques the successful tuning as well as the elimination of the QCSE in strongly polar [(cid:4)(cid:4)(cid:4)(cid:5)(cid:6)] wurtzite GaN/AlN nanodiscs while reducing the exciton life times by more than two orders of magnitude. The IFGARD based elimination of the QCSE is independent of any specific crystal growth procedures. Furthermore, the cone-shaped geometry of the utilized nanowires (which embeds the investigated IFGARD nanodiscs) facilitates the experimental differentiation between quantum confinement- and QCSE-induced emission energy shifts. Due to the IFGARD, both effects become independently adaptable. I. INTRODUCTION the exciton emission energy Group-III-nitride semiconductors are key materials for visible and ultraviolet LEDs, LDs,1–3 and quantum- light sources.4–6 In particular GaN as well as AlN favorably crystalize in the wurtzite crystal structure.7 Hence, heterostructures based on these materials suffer from a strong electric field induced by a piezo- and pyroelectric polarization parallel to the most natural crystal growth direction [0001], the so called c axis.8–12 The polarization-induced internal fields cause a redshift of these heterostructures, known to be the prominent feature of the quantum-confined Stark effect (QCSE),13–17 which is accompanied by a drastic decrease of the spatial electron-hole overlap in the direction of the c axis.4,9,18– 27 Different approaches to eliminate or to diminish the electric field in group-III-nitride heterostructures (across the optically active region) have been investigated, such as growth on non- or semi-polar crystal planes,28,29 forcing the growth of the cubic zincblende phase,30 or by screening the fields with inside doping-induced free carriers.31 Generally, approaches to avoid the preferential [0001] wurtzite crystal growth are challenging, slow, and they often produce a reduced crystal quality.32–37 A more promising method to control the internal electric field is the Internal-Field- (IFGARD),38,39 Guarded-Active-Region Design theoretically developed by Hönig et al..40 As described in detail in Ref. 40, a conventional structure of a GaN quantum well (QW) embedded in AlN barriers is complemented in the IFGARD structure by additional GaN guard layers enclosing the AlN barriers [see Fig. 1(a) + (c)]. This is not intuitive as the additional GaN guard layers reabsorb a particular percentage of the photons generated in the QW. But as discussed by Hönig et al.,40 the overall gain based on the elimination of the polarization field in the QW and the resulting boost in the exciton recombination probability, can overcompensate the reabsorption losses by the guard layers, if the thickness of the guard layer in the emission direction is below the emitted wavelength.41–44 It was numerically demonstrated that this concept leads to a [email protected] † Present address: Institute of Physics, École Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland 1 (NWs) containing GaN nanodiscs (NDs). In particular, we demonstrate the reduction of the internal electric field inside of the GaN NDs down to zero, which reveals the immense advantage of the IFGARD over alternative, more conservative design concepts. Furthermore, we identify the individual contributions of the pure QCSE and the pure confinement effect based on two sample series with varying barrier and ND thickness. II. SAMPLES beam epitaxy The investigated NWs were grown by plasma-assisted (Tsubstrate = 790 °C, molecular TGa = 916 °C, TAl = 1069 °C) on Si (111) substrates. 21,31 Figure 1 illustrates the GaN NWs [Fig. 1(a+b)] with the corresponding band profiles along the growth axis [Fig. 1(c)] as well as corresponding Scanning- Transmission-Electron-Microscopy (STEM) images [Fig. 1(d)]. The STEM images were recorded with a HAADF (High-Angle Annular Dark Field) detector showing single NWs with 1-nm-thick AlN barriers (left, black) and 6-nm-thick AlN barriers (right, black) at a constant ND (white) thickness of 4 nm. The NWs grow along the polar [0001(cid:3)] axis of the wurtzite crystal structure. Each NW embeds a stack of 40 GaN NDs separated by AlN barriers. The first GaN guard layer has a length (in growth direction) of 650 nm. The NW ends with a 20-nm-thick GaN guard layer [Fig. 1(a+b)] in order to realize the fundamental IFGARD symmetry. It is important to underline that the most simplistic stack element consists of one AlN barrier followed by one GaN ND40 being repeated 40- times in order to clearly separate the ND from the guard layer signal. This sequence is encapsulated by the aforementioned GaN guard layers. In total, the NWs have a length of up to 1 µm. The formation of an AlN shell and an increasing diameter of successively grown GaN NDs is observed due to the significant lateral growth of AlN at the applied temperatures [Fig. 1(d)].45 Consequently, the NWs exhibit a cone-shaped structure46 with the diameter increasing from 28 to 36 nm for the NWs with 1-nm-thick AlN barriers [Fig. 1(d), left] and from 38 to 74 nm for the NWs with 6-nm-thick AlN barriers [Fig. 1(d), right]. Considering the conservation of a constant interface-charge density, the particular cone shape leads to different net interface charges [red +, blue - in Fig. 1(a+b)] at either side of each AlN barrier; because the top interface to the GaN is larger than the interface to the GaN below each AlN barrier. This total interface- charge discrepancy is the perfect feature to proof the influence of the IFGARD on the electric field within [Fig. 1 FIG. 1. Schematic sketch of nanowires (NWs) with embedded IFGARD nanodiscs with AlN barriers of 1 nm (a) and 4 nm (b) thickness. NWs with an increased barrier thickness are accompanied by an increased cone-shape angle α due to lateral growth of the barrier material. Red + and blue - mark the positions of interface charges induced by the wurtzite crystal polarization. Due to the cone-shape angle α of the NWs (b), the individual net interface charges of each AlN-barrier interface differ. In (c), a sketch of the black IFGARD band profile is compared to the corresponding conventional nanodisc band profile in red. (d) shows HAADF-STEM images of the stacked GaN/AlN nanodiscs with 1 nm (left) and 6 nm barriers (right). constant piezo- and pyroelectric polarization potential inside of a QW, which results in flat conduction- and valence-band edges (c)]. This elimination of the electric field inside the QW results in a higher spatial overlap of the electron and hole wave function, and hence in a drastic rise of the exciton recombination probability. As calculated in Ref. 40 this maximized probability implicates a gain in the emitted single photons by 2 orders of magnitude. It is important to note that the IFGARD approach does not require a change in the underlying growth procedure; hence, no degradation of the crystal quality occurs as frequently observed for non- and semi-polar growth. Maintaining the initial crystal growth method and substrate material is a big advantage of the present concept compared to any other alternative approaches that aim to reduce or even eliminate the internal polarization semiconductor heterostructures.32–37 In this work, we show the successful implementation of the IFGARD concept for the example of nanowires recombination exciton therein fields in 2 the NDs, as in this case the flatness of the band-edge [Fig. 1(c)] depends on the barrier thickness - similar to what Hönig et al. discussed in Ref. 40 for IFGARD- based quantum dots. Hence, it becomes possible to control the built-in electric field in the NDs by adjusting the AlN barrier thickness and thus the net interface charges. Furthermore, thicker AlN barriers also have an enhanced lateral dimension, which leads to a more pronounced cone-shape of such NWs with thicker AlN barriers, further enhancing their built-in electric field strengths. The angle α of the cone-shaped NW, delineated in Fig. 1(b), gets smaller with reduced AlN barrier thickness until the cone-shape turns into a straight shape [Fig. 1(a) vs. 1(b, d)]. Consequently, we produced two sample series: i) samples with decreasing AlN barrier thickness from 6 nm to 1 nm with a constant GaN ND thickness of 4 nm and ii) samples with a constant AlN barrier thickness of 1 nm and a GaN ND thicknesses of 1, 2, and 3 nm. Since a ND thickness of 4 nm is large compared to the exciton-Bohr (cid:10)(cid:11) radius in GaN (effective Bohr-radius (cid:7)∗ = (cid:10)∗ ɛ(cid:13)(cid:7)(cid:14) ≤ 3 (cid:18)(cid:19)),4,47–51 we would expect an emission energy (without built-in electric fields) close to the free exciton transition energy in bulk GaN of approximately temperatures.52,53 Hence, 3.48 eV at cryogenic decreasing the thickness of the AlN barriers is expected to blueshift the ND emission towards the free exciton transition energy in bulk material by reducing the internal electric fields in the NDs. After having counteracted the QCSE by the discussed IFGARD structure with minimal barrier widths, we tune the energy to even higher energies by confining the excitons in thinner NDs. III. EXPERIMENTAL DETAILS The optical properties of the NDs were investigated by micro-photoluminescence (µ-PL) and time-resolved µ- PL (µ-TRPL) measurements, the latter with two different experimental setups to explore the exciton life times from the ps- up to the µs-range. The samples were mounted into a micro helium-flow cryostat providing measurement temperatures down to 5 K. In the µ-PL setup the NWs were excited by a pulsed (76 MHz repetition rate, 5 ps pulse width and 2 µm spot size) and frequency-quadrupled fiber laser resulting in an excitation wavelength of 258 nm. The spectra were dispersed by a 0.85 m single monochromator by a 150 lines/mm grating (500 nm blaze). The photons were detected with an UV-enhanced Si-charge-coupled device (CCD) array. All spectra were calibrated with an Hg-lamp and are corrected for the refractive index of FIG. 2. The µ-PL spectra of all investigated samples are excited by a pulsed fiber laser with 258 nm wavelength and an excitation power of 15 µW at 7 K. In all samples the dominant GaN ND luminescence is accompanied by the GaN-guard-layer luminescence (marked by the dashed, blue line), which divides the samples series into the so called QCSE regime and the confinement regime. air. µ-TRPL measurements in the ps- to ns-range were performed with the same laser as used for the µ-PL measurements, here coupled into a subtractive 0.35-m double-monochromator with 2400 lines/mm gratings as dispersing elements (300 nm blaze). The time- correlated single-photon-counting technique is used in combination with a hybrid photomultiplier detector assembly for recording the temporal decays. The µs-range TRPL measurements are performed with a pulsed dye laser (350 nm, 100 Hz, 20 ns pulse width) pumped by a 308 nm XeCL-Excimer laser (100 Hz, 20 ns pulse width). For the dispersion an additive double monochromator with a focal length of 1 m and a holographic 1800 lines/mm grating are used. For recording the PL transients a multichannel-plate photomultiplier with an S20 cathode combined with time-correlated single-photon-counting and multi- channel-scaling (including multi-stop capability) are utilized. The recorded temporal decays are all corrected with response characteristics of the respective setup.54 technique the specific IV. EXMPERIMENTAL OBSERVATIONS Figure 2 displays µ-PL spectra, which are recorded with an excitation power of 15 µW at a temperature of 7 K. 15 µW excitation-power spectra are chosen for this plot as they represent a trade-off between energetically well separated luminescence maxima for all samples with enough intensity to be plotted on a linear scale (compare the yellow bar in Fig. 3). The spectra are not 3 increase normalized and therefore quantitatively comparable in absolute intensities. Please note that the excitation power of 15 µW does not represent the lowest value, which is marked by a green bar in Fig. 3. Besides the dominant ND emission signature, the luminescence band of the bulk-like GaN guard layers - close to the free exciton emission energy in bulk GaN 3.48 eV52,53 is visible and marked in Fig. 2 by the dashed, blue line. As shown in Fig. 2, the emission peak energy of the GaN NDs in the QCSE regime (GaN ND thickness is constant at 4 nm with varied AlN-barrier thickness: 6 nm, 4 nm, 2 nm and 1 nm) shifts from 2.76 eV to 3.34 eV for decreased barrier thicknesses. This shift is accompanied by a strong in absolute luminescence intensity combined with a decrease in the full width at half maximum (FWHM) (750 meV to 75 meV) of the ND-luminescence peaks in Fig. 2 for decreasing barrier thicknesses. For decreasing ND thicknesses an even further shift to higher emission energies from 3.34 eV to 4.04 eV occurs due to a stronger confinement of the charge carriers in the thinner NDs. At the same time, the absolute luminescence intensity decreases while the FWHM of the ND-luminescence peaks increases from 83 meV to 330 meV. The luminescence intensity decrease is caused by the reduced photon-absorption volume of the thinner NDs, while the stronger impact of monolayer-thickness variations in the thinner ND ensembles broadens the emission bands. As motivated at the beginning of this section, Fig. 2 was recorded with an increased excitation power of 15 µW causing a blue-shifted photoluminescence-peak position relative to the lowest pump powers in Fig. 3 due to partial screening of the QCSE by excited charge carriers.55–57 The energetically smallest, measureable luminescence- peak positions are therefore displayed in Fig. 3. Figure 3 reveals the influence of the excitation-pump power that was varied from 0.035 µW to 500 µW on the emission-peak respective ND luminescence. On the right, vertical axis (blue) the emission energy shift is noted, representing the quantitative screening of the QCSE by the excited charge carriers plus a ND-band-filling effect converging for thinner NDs towards 12 meV. The grey shading visually divides Fig. 3 into a region either containing the results for IFGARD samples with a negligible QCSE (with total emission-energy shifts ≤ 14 meV) or those samples, whose emission characteristics are evidently affected by the QCSE. This grey shading is transferred to Fig. 2 including the notations: "QCSE" and "confinement regime". The energy range where the QCSE dominates the emission properties the bulk-like GaN luminescence of the guard layers. As visible in Fig. 3, located below energy of the is FIG. 3. The emission peak positions of the nanodisc (ND) samples are plotted for excitation powers between 0.035 µW and 500 µW. The total emission energy shift between the lowest (min) and highest (max) excitation power is given on the right, vertical axis (blue) representing the achieved polarization field screening by excited charge carriers for the QCSE regime and the effect of ND-band filling for the confinement regime. the total emission-energy shift reduces with decreasing AlN-barrier widths from 617 meV (6 nm AlN barrier) to 43 meV (1 nm AlN barrier). The apparent trends for the intensity, the energy shifts and the evolution of the FWHM values are accompanied by corresponding changes in the exciton lifetimes. The results of the µ- TRPL measurements are presented for all samples within the QCSE regime in Fig. 4 (a) and for the samples within the confinement regime in Fig. 4(b). As the transients exhibit a biexponential decay,58 the fit parameters, ꚍ1 and ꚍ2, are shown in the legends of Fig. 4(a) and (b). These transients are recorded at the individual ND luminescence maxima and the dominant decay time (highest intensity in the temporal decay) for each sample is highlighted by bold letters in the legends. Remarkably, for a fixed GaN ND thickness of 4 nm, the decay time decreases by more than two orders of magnitude from (7.34 ± 0.10) µs for the thickest AlN barriers down to (40 ± 7) ns for its thinnest counterpart, as predicted for the IFGARD approach.40 Subsequently, if one decreases the ND thickness (4 – 1 nm) for the samples with the thinnest AlN barrier of 1 nm, to approximately 0.7 ns for all three samples. All those samples that are part of the confinement regime feature very similar transients, presented in Fig. 4(b). further down time drops the decay 4 V. DISCUSSION reduces It is important to note that a 4-nm-thick GaN ND is not able to significantly quantum-confine excitons by its dimensions in absence of the QCSE.4,49–51 Therefore, all IFGARD samples with a ND thickness of 4 nm would (without the cone-shape of the embedding NWs) emit at a luminescence energy close to the free-exciton- emission energy of bulk GaN (3.48 eV52,53). This free exciton emission energy is marked by the luminescence of the bulk-like GaN guard layers in all recorded spectra and represents the transition between the QCSE and the confinement regime. In the confinement regime, the NDs show no indication of built-in fields and the overall energy shift to higher values for thinner NDs is solely caused by the quantum confinement without being affected by a counteracting QCSE red shift. This fact is demonstrated by the power-dependent µ-PL measurements analyzed in Fig. 3. Here, a screening of the internal electric field by excited charge carriers at high excitation powers becomes visible for the QCSE regime, only. The screened QCSE energy drastically for decreasing AlN barrier thicknesses. Within the QCSE regime, the internal electric field strength is a consequence of two effects: first, the unequally sized AlN-GaN interfaces due to the cone-shape of the NWs on either side of each AlN barrier lead to a non-zero net-polarization field inside the NDs, similar to the IFGARD quantum dot case theoretically described by Hönig et al.40 as the total number of space charges on both interfaces differ from each other. Second, the pronounced cone-shape angle α [compare Fig. 1(b)] in the samples with thicker AlN barriers causes more unequal AlN-GaN interfaces further strengthening the net-polarization as explained in the first case. Hence, the whole QCSE regime features a strong dependence of the ND luminescence energy on the AlN barrier thickness in Fig. 2 and strong screening effects by enhanced excitation powers, cf. Fig. 3. Here, even the highest pump powers (already leading to a moderate sample degeneration under long- time exposure) cannot fully screen the built-in polarization fields in the samples with AlN barrier widths exceeding 1 nm - similar to the unattainable full carrier screening in conventional GaN/AlN QW structures.31 Decreasing the ND thickness further reduces the field strength within the NDs [compare Fig. 1(a)] as the finite lateral dimensions of the equally sized "capacitor plates" (due to the finite dimensions of the NWs) become less important. Therefore, the QCSE in the whole ND series becomes negligible resulting in the convergence of the total energy shifts against 12 meV (visible in Fig. 3) indicating a ND-band-filling effect.59– 61 Hence, the overall emission-peak positions of the confinement regime samples in Fig. 2 are purely caused by quantum confinement plus a ND-band filling. This is similar to what was achieved earlier for non-polar GaN/AlN QWs,29,62–64 which still exhibit lower emission energies than the investigated IFGARD NDs. Besides the aforementioned influence of the decreasing FIG. 4. The time-resolved µ-photoluminescence (µ-TRPL) transients of the samples within the QCSE regime shown in (a) reveal the decreasing exciton decay times due to smaller residual electric field strengths in samples with decreased barrier widths. (b) µ-TRPL transients of the samples within the confinement regime show a much smaller, constant decay time due to the eliminated electric field strength. The exciton lifetimes for each biexponential decay are given in the legend; the dominant lifetime is highlighted by bold numbers. All µ- TRPL individual ND luminescence maxima. transients are recorded at the 5 for their non-visibility QCSE on the spectra in Fig. 2, we obtain an increased overlap between the electron and hole occupation probability density, leading to an enhanced exciton- decay rate. In turn, this leads to a strong increase in absolute intensity for the samples with decreased AlN- barrier widths in the QCSE regime. Additionally, the decreased electric dipole moment of the exciton couples less efficiently to defect charge fluctuations in the ND vicinity, which is one mechanism that reduces the FWHM for the emission peaks of the samples related to the QCSE regime.18 Furthermore, the 1-nm- thick AlN barriers and 4-nm-thick GaN NDs (Fig. 2, blue) exhibit multiple LO-phonon replicas (ELO = 94 meV) whereas all samples within the confinement regime do not show any phonon replicas. While the visibility of multiple LO-phonon replicas implicates a large Huang-Rhys-factor and confirms the existence of built-in excitonic electric dipole moments65 within the QCSE-regime NDs, the confinement-regime NDs indicates the absence of a built-in excitonic electric dipole moment. When the thickness of the NDs in the confinement regime is decreased, the emission energy even exceeds 4 eV due to the strong confinement in absence of the QCSE. The reduction of the absolute intensity (for a constant pump power) can straightforwardly be understood: As the number of NDs (40) is conserved, the photon-absorption volume is reduced. At the same time, the FWHM of the NDs' luminescence peaks within the confinement regime rises. This can be understood considering our quantitative STEM analysis showing random ND-thickness fluctuations of ±1 monolayer thickness that have the strongest impact on the confinement energy of the excitons within the thinnest NDs.66 After having separated the QCSE from the confinement effect, the remaining internal electric field strength can be estimated for individual NDs of the QCSE regime simply by the energetic difference between the GaN- guard-layer-luminescence energy and the respective ND emission energy. This difference represents the energy of the electric dipole moment of one exciton in the built-in polarization field. As shown by Nakaoka et al.,67 the quadratic polarizability of the exciton is only of relevance for electric field strengths up to 100 kV/cm, causing energetic shifts of up to 3 meV. Therefore, we can safely neglect the polarizability of the ND exciton and roughly approximate it by a solid electric dipole moment. Here, we estimate the maximal possible dipole moment μ(cid:22) by the thickness of the NDs times the elemental charge (cid:23), being μ(cid:22) ≈ 4 nm ∙ (cid:23) for the QCSE regime.68 Hence, the energetic luminescence offset to the 3.47 eV of the bulk-like GaN guard layers (cid:29)(cid:30)(cid:31) − (cid:29) " = μ(cid:22)#(cid:22) 69 has to be divided by μ(cid:22) to give FIG. 5. The graphical overview summarizes all results of this work and illustrates the correlation between barrier and nanodisc thicknesses, the nanodisc emission energies (within the confinement regime, marked by black squares and the QCSE regime by red dots), the unscreened electrical field strengths (green values), the exciton confinement energies (black values), and the dominant decay times (orange values). Additionally, the geometric nanowire structure in the respective regimes is indicated. an easy approximation of the built-in field strength, that is noted in green in Fig. 5. This approximation should become more reliable for higher field strengths, as the true electric dipole moment of the exciton converges towards the approximated 4 nm ∙ (cid:23) with higher field strengths. Nevertheless, the estimated 2.61 MV/cm in the IFGARD QW with 6 nm AlN barriers are still more than two-times smaller than the field strengths in comparably thick conventional GaN/AlN QWs.26,70,71 Although this discrepancy might lead to the assumption of smaller electric dipole moments, i.e., μ(cid:22) < 2 nm ∙ (cid:23) this would be unrealistically small for a 4 nm thick ND, demonstrating the IFGARD effectively reducing the field strength in all samples. Moreover, the overall decrease of the decay time from 7.34 µs down to 0.7 ns [in Fig. 4] correlates with the decreasing electric field strength of the NDs from (a) 2.61 MV/cm over 0.38 MV/cm down to approximately zero for the whole confinement regime (b) with a constant effective decay time of approximately 0.7 ns. Figure 5 summarizes all discussed results: As expected, for a pure confinement effect,62–64 the luminescence energies (black squares) in the confinement regime converge towards the bulk-like GaN-guard-layer luminescence energy of 3.47 eV. Finally, the (111)Si-(0001)GaN interface is expected to build up space charges at the bottom of the NWs. The resulting net-bottom-interface charge causes a field strength that strongly diminishes over the 650 nm distance of the IFGARD-ND stack according to #~ (cid:13)( and has therefore only a negligible influence on the entire ND to understand important ' it is that 6 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. luminescence. This point-charge approximation would not be valid for IFGARD QWs, where the distance to the wafer interface would be small if compared to the wafer diameter. Hence, in such QW samples the bottom hetero-interface should be avoided, e.g., by homo- epitaxial growth or a subsequent laser lift-off process,72 seeking to reestablish the fundamental symmetry of the IFGARD approach. VI. SUMMARY In summary, we experimentally demonstrate that the IFGARD represents a beneficial method to suppress the internal electric fields in polar heterostructures such as nanodisc (ND) structures. As a result, the QCSE is eliminated, reflected by a shift of the emitted photons to higher energies as compared to conventional quantum wells with matchable thickness and by an enhanced oscillator strength that causes a distinct intensity boost and a decrease in decay times by more than two orders of magnitude for a constant ND thickness. Furthermore, we show that in the presented cone-shaped NW structures, the internal electric field becomes adjustable by the thickness of the AlN barriers. Therefore, the IFGARD opens the possibility to effectively tune the QCSE in polar semiconductor heterostructures without the need for the growth of extraordinary phases or epitaxy on non-polar surfaces. ACKNOWLEDGMENTS support from (DFG) We acknowledge the Deutsche Forschungsgemeinschaft within the Collaborative Research Center 787 (CRC 787). J.M. acknowledges financial support from the JLU Giessen via the graduate scholarship. AUTHOR CONTRIBUTIONS S. S. wrote the manuscript, performed the measurements, and analyzed the presented data. S. S., G. H., and G. C. discussed and interpreted the data. G. H., G. C., and A. H. are the inventors of the IFGARD. J. M., P. H., J. T, J. S., and M. E. developed the NDs in NW sample structure, produced the samples and contributed to the overall data interpretation. T. G. and A. R. supplied the STEM analysis. G. H., G. C., A. H., and M. W. supervise the IFGARD project. Gil, B. III-Nitride Semiconductors and Their Modern Devices. (Oxford University Press, 2013). Nakamura, S., Pearton, S. & Fasol, G. The Blue Laser Diode. The Complete Story. (Springer, 2013). Nakamura, S. & Fasol, S. The blue laser diode-GaN light emitters and lasers. (Springer, 1997). 1. 2. 3. Hönig, G. et al. Manifestation of unconventional biexciton states in quantum dots. Nat. Commun. 5, 5721 (2014). Callsen, G. et al. Steering photon statistics in single quantum dots: From one-to two-photon emission. Phys. Rev. B 87, 245314 (2013). Holmes, M. et al. Measurement of an Exciton Rabi Rotation in a Single GaN/Al(x)Ga(1-x)N Nanowire- Quantum Photoluminescence Spectroscopy: Evidence for Coherent Control. Phys. Rev. Lett. 111, 57401 (2013). Using Dot Schliwa, A., Hönig, G. & Bimberg, D. in Multi-Band Effective Mass Approximations 57–85 (Springer, 2014). Ambacher, O. & Majewski, J. Pyroelectric properties of Al (In) GaN/GaN hetero-and quantum well structures. J. Phys. Condens. Matter 14, 3399 (2002). Langer, R. et al. Giant electric fields in unstrained GaN single quantum wells. Appl. Phys. Lett. 74, 3827 (1999). Ambacher, O. et al. Electronics and sensors based on pyroelectric AlGaN/GaN heterostructures. Phys. Status Solidi C 0, 1878 (2003). Bernardini, F. & Fiorentini, V. Macroscopic polarization nitride heterojunctions. Phys. Rev. B 57, 9427 (1998). offsets band and at Ambacher, O. et al. Role of Spontaneous and Piezoelectric Polarization Induced Effects in Group- III Nitride Based Heterostructures and Devices. Phys. Status Solidi B 216, 381 (1999). built-in Leroux, M. et al. Quantum confined Stark effect due to in (Al,Ga)N/GaN quantum wells. Phys. Rev. B 58, 13371 (1998). polarization internal fields Grandjean, N. et al. Built-in electric-field effects in wurtzite AlGaN/GaN quantum wells. J. Appl. Phys. 86, 3714 (1999). Nakaoka, T., Kako, S. & Arakawa, Y. Quantum confined Stark effect in single self-assembled GaN/AlN quantum dots. Phys. E Low-Dimensional Syst. Nanostructures 32, 148 (2006). Cingolani, R. et al. Spontaneous polarization and piezoelectric field in GaN/Al 0.15 Ga 0.85 N quantum wells: Impact on the optical spectra. Phys. Rev. B 61, 2712 (2000). Fiorentini, V., Bernardini, F., Della Sala, F., Di Carlo, A. & Lugli, P. Effects of macroscopic polarization in III-V nitride multiple quantum wells. Phys. Rev. B 60, 8849 (1999). Kindel, C. et al. Spectral diffusion in nitride quantum dots: Emission energy dependent linewidths broadening via giant built-in dipole moments. Phys. Status Solidi RRL 8, 408 (2014). Kako, S. et al. Size-dependent radiative decay time of excitons in GaN/AlN self-assembled quantum dots Size-dependent radiative decay time of excitons in GaN/AlN self-assembled quantum dots. Appl. Phys. Lett. 83, 984 (2003). 7 Renard, J. et al. Evidence for quantum-confined Stark effect in GaN/AlN quantum dots in nanowires. Phys. Rev. B 80, 121305 (2009). Müssener, J. et al. Probing the internal electric field in GaN/AlGaN nanowire heterostructures. Nano Lett. 14, 5118 (2014). Kindel, C. et al. Exciton fine-structure splitting in GaN/AlN quantum dots. Phys. Rev. B 81, 241309(R) (2010). Ostapenko, I. A. et al. Large internal dipole moment in InGaN/GaN quantum dots. Appl. Phys. Lett. 97, 63103 (2010). Bretagnon, T. et al. Radiative lifetime of a single electron-hole pair in GaN/AlN quantum dots. Phys. Rev. B 73, 113304 (2006). Bretagnon, T. et al. Time dependence of the photoluminescence of GaN/AlN quantum dots under high photoexcitation. Phys. Rev. B 68, 205301 (2003). Adelmann, C. et al. Growth and optical properties of GaN/AlN quantum wells. Appl. Phys. Lett. 82, 4154 (2003). Fiorentini, V., Bernardini, F. & Ambacher, O. Evidence for nonlinear macroscopic polarization in III–V nitride alloy heterostructures. Appl. Phys. Lett. 80, 1204 (2002). Paskova, T. Nitrides with Nonpolar Surfaces: Growth, Properties, and Devices. Nitrides with Nonpolar Surfaces: Growth, Properties, and Devices (WILEY-VCH, 2008). Waltereit, P. et al. Nitride semiconductors free of electrostatic fields for efficient white light-emitting diodes. Nature 406, 865 (2000). Paisley, M. J., Sitar, Z., Posthill, J. B. & Davis, R. F. Growth of cubic phase gallium nitride by modified molecular‐beam epitaxy. J. Vac. Sci. Technol. A 7, 701 (1989). Hille, P. et al. Screening of the quantum-confined Stark effect in AlN-GaN nanowire superlattices by germanium doping. Appl. Phys. Lett. 104, 102104 (2014). Kemper, R. M. et al. Anti-phase domains in cubic GaN. J. Appl. Phys. 110, 123512 (2011). Bürger, M. et al. Lasing properties of non-polar GaN quantum dots in cubic aluminum nitride microdisk. Appl. Phys. Lett. 103, 21107 (2013). Kako, S. et al. Single-photon emission from cubic GaN quantum dots. Appl. Phys. Lett. 104, 11101 (2014). Schupp, T. et al. Zinc-blende GaN quantum dots grown by vaporliquidsolid condensation. J. Cryst. Growth 323, 286 (2011). Sergent, S., Kako, S., Bürger, M., As, D. J. & Arakawa, Y. Narrow spectral linewidth of single zinc-blende GaN/AlN self-assembled quantum dots. Appl. Phys. Lett. 103, 151109 (2013). 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. Farrell, R. M., Young, E. C., Wu, F., DenBaars, S. P. & Speck, J. S. Materials and growth issues for high-performance nonpolar and semipolar light- emitting devices. Semicond. Sci. Technol. 27, 24001 (2012). Pahn, G. M. O., Callsen, G. & Hoffmann, A. Semiconductor device having an internal-field- guarded International Patent Publication Number WO 2017/042368 A1. (2017). region, active Pahn, G. M. O., Callsen, G. & Westerkamp, S. Semiconductor device having an internal-field- guarded active region, European Patent Office Application Number EP17160160.2. (2017). Hönig, G. M. O., Westerkamp, S., Hoffmann, A. & Callsen, G. Shielding Electrostatic Fields in Polar Semiconductor Nanostructures. Phys. Rev. Appl. 7, 24004 (2017). Vico Triviño, N. et al. High quality factor two dimensional GaN cavity membranes grown on silicon substrate. Appl. Phys. Lett. 100, 71103 (2012). photonic crystal Mead, D. M. Wave Propagation in Continuous Periodic Structures: Research Contributions From Southampton, 1964-1995. J. Sound Vib. 190, 495 (1996). Chávez-Ángel, E. et al. Reduction of the thermal conductivity silicon nano- membranes investigated by non-invasive Raman thermometry. APL Mater. 2, 12113 (2014). free-standing in Muth, J. F. et al. Absorption coefficient, energy gap, exciton binding energy, and recombination lifetime of GaN obtained from transmission measurements. Appl. Phys. Lett. 71, 2572 (1997). Furtmayr, F. et al. Carrier confinement in GaN/Al(x)Ga(1−x)N nanowire heterostructures (0<x<1). Phys. Rev. B 84, 205303 (2011). Songmuang, R. et al. Identification of III – N nanowire growth kinetics via a marker technique. Nanotechnology 21, 295605 (2010). Shan, W. et al. Time-resolved exciton luminescence in GaN grown by metalorganic chemical vapor deposition. Appl. Phys. Lett. 67, 2512 (1995). Im, J. S. et al. Radiative carrier lifetime, momentum matrix element, and hole effective mass in GaN. Appl. Phys. Lett. 70, 631 (1997). Efros, A. L. Interband absorption of light in a semiconductor sphere. SPIE milestone Ser. 180, 71 (2005). Banyai, L., Hu, Y. Z., Lindberg, M. & Koch, S. W. Third-order optical nonlinearities in semiconductor microstructures. Phys. Rev. B 38, 8142 (1988). Que, W. Excitons in quantum dots with parabolic confinement. Phys. Rev. B 45, 11036 (1992). Volm, D. et al. Exciton fine structure in undoped GaN epitaxial films. Phys. Rev. B 53, 16543 (1996). Korona, K. P. Dynamics of excitonic recombination and interactions in homoepitaxial GaN. Phys. Rev. B 65, 235312 (2002). 8 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. Pal, J., Tse, G., Haxha, V., Migliorato, M. A. & Tomić, S. Second-order piezoelectricity in wurtzite III-N semiconductors. Phys. Rev. B 84, 85211 (2011). Pal, J., Tse, G., Haxha, V., Migliorato, M. A. & Tomić, S. Erratum: Second-order piezoelectricity in wurtzite III-N semiconductors [Phys. Rev. B 84, 085211 (2011)]. Phys. Rev. B 84, 159902(E) (2011). Hahn, B., Galler, B. & Eng, K. Development of high-efficiency and high-power vertical light emitting diodes. Jpn. J. Appl. Phys. 53, 100208 (2014). O'Connor, D. V., Ware, W. R., Andre, C. Deconvolutiori of Fluorescence Decay Curves. A Critical Comparison. J. Phys. Chem. 83, 1333 (1979). Reale, A. et al. Comprehensive description of the dynamical screening of the internal electric fields of AlGaN/GaN quantum wells time-resolved photoluminescence experiments. J. Appl. Phys. 93, 400 (2003). in Della Sala, F. et al. Free-carrier screening of polarization fields in wurtzite GaN/InGaN laser structures. Appl. Phys. Lett. 74, 2002 (1999). 70. 71. 72. Kuokstis, E. et al. Polarization effects in photoluminescence of C-and M-plane GaN/AlGaN multiple quantum wells. Phys. Rev. Lett. 81, 4130 (2002). Beeler, M. et al. Long-lived excitons in GaN/AlN nanowire heterostructures. Phys. Rev. B 91, 205440 (2015). Casey, H. C., Muth, J., Krishnankutty, S. & Zavada, J. M. Dominance of tunneling current and band filling in InGaN/AlGaN double heterostructure blue light-emitting diodes. Appl. Phys. Lett. 68, 2867 (1996). Nippert, F. et al. Polarization-induced confinement of continuous hole-states in highly pumped, industrial-grade, green InGaN quantum wells. J. Appl. Phys. 119, 215707 (2016). Burstein, E. Anomalous Optical Absorption Limit in InSb. Phys. Rev. 93, 632 (1953). Craven, M. D., Waltereit, P., Speck, J. S., Denbaars, S. P. & Waltereit, P. Well-width dependence of photoluminescence a-plane GaN/AlGaN multiple quantum wells. Appl. Phys. Lett. 84, 496 (2004). emission from Bhattacharyya, A. et al. Comparative study of GaN/AlGaN MQWs grown homoepitaxially on (1- 100) and (0001) GaN. J. Cryst. Growth 251, 487 (2003). Koida, T. et al. Improved quantum efficiency in nonpolar (1120) AlGaN/GaN quantum wells grown on GaN prepared by lateral epitaxial overgrowth. Appl. Phys. Lett. 84, 3768 (2004). Callsen, G. et al. Analysis of the exciton–LO- phonon coupling in single wurtzite GaN quantum dots. Phys. Rev. B 92, 235439 (2015). Rigutti, L. et al. Origin of energy dispersion in Al(x)Ga(1 − x)N/GaN nanowire quantum discs with low Al content. Phys. Rev. B 82, 235308 (2010). Nakaoka, T., Kako, S. & Arakawa, Y. Unconventional quantum-confined Stark effect in a single GaN quantum dot. Phys. Rev. B 73, 121305(R) (2006). Hönig, G. M. O. Mehrteilchenzustände Halbleiter-Quantenpunkten. (2015). in Hönig, G. et al. Identification of electric dipole moments of excitonic complexes in nitride-based quantum dots. Phys. Rev. B 88, 45309 (2013). 9
1711.09249	1	1711	2017-11-25T15:19:14	Interfaces of high efficient kesterite Cu2ZnSnS(e)4 thin film solar cells	[ "physics.app-ph" ]	Cu2ZnSnS(e)4 (CZTS(e)) solar cells have attracted much attention due to the elemental abundance and the non-toxicity. However, the record efficiency of 12.6% for Cu2ZnSn(S,Se)4 (CZTSSe) solar cells is much lower than that of Cu(In,Ga)Se2 (CIGS) solar cells. One crucial reason is the recombination at interfaces. In recent years, large amount investigations have been done to analyze the interfacial problems and improve the interfacial properties via a variety of methods. This paper gives a review of progresses on interfaces of CZTS(e) solar cells, including: (1) the band alignment optimization at buffer/CZTS(e) interface, (2) tailoring the thickness of MoS(e)2 interfacial layers between CZTS(e) absorber and Mo back contact, (3) the passivation of rear interface, (4) the passivation of front interface, and (5) the etching of secondary phases.	physics.app-ph	physics	Interfaces of high efficient kesterite Cu2ZnSnS(e)4 thin film solar cells* Shoushuai Gao(高守帅) 1, Zhenwu Jiang(姜振武)1, Li Wu(武莉)2, Jianping Ao(敖建平)1 †, Yu Zeng(曾玉)1, Yun Sun(孙云)1, and Yi Zhang(张毅)1‡ 1. Institute of Photoelectronic Thin Film Devices and Technology,Tianjin Key Laboratory of Photoelectronic Thin Film Devices and Technology, Nankai University, Tianjin 300071, PR China 2. The MOE Key Laboratory of Weak-Light Nonlinear Photonics, School of Physics, Nankai University, Tianjin 300071, PR China Abstract Cu2ZnSnS(e)4 (CZTS(e)) solar cells have attracted much attention due to the elemental abundance and the non-toxicity. However, the record efficiency of 12.6% for Cu2ZnSn(S,Se)4 (CZTSSe) solar cells is much lower than that of Cu(In,Ga)Se2 (CIGS) solar cells. One crucial reason is the recombination at interfaces. In recent years, large amount investigations have been done to analyze the interfacial problems and improve the interfacial properties via a variety of methods. This paper gives a review of progresses on interfaces of CZTS(e) solar cells, including: (1) the band alignment optimization at buffer/CZTS(e) interface, (2) tailoring the thickness of interfacial layers between CZTS(e) absorber and Mo back contact, (3) the passivation MoS(e)2 of rear interface, (4) the passivation of front interface, and (5) the etching of secondary phases. Keywords: Cu 2ZnSnS4 solar cells, kesterite, interface, passivation PACS:88.40.H-, 79.60.Jv, 73.40.Cg, 88.40.hj 1. Introduction * Project supported by the National Science Foundation of China (51572132, 51372121, 61674082), Tianjin Natural Science Foundation of Key Project (16JCZDJC30700), YangFan Innovative and Entrepreneurial Research Team Project (2014YT02N037), and 111 Project (B16027). † Corresponding author. E-mail: [email protected] ‡ Corresponding author. E-mail: [email protected] Cu(In,Ga)Se2 (CIGS) solar cells are one of the most promising thin film photovoltaic devices with the record efficiency of 22.6%.[1] However, the reserves of indium (In)and gallium (Ga) in the earth is rare and the demand of In in display industry is huge, thus the maximum power production capacity of CIGS solar cells will be limited to ~ 100 GW per year.[2-3] Kesterite structure Cu2ZnSnS(e)4 (CZTS(e)) is derived from chalcopyrite structure CIGS with the more earth-abundant and low-cost elements Zn and Sn to replace the rare elements In and Ga.[4-5] Thus, CZTS(e) solar cells will be much cheaper. Besides, the optoelectronic properties of CZTS(e), such as appropriate band gap (1.0-1.5 eV varying with S/Se ratio) and high absorption coefficient (~104 cm-1), make it ideal for use as an absorber material.[6-7] In the past several years, CZTS(e) solar cells have attracted much attention, and the world record efficiency of Cu 2ZnSn(S,Se)4 (CZTSSe) solar cells has achieved to 12.6% by Mitzi et al. in 2013.[8] But it still has a big gap with the Shockley-Queisser (S-Q) limit of the conversion efficiency of 32.8% for a CZTSSe solar cell.[9] Comparing the device performances of record CZTSSe and CIGS solar cells, as shown in Table 1, the major performance gap between CZTSSe solar cells and CIGS solar cells is the large open circuit voltage (VOC) deficit, expressed as Eg/q- VOC, where Eg is the band gap and q is the elemental charge. Therefore, the VOC deficit is the key limiting factor for high efficient CZTS(e) solar cells. To identify the reasons of big VOC deficit for CZTSSe solar cells, Gunawan et al. studied the temperature dependence of VOC for CIGS and CZTSSe solar cells to obtain the dominant recombination mechanism, as shown in Fig. 1.[10] The relationship between VOC and T can be expressed as[11] 𝑉𝑉OC=𝐸𝐸A𝑞𝑞−𝐴𝐴𝐴𝐴𝐴𝐴𝑞𝑞 ln⁡(𝐽𝐽00𝐽𝐽L) (1) where EA, q, A, k, J00, and JL are the activation energy of the dominant recombination mechanism, elemental charge, diode ideality factor, Boltzmann constant, reverse saturation current prefactor, and photo-current, respectively. Assuming A, J00, and JL to be temperature independent, the VOC almost increase linearly with the decrease of T and the intercept of VOC vs. T plot at 0 K yields the activation energy EA of the dominant recombination mechanism.[11] As shown in Fig. 1, the activation energy EA of CIGS solar cell is almost the same as the band gap of the absorber, which indicates the dominant recombination process in CIGS solar cells occurs in the space charge region or bulk of the absorber.[12] However, the activation energy EA of both CZTSSe solar cells are significantly lower than their corresponding band gap, which is usually attributed to dominant recombination process at interface.[12-13] The interface recombination can be separated into three main categories: (1) A cliff-type band alignment at buffer/absorber interface, where the conduction band edge of the buffer layer is lower than that of the absorber, increasing recombination at interface; (2) Secondary phases at the interface trapping minority charge carriers or producing shunt paths; (3) A non-Ohmic back contact because of the over thick MoS(e)2 between CZTS(e) and Mo interface, increasing the series resistance. This paper review the progress in the interface, including band alignment optimization, tailoring the interfacial layer, interface passivation, and removal of secondary phases to thickness of MoS(e)2 show the perspective of how to improve the performance of kesterite solar cells. 2. Band alignment optimization at buffer/CZTS(e) interface The band alignment at buffer/absorber interface is crucial for high efficient solar cells. There are two kinds of conduction band offset (CBO): cliff like and spike like band alignment, as shown in Fig. 2. For the cliff like band alignment, where the conduction band edge of buffer layer is below than that of absorber layer, the potential difference that can be generated between [14] Besides, the cliff barrier will block the flow of injection electrons under forward bias, causing the quasi-Fermi levels of buffer layer and absorber layer under illumination will be reduced. the increase of recombination between majority carriers via defects at buffer/absorber interface.[14-15] As a result, the VOC and FF will be reduced. However, the spike like band alignment, where the conduction band edge of buffer layer is above that of absorber layer, is less detrimental to VOC, and VOC is nearly constant despite the increase of CBO.[14-15] Besides, the simulation suggested that the JSC is nearly constant when the CBO is below 0.4 eV.[15] However, the photo-generated electrons cannot cross over the spike barrier when the CBO is too large (> 0.4 eV), leading to the decrease of JSC abruptly.[15] Therefore, a well band alignment at buffer/absorber interface is crucial for high efficient solar cells and the optimal range of CBO is 0 eV to 0.4 eV.[14-15] 2.1. CdS/CZTS(e) interface Like CIGS solar cells, CdS is generally used as buffer layer in CZTS(e) solar cells. The champion CZTSSe solar cell with efficiency of 12.6% also used CdS as buffer layer.[8] To explore the reasons why the VOC-deficit of CZTSSe solar cells is so large and further improve the efficiency, it is important to figure out the band alignment between CdS buffer and CZTSSe absorber. Table 2 shows the summary of CBO at CdS/CZTS or CdS/CZTSe interface. In 2011, Bär et al. from Helmholtz-Zentrum Berlin für Materialien und Energie GmbH (HZB) studied the band alignment of CdS/CZTS heterojunction by means of X-ray photoelectron spectroscopy (XPS), ultraviolet photoelectron spectroscopy (UPS), and inverse photoelectron spectroscopy (IPES) spectra.[16] They found that the conduction band offset (CBO) and valence band offset (VBO) at CdS/CZTS interface is -0.33 eV and -1.2 eV, respectively, which means the CBO is cliff like. However, in the same year, Haight et al. from IBM also studied the band alignment at the interface between CdS buffer and Cu2ZnSn(SxSe1-x)4 (CZTSSe) absorber as a function of S/(S+Se) ratio by UPS.[7] Their study indicated that the valence bandminimum (VBM) moves down with the increase of S/(S+Se) ratio, whereas, the conduction band minimum (CBM) is weakly dependent on S content. As a consequence, the VBO at CdS/CZTSSe interface decreases from 0.97 eV to 0.54 eV with the increase of S content, while the CBO is in the range of 0.4-0.5 eV, that is spike like and a little bit bigger than the optimal range (0-0.4 eV).[7] Afterwards, Tajima et al. measured the VBO at CdS/CZTS interface using X-ray photoelectron spectroscopy (HAXPES) and found that the value of VBO is ca. 1.0 eV, then the CBO were deduced as 0.0 eV.[17] Li et al. studied the CBO at CdS/CZTS and CdS/CZTSe interface and found that they are -0.06 and +0.34 eV, respectively.[18-19] Santoni et al. investigated the VBO at CdS/CZTS interface by XPS, and then the CBO were calculated, which is -0.3 eV.[20] Recently, Udaka et al. also investigated the band alignment at the interface between CdS buffer and Cu2ZnSn(SxSe1-x)4 (CZTSSe) absorber with x=0, 0.3, and 1 by means of XPS, UPS, and IPES.[21] On contrary to Haight et al.,[7] they found that the CBM of CZTSSe obviously increases with the S-substitution of Se, and the CBM of CZTS is 0.40-0.45 eV higher than that of CZTSe, while the VBM only moves down 0.07 eV.[21] As a result, the conduction band offset (CBO) at CdS/CZTSSe interface decreases from +0.55 eV to -0.14 eV with the S-substitution of Se.[21] The CBO of +0.55 eV at Cd/CZTSe (x=0) interface is a rather high spike, which will block the transportation of photo- generated electrons and lead to a serious current-loss. Whereas, the negative CBO of -0.14 eV at CdS/CZTS (x=1) interface will cause serious voltage-loss. It is indicated that the CBO can be optimized by modifying the S/(S+Se) ratio, and the optimal ratio is in the range of 0.3-0.4. The conversion efficiencies of Cu2ZnSn(SxSe1-x)4 solar cells with x= 0, 0.3, and 1are 7.2%, 11.4%, and 7%, respectively.[21] The CBO are also calculated using first principles. Bao et al. indicated that there are two inequivalent interfaces between CdS and CZTS, Cu-Zn layer or Cu-Sn layer adjacent to interface, respectively.[22] And the corresponding CBO calculated by first principles are -0.2 eV and -0.7 eV, respectively. If the CBO is averaged, then it is -0.3 eV. Palsgaard et al. calculated the CBO at CdS/CZTSe interface using first principles, which is spike like +0.6 eV.[23] The different results of CBO for the studies may be attributed to the difficulty in getting the accurate values of band bending near the surface, the different process conditions of absorber and buffer layer, and the different composition of the absorber layerespecially at the interface.[14] Thus, the results are inconclusive and more further studies are essential. Overall, the CBO between CdS buffer and CZTSor CZTSe absorber is not optimal.There are following several methods to optimize the conduction band alignment at the buffer/absorber interface. As indicated in the above works, first, the CBO can be optimized by modifying the S/(S+Se) ratio.The optimal S/(S+Se) ratio is in the range of 0.3-0.4.[21] Udaka et al. indicated that the conversion efficiencies increased from 7.2% for CZTSe and 7% for CZTS to 11.4% for Cu2ZnSn(S0.3Se0.7)4solar cells because of the optimization of CBO.[21] Besides, modifying the chemical bath depsition (CBD) process of CdS is another method to optimize the CBO of CdS/CZTSSe interface. Kato et al. from Solar Frontier K.K. indicated that the Zn and Se would diffuse out from CZTSe to CdS during deposition of CdS,therefore, the band edge of CdS would bend upward at the interface and the CBO at the CdS/CZTSe interface would be increased to 0.6 eV, which leads to S-shaped J-V curves with poor FF and efficiency.[24] They modified the CBD process of preparing CdS film to suppress the outdiffusion of Zn and Se, and thus the CBO was reduced to 0.3 eV, which is in the optimal range of 0-0.4 eV.[24] As a result, a CZTSSe submodule were prepared with a record efficiency of 10.8% and a low VOC deficit 561 mV.[24] Neuschitzer et al. optimized the CBD process of CdS by replacing of cadmium sulfate (CdSO4·H2O) with cadmium nitrate (Cd(NO3)2) as Cd precursor source to reduce the deep level acceptor-like traps states in the CdS and then reduce the CBO at CdS/CZTSe interface. As a result, the crossover and red kink of J-V curves were eliminated and the efficiency were improved from 7.0% to 8.2%.[25] 2.2. Zn(O,S)/CZTS(e) interfaces CBO of CdS/CZTS(e) is sensitive to S/(S+Se) ratio. To optimize the band alignment between buffer layer and absorber layer, an alternative buffer layer, Zn(O,S), (Zn,Cd)S, Zn 1-xSnxOy, Zn1- xMgxO, and In2S3,etc., is used to optimize the band alignment at buffer/absorber interface. Among them, Zn(O,S) is studied as an alternative buffer layer due to the advantages of large and tunable band gap (2.6-3.8 eV), high transparency, low toxicity, and earth abundant raw materials.[26-28] Zn(O,S) buffer layer has been successfully used in CIGS solar cells, and world record efficiency 22.8% of CIGSSe solar cell with CBD-Zn(O,S) buffer layer has been achieved by Solar Frontier K.K., which is comparable with the devices with CdS-references.[29] It is indicated that the CBO between ZnS and CZTSSe is spike like, which is as high as 1.1 eV current is nearly zero. (Fig. 3(a)). This high CBO will introduce a high barrier to electron flow and thus the photo- [30] However, the CBO between ZnO and CZTSSe is cliff like (Fig. 3(b)), thus the photo-current is much higher than that with ZnS buffer layer, and the VOC and FF is much lower than that with CdS buffer layer. The band structure and band gap of Zn(O,S) are determined by the O/(S+O) ratio, as shown in Fig. 4.[31-32] So, it is crucial to optimize the band alignment between Zn(O,S) and CZTSSe via modifying the O/(S+O) ratio. Neuschitzer et al. studied the influence of thiourea (TU) concentration during the CBD process on the characteristics of Zn(O,S) layer as well as the influence on the device performances of CZTSe solar cells.[33] As shown in Fig. 5, the J-V curves show a kink-like shape and the distortion of the J-V curves are less pronounced with the decrease of TU concentration. The distortion of J-V curves can be attributed to the high conduction band spike, which acts as a barrier for photo- generated electrons.[33] The O/(S+O) ratio in the Zn(O,S) buffer layer increase with the decrease of sulfur concentration in the CBO solution, and thus the barrier height decrease and even invert into cliff like for the sample with low TU concentration,causing the decrease even elimination of distortion (Fig. 5). As shown in Fig. 6, the distortion can be reduced and the VOC, JSC, and FF increase significantly after light soaking, which can be attributed to the decrease of spike barrier after light soaking because of the photo-doping of Zn(O,S).[33] Grenet et al. also reported that the JSC is nearly zero with Zn(O,S) buffer layer due to the high spike barrier and the conversion efficiency increased significantly to 5.8% after light soaking due to the decrease of spike barrier.[34] The Zn(O,S)/CZTSSe interface can also be improved by modifying the surface of CZTSSe absorber other than modifying the O/(S+O)ratio of Zn(O,S). Sakai et al. indicated that the absorber. efficiency of CZTS solar cells with Zn(O,S) buffer layer decreases with the Zn/Snratio of CZTS [35] Then, they reduced the surface Zn/Sn ratio from 1.1 to 0.96 via Tin-based chemical treatment (TCT) before CBD of Zn(O,S) buffer layer. As a result, a higher efficiency of 5.85% was achieved for a 5 × 5 cm2 CZTS submodule.[36] The results suggest that the surface properties of absorbers are important to the properties of buffer/absorber interface and thus determine the device performance. On the other hand, Zhang group in Nankai University China indicated the co-exist of ZnO and Zn(OH)2 secondary phases in the Zn(O,S) layer will cause band fluctuations due to the large difference of conduction band edge between Zn(O,S) and ZnO, which is detrimental to the device performance, as shown in Fig. 7(a).[28] They proposed a treatment process of etching by concentrated ammonium and followed by soft annealing to eliminate ZnO and Zn(OH)2 secondary phases in the Zn(O,S) layer.[28] As a result, the efficiency of the Zn(O,S)/CZTSe solar cells was improved from 1.17% to 7.2%. According to their study, the CBO between etched & annealed Zn(O,S) and CZTSe is spike like, as high as 1.22-1.28 eV, which will block the transportation of photo-generated electrons in theory. After study the low temperature transportation, they indicated that there is a defects energy level close to the Fermi level, which acts as a shortcut for the transportation of photo-generated electrons across the Zn(O,S) layer, as shown in Fig. 7(b).[28] 2.3. Zn1-xCdx S/CZTS(e) interface As we have mentioned, the CBO at CdS/CZTS and ZnS/CZTS is about -0.3 eV and +1.0 eV, respectively. Thus, Zn 1-xCdxS, a mixture of ZnS and CdS, can be used as buffer layer and the CBO between Zn1-xCdxS and CZTS can be optimized by modifying the Cd/(Zn+Cd) ratio. Bao et al. calculated the band offsets at Zn1-xCdxS/CZTS interface by first principles and found that the CBO at Zn1-xCdxS/CZTS interface is spike like and smaller than 0.3 eV, when the Cd/(Zn+Cd) ratio is 0.625 – 0.75.[37] Sun et al. prepared Zn1-xCdxS as buffer layer for CZTS solar cells by successive ionic layer adsorption and reaction (SILAR) method.[38] Their study indicated that all the performance parameters (including VOC, JSC, and FF) of CZTS solar cells with Zn0.35Cd0.65S buffer are improved compared to CdS-references and the highest efficiency of 9.2% is achieved with the VOC of 747.8 mV, which is significantly improved about 100 mV compared to CdS-references.[38] The improvement is attributed to the spike like CBO (0.37 eV) at buffer/absorber interface, which will reduce the interface recombination. Messaoud et al. also used Zn1-xCdxS buffer layer for CZTSe solar cells and the Zn1-xCdxS buffers were deposited by chemical bath deposition.[39] It is indicated that the VOC and FF decrease with the increase of Zn content in Zn1-xCdxS buffer due to the presence of Zn(OH)2 and ZnO, which bring into the decrease of CBO. Thus, a Cd partial electrolyte (Cd PE) treatment before the deposition of Zn1- xCdxS were adopt to form a thin Cd(OH)2 layer (2-3 nm) to prevent the adsorption of the negatively charged Zn(O,OH)x particles, resulting in the decrease of interface recombination.[39] As a consequence, all the performance parameters of CZTSe solar cells with Cd PE/Zn0.02Cd0.98 8.3% is achieved. S buffer are improved compared to CdS-references, and a highest efficiency of 2.4. (Zn,M)O/CZTS(e) interface (M= Sn, Mg) Zn 1-xSnxOy (ZTO) is also a promising buffer material for CZTS(e) solar cells. The conduction band alignment between ZTO and CZTS(e) can be optimized via tuning the Sn/(Zn+Sn) ratio. Platzer-Björkman et al. prepared ZTO buffer on CZTS absorber by atomic layer deposition (ALD) and optimized the band alignment between ZTO and CZTS by varying the deposition temperature.[40-41] It is worth noting that the highest efficiency of CZTS solar cells with ZTO buffer is 9.0%, which is higher than that of CdS-references.[41] Tajima et al. also used ZTO as buffer layer for CZTS solar cells and the best efficiency of CZTS solar cells with ZTO buffer is 5.7%, which is lower than its CdS-references.[42] They attribute the deterioration to the large lattice mismatch between ZTO and CZTS, which introduces interface recombination. To reduce the interface recombination, a ZTO/CdS double buffer layer was introduced.[42] Before the deposition of ZTO layer, a thin CdS buffer layer (10 nm) was deposited by CBD, followed by annealing in N2 atmosphere. As a result, a high VOC of 0.81 eV was achieved and the efficiency was improved. Zn1-xMgxO (ZMO) is an attractive material as buffer layer due to its big and tunable band gap (3.3-7.7 eV). Besides, the conduction band alignment between ZMO and CZTS(e) can be optimized via tuning the Mg/(Zn+Mg) ratio. Hironiwa et al. prepared ZMO layer as buffer layer for CZTSSe solar cells, where the ZMO were deposited by co-sputtering of ZnO and MgO.[43-45] The efficiency of CZTSSe solar cells with ZMO buffer layer is significantly lower than that of CdS-references, which is attributed to the sputtering damage on CZTSSe absorber. To eliminate the sputtering damage, a ZMO/CdS double buffer layer was introduced.[43-45] As a result, the efficiency of solar cells with double buffer layer is improved to almost the same level of CdS- references. Besides, the efficiency of CIGS solar cells with double buffer layer is higher than the CdS-references. 2.5. In2S3/CZTS(e) interface In2S3 is also a potential material as buffer layer for CZTS(e) solar cells. In2S3 is an indirect semiconductor with a band gap of 2.1 eV, which is slightly smaller than that of CdS (2.4 eV).[30, 46] However, the transparency of In2S3 is higher than that of CdS due to the indirect band gap of In2S3, which is beneficial to the improvement of JSC.[30] Barkhous et al. indicated that the CBO between In2S3 and CZTSSe (for a sulfur content S/(S+Se) ~ 0.4 eV) is 0.15 ± 0.1 eV, the spike like in the optimal range.[30] The CZTSSe solar cells with In2S3 buffer layer yield a highest efficiency of 7.6%, which is comparable with the CdS-references. Yan et al. also used In2S3 as buffer layer in CZTS solar cells and conclude that the CBO between In2S3 and CZTS is 0.41 eV ± 0.1 eV, which is spike like and slightly higher than the optimal value. Thus, they indicated that the high CBO will block the flow of photo-generated electrons.[47] As a result, the CZTS solar cells with In2S3 buffer have a better VOC and lower JSC and FF than the CdS-references. Yu et al. prepared In2S3 by thermal evaporation at different temperatures (30~200 °C).They found that the values of CBO between In2S3 and CZTS are dependent on the deposition temperature of In2S3 and variable in the range of 0.01 ~ 0.41 eV.[48] Besides, co-evaporated and sprayed In2S3 is used as buffer layer in CZTSe solar cells, and the efficiency of solar cells is around 5.7%.[49-50] On the other hand, In 2S3 can combined with CdS as double buffer layer to further improve the efficiency of CZTS(e) solar cells. Kim et al. indicated that a key reason for the low VOC of CZTSSe solar cells is the low majority carrier density and mobility product (i.e. the low conductivity) of the CZTSSe absorber.[51] The mobility of the CZTSSe absorbers is much lower than that of CIGS, thus the crucial to improve VOC is to increase the carrier concentration of absorber. Kim et al. prepared a thin In2S3 layer on CdS as a double buffer followed by rapid thermal annealing (RTA).[51] After RTA, In diffused into CdS and CZTSSe, leading to the formation of n-type doping defects InCd in CdS and p-tye doping defects InSn in CZTSSe, and resulting in the enhancement of carrier concentration of both CdS buffer and CZTSSe absorber. As a consequence, a record efficiency of 12.7% was achieved by employing an In2S3/CdS double buffer layer in CZTSSesolar cell.[51] Hiroi et al. also prepared a hybrid buffer layer with In2S3 deposited on CdS and an efficiency of 9.2% was obtained on CZTS submodule.[52] The improvement of device performance is attributed to the Cu and O diffusion as well as Cd and In diffusion, and the reduced carrier recombination at i-ZnO/buffer interface for the lower VBM at the interface.[52] Differently from Kim et al. and Hiroi et al.,[36, 51] Yan et al. prepared CdS/ In2S3 hybrid buffer with In2S3 inserted between CdS and CZTS, followed by a RTP annealing, yielding an improvement of VOC and efficiency from 5.5% to 6.6%.[53] It is indicated that the In can effectively diffuse into CdS and CZTS to increase the carrier concentrations of CdS and CZTS, and the carrier concentration of CZTS increased by on order of magnitude than CdS- references, resulting in the improvement of VOC.[53] In summary, the band alignment at CdS/CZTS(e) interface is not optimal and it is necessary to optimize the band alignment at buffer/CZTS(e) interface. First, the CBO can be optimized by modify the composition of CZTS(e) absorber, especially the surface composition. Secondly, the CBD process of CdS should be optimized to tailor the interfacial diffusion and reduce the interfacial defects. On the other hand, alternative buffer layers, including Zn(O,S), (Zn,Cd)S, Zn 1-xSnxOy, Zn1-xMgxO, and In2S3 ,etc., can be used to optimize the band alignment at buffer/absorber interface. However, the photovoltaic performances of CZTS(e) solar cells with an alternative buffer are usually not ideal and even worse than the CdS-references. To further improve the device performances of CZTS(e) solar cells with an alternative buffer, much more investigations are needed, including the optimization of the compositions and preparation processes of alternative buffer layers. 3. Mo/CZTS(e) interface Mo films are the most commonly used back contact in CIGS solar cells because of a series advantage of Mo films as an excellent back contact. These merits include low resistivity, stability during growth at high temperature, and good adhesion between CIGS absorber and glass substrate.[54-58] Fig. 8 shows the schematic diagram of the band structure of the CIGS solar cells. As shown in Fig.8(a), the CIGS/Mo interface is a Schottky contact which would increase the series resistance RS and lead to the roll-over phenomenon in the I-V curve of the solar cell.[59] Generally, an interfacial layer of MoSe2 can be formed between CIGS absorber and Mo back contact during the growth of CIGS film. As shown in Fig.8(b), the CIGS/MoSe2 interface is Ohmic contact, whereas the MoSe2/Mo interface is still a Schottky contact because of the large downward band bending at the MoSe2/Mo interface. The MoSe2 interfacial layer formed during the growth of CIGS films is usually very thin (10-100nm). Tunneling would occur across the MoSe2/Mo interface when the thickness of MoSe2 is very thin. Thus, the very thin MoSe2 interfacial layer can act as a buffer layer to convert the Schottky contact to a quasi-Ohmic contact.[59-61] However, the transportation of holes will be suppressed when the MoSe2 interfacial layer are very thick. Zhang's group indicated that that the back contact barrier height in CZTSe solar cells is 135 meV as MoSe2 is very thick, while it is only 24 meV without MoSe2 between Mo and CZTSe film.[62] Therefore, the thickness of MoSe2 layer should be well-controlled to turn the back contact to a quasi-ohmic contact but not Schottky contact. The structure of CZTS(e) solar cells is inherited from CIGS solar cells. Thus, Mo films are Besides, another decomposition process is easily to occur at the interface between CZTS(e) and Mo back contact at high temperature according to the following reaction [65] also widely used as back contact in CZTS(e) solar cells.However, different from CIGS films, the CZTS(e) films are easily to decompose at high temperature as shown below[63-64] 𝐶𝐶𝐶𝐶2𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)4⇌𝐶𝐶𝐶𝐶2𝑍𝑍(𝑒𝑒)+𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)+𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)+12𝑍𝑍(𝑒𝑒)2 (1) 2𝐶𝐶𝐶𝐶2𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)4+𝑀𝑀𝑀𝑀⇌2𝐶𝐶𝐶𝐶2𝑍𝑍(𝑒𝑒)+2𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)+2𝑍𝑍𝑍𝑍𝑍𝑍(𝑒𝑒)+𝑀𝑀𝑀𝑀𝑍𝑍(𝑒𝑒)2 (2) It is calculated that change of the Gibbs energy of Reaction (2) is -150 kJ for CZTS and -100 kJ for CZTSe at 550 oC, which means that it is thermodynamically favorable for the decomposition at CZTS(e)/Mo interface and the MoS(e)2 interfacial layer is easily to be formed between CZTS(e) absorber and Mo back contact.[63, 65] To suppress the formation of MoSe2, Shin et al. modified the selenizing temperature.[66] They indicated that the thickness of MoSe2 are reduced with the selenizing temperature decrease, whereas, the average grains size of CZTSe film are also reduced.[66] Another path to suppress the formation of MoSe2 is selenizing precursors under low Se partial pressure. Yao et al. fabricated 8.2% CZTSe solar cells with a very thin MoSe2 layer by selenizing electrodeposited Cu/Sn/Zn precursor under low Se pressure.[67] However, deep defects (e.g. Se deficiency) are formed when the Se partial pressure is low during selenizing at high temperature.[66] To promote the growth of grains, suppress the decomposition according to Reaction (1), and at high temperature under high S(e) and SnS(e) pressure. avoid the formation of deep defects (e.g. Se deficiency), CZTS(e) absorbers are usually prepared [65-66, 68-70] As a result, an excessively thick MoS(e)2 interfacial layer, which would increase the contact resistance between CZTS(e) absorber and Mo back contact, is easily to be formed according to Reaction (2).[66, 69] Thus, one of the challenges for the improvement of the current conversion efficiencies is to reduce the thickness of MoSe2 interfacial layer. After studying the kinetics of MoSe2 interfacial layer formation during annealing CZTSe thin films under Se ambient, Shine et al. indicated that the formation of MoSe2 is modeled with three step processes, as shown in Fig.9-diffusion of Se through CZTSe film, diffusion of Se through the already formed MoSe2, and then reaction between Se and Mo.[69] It reveals that the key to suppress the formation of MoSe2 is to block the diffusion of Se. Thus, there are following several ways to suppress the formation of MoSe2 . In accordance with the characteristic that Se vapor firstly diffuse through CZTS(e) film, an intermediate layer, such as TiN, TiB 2, MoO2, and ZnO etc., inserted between CZTS(e)absorber and Mo back contact are introduced as a S(e) diffusion barrier to suppress the formation of MoS(e)2.[64, 66, 71-72] TiN is used as the barrier material to prevent the S(e) diffusion for its stability and minor diffusion at high temperature. Shin et al. reduced the thickness of MoSe 2 layer from 1300 nm to 220 nm by a thin TiN diffusion barrier (~20 nm) and thereby the series resistance decreased from 3.4 (cid:127) cm2 to 1.8 (cid:127) cm2 and the efficiency increased from 2.95% to 8.9%.[66] However, Scragg et al. found that TiN intermediate layer would induce a rather high series resistance, causing the decrease of fill factor.[64] Therefore, contrast to the results of Shin,[66] the efficiency of CZTS solar cells was reduced when a TiN intermediate layer was introduced although the MoS2 were out of growth.[64] Schnabel et al. further investigated the influence of different back contact configurations of Mo and TiN (shown as Fig. 10) on the formation of MoSxSe2-x, the crystallinity of the absorber and thereby the performance of solar cell devices.[72] They found that the thin TiN intermediate layer can effectively inhibit the formation of MoSxSe2-x, but is detrimental to the crystallinity of the absorber layer. The photovoltaic performances of solar cell demonstrated that the existence of MoSxSe2-xis necessary for achieving high VOC. Thus, the best performance solar cells were the ones using a combined Mo/TiN/Mo stack as back contact, which can easily control the thickness of MoSxSe2-x layer by the thickness of the upper Mo layer. In addition to TiN, TiB 2 is introduced as the intermediate layer. Liu et al. inhibited the formation of MoS2 layer with TiB2 film as intermediate layer in CZTS thin film solar cells.[71] As a result, the series resistance was greatly reduced and thereby the efficiency of solar cells were increased from 3.06% to 4.40% with the improvement ofJSC and FF. However, the TiB2 intermediate layer would degrade the absorber's crystallinity, which would be detrimental to the device performance especially VOC.[71] Besides, Duchatelet et al. prepared a MoO 2 layer by oxidizing the Mo coated glass substrates to suppress the formation of MoSe2.[73] It is indicated that the Gibbs free energy of the reaction of MoO2 with selenium is +294 KJ/mol which is thermodynamically unfavorable.[73] As a result, the MoO2 can passivate Mo against selenization, causing the significant decrease of thickness of MoSe2 layer.[73] It is known that MoO2 exhibit a metallic electrical conductivity,[74-75] and thus the presence of MoO2 with a thin MoSe2 has little effect on the sheet resistance of back contact.[73] Lopez-Marino et al. found that the presence of MoO2 will change the preferred orientation of MoSe2 from (100) peak to (002) peak, i.e. the Se-Mo-Se sheets changing from perpendicular to parallel to Mo substrate, as shown in Fig.11(a) and (c).[76] The MoSe2 with Se- Mo-Se parallel to Mo substrate can act as a natural Se barrier and thus the thickness of MoSe2 are reduced.[76] However, the conductivity of MoSe2 is anisotropic, and the electrical conductivity along the Se-Mo-Se sheets is two orders of magnitude bigger than that along the direction perpendicular to the Se-Mo-Se sheets, i.e. the electrical conductivity of MoSe2 with Se- Mo-Se sheets parallel to Mo substrate is very poor.[77] To prepare sufficient MoSe2 promote an Ohmic contact between Mo and absorber, another top sacrificial Mo cap layer (20-70 nm) was deposited on the MoO2 layer.[76] The thickness of MoSe2 can be tailored by modifying the thickness of the top sacrificial Mo cap layer and the photovoltaic performances can be improved. Besides, the MoO2 layer can not only reduce the thickness of MoSe2 effectively but also can obviously promote the grain growth, and thus the shunt resistance (RSh) was increased.[76] An intermediate ZnO layer on Mo back contact can also prevent the interaction between CZTS(e) absorber and Mo film during the high temperature annealing process and thereby the decomposition of CZTS(e) would be inhibited. In 2013, López-Marinoet al. introduced an ultrathin (10nm) ZnO intermediate layer to minimize/eliminate the decomposition reaction between CZTSe and Mo interface.[78] Consequently, the CuxSe binary species were eliminated and the density of voids was reduced. As a result, the series resistance were significantly decreased and thereby the efficiency increased from 2.5% to 6.0%.[78] Li et al. and Liu et al. also inhibited the formation of MoS2 by introducing an ultrathin (10 nm) ZnO intermediate layer.[79- 80] It needs to be indicated that the path to suppress the formation of MoS(e) 2 by inserting an intermediate layer can work well in some extent. However, it will make the process complexity and the series resistance of solar cells cannot be decreased to a low level. To simplify the process to suppress the formation of MoSe2 and decrease the series resistance of solar cells, Zhang's group presented two means to suppress the formation of MoSe2 successfully without any additional intermediate layer used.[62, 81]One is that the metal precursors were annealed at 300 oC under Ar atmosphere to form a dense temporary alloy layer, the diffusion of Se was blocked in the following selenization process at high temperature and the thickness of MoSe2 interfacial layer were reduced from 1200 nm to less than 10 nm. The series resistance was reduced from 3.4 (cid:127) ·cm2 to 0.61 (cid:127) ·cm2, and the conversion efficiency were improved from 5.6% to 8.7%.[62] The other is that a thin (103) peak preferred Mo(S,Se)2 layer with tilted Se-Mo-Se sheets were prepared by modifying the surface morphology of Mo films.[81] As shown in Fig. 11 (d), the tilted Se-Mo-Se sheets can act as a natural barrier to inhibit the formation of Mo(S,Se)2 interfacial layer and provide a good electrical conductivity along the Se-Mo-Se sheets.[81] As a result, the thickness of Mo(S,Se)2 sharply decreased from 1500 nm to 200 nm with the surface morphology change of Mo back contact, which resulting in the decrease of series resistance of Cu2ZnSn(S,Se)4 solar cells from 2.94 (cid:127) ·cm2 to 0.49 (cid:127) ·cm2, and the increase of conversion efficiency of Cu2ZnSn(S,Se)4 solar cells from 6.98% to 9.04%. In summary, a thin MoS(e)2 layer between CZTS(e) absorber and Mo back contact can act as a buffer layer to convert the Schottky contact to a quasi-ohmic contact. However, an over thick MoS(e)2 layer is easily to be formed during selenaziton process at high temperature under high S(e) pressure, which will suppress the transportation of holes. Thus, it is crucial to take effective measures to tailor the thickness of MoS(e)2. The formation of MoS(e)2 can be suppressed by a dense temporary precursor, a thin (103) peak preferred Mo(S,Se)2layer with tilted Se-Mo-Se sheets, or an intermediate barrier layer, including TiN, TiB2, and MoO2 4. Passivation 4.1. Passivation of rear interface , etc.. interface. The dielectric To reduce the rear recombination at the semiconductor-metal interface to increase the V OC, and meanwhile to improve the JSC by increasing the rear reflection, a dielectric layer with patterned local contacts has been introduced between CZTS(e) and Mo to passivate the rear [84], or [86] have been used as a passivation layer in Si solar cells for a long time. Al2O3(SiO)/SiNX Recently, the dielectric layer has been adopted in CZTS(e) solar cells. Vermang et al. used an Al2O3 with nanosized point openings to passivate ultrathin CZTS solar cells.[87] The results indicated that the charge carrier recombination and the secondary phases (SnS and ZnS) were layer, such as SiO2 [82-84], SiCX [85], SiNX [83], Al2O3 reduced. As a consequence, compared with CZTS solar cells without rear passivation at CZTS/Mo interface, photovoltaic performances of the CZTS solar cells with an Al2O3 rear passivation layer were obviously increased.[87] Similarly, Kim et al. prepared a dielectric layer (Al2O3 or SiO2) with nano-patterned local contacts by nanosphere lithography to passivate the Mo/CZTSe interface, as shown in Fig. 12.[88] All photovoltaic parameters were significantly improved and the luminescence intensity of LT-PL spectra was increased for the CZTSe solar cells with a dielectric layer, which indicates that the Al2O3 or SiO2 dielectric layer passivated the bottom interface effectively.[88] Different from the study of Vermang[87] and Kim[88], Liu et al. sputtered a Al2O3 intermediate layer on the Mo back contact, and then the Al2O3 intermediate layer turns into a self-organized nanopattern with a nanoscale opening between CZTS and MoS2for their connection.[89]The Al2O3 intermediate layer prevents the interface reaction between CZTS and Mo at the initial stage of sulfurization, and thus the thickness of MoS2 is reduced, the voids and ZnS secondary phase at the back contact region are eliminated. And then the initially continuous Al2O3 intermediate layer breaks into discontinuous aggregates, producing self-organized nanopattern with nanoscale opening.[89] The CZTS absorber and MoS2 can connect directly at the openings and it is beneficial to the electrical contact and avoids high series resistance of the solar cell devices. Besides, the crystallinity of CZTS absorber becomes better, and the minority lifetime is increased. As a result, the conversion efficiency of the ultrathin CZTS solar cells is improved from 7.34% to 8.65%.[89] An electrostatic field at the back contact can also prevent rear recombination, which can be created by depositing a high work function (HWF) material, such as MoO 3, WO3, etc., between CZTSSe absorber and Mo back contact. It is indicated that the work function of MoO3 is 6.5 eV, which is much higher than that of CZTSSe (5.2 eV).[90] Thus, electrons flow from the lower work function CZTSSe to the higher work function MoO3 to equilibrate the Fermi level while MoO3 and CZTSSe are contacted and thus an electrostatic field is created. Thereby, electrons will be driven to the front hetero-junction, holes will be attracted to the back contact, and the electron-hole recombination at back contact will be reduced, as shown in Fig. 13. Antunez et al. presented a method to modify the back contact by exfoliating the fully CZTSSe devices from Mo/glass substrate, followed by the deposition of HWF MoO3 and reflective Au contact.[90] The wxAMPS simulation results indicate that the VOC increases with the absorber thickness decrease while the high work function material (MoO3) are applied and it is true for actual devices.[90] It is because that more electron-hole pairs are created at and near the back contact with the decrease of the thickness of absorber. The carrier recombination is reduced when an electrostatic field created by HWF MoO3. It is indicated that the VOC increases of up to 49 mV for the non-etched samples, and even up to 61 mV for the bromin-methanol etched samples with 1 µm absorber thickness.[90] Ranjbar et al. also introduced an ultra-thin MoO3 layer between CZTSe absorber and Mo back contact and found that the minority carrier life time and VOC are increased because of the decrease of rear recombination.[91] Besides, to minimize/eliminate the blocking of carrier transportation by voids, Zhou et al. introduced an ultrathin carbon layer between CZTS and Mo layer.[92] It is indicated that the carbon sticks on the inner wall of voids and connects the CZTS and MoS2 at the void region after sulphuration, which will boost the transportation of free carriers and improve the JSC.[92] Besides, the introduction of the ultrathin carbon intermediate layer has no obvious influence on the thickness of MoSe2 and the crystallinity of CZTS film, and thereby the VOC and FF are nearly unchanged.[92] 4.2. Passivation of front interface An ultrathin dielectric film, such asTiO 2 and Al2O3,can be introduced between CdS and CZTSSe, acting as a passivation layer to reduce the recombination at CdS/CZTSSe interface.[88, 93-96] Wu et al. introduced an atomic layer deposition (ALD) ultrathin TiO2 layer between CdS and CZTSSe to reduce the interface recombination.[93] It is indicated that the activation energy (EA) of the main recombination process of the CZTSSe device with TiO2 passivation layer is ca. 40 meV higher than that without a passivation layer, which suggests that the interfacial recombination is reduced for the passivation of TiO2 layer.[93] Besides, the DLCP and CV profiling suggests the density of defects at CdS/CZTSSe interface is reduced for the CZTSSe devices with TiO2 passivation layer.[93] As a result, the VOC, FF, and efficiency of CZTSSe solar cells with TiO2 layer are improved. However, the JSC is decreased slightly due to the impact of insulating TiO2 layer that the charge transportation is suppressed.[93] Therefore, it is critical to optimize the thickness of the passivation layer to minimize the blocking of charge transportation. It is convenient to control the thickness of the ultrathin passivation layer by ALD, while it is difficult for other deposition technique. Ranjbar et al. adopted a solution processed TiO2 layer to passivate the front interface.[94] The introduction of TiO2 layer improved the VOC, whereas, the FF and efficiency decreased which is because of a high barrier at the conduction band by the introduction of TiO2 layer and therefore the charge transportation is suppressed.[94] To boost the charge transportation, opens were created by an additional KCN treatment after TiO2 layer deposited, and the JSC and FF are also improved, and the efficiency was improved from 5.4% to 6.9%.[94] In addition, introduction of Al 2O3 layer is an effective interface-passivation strategy to improve device performance of CZTS(e) solar cells. Lee et al. introduced an ultrathin ALD- Al2O3film as a passivation layer.[95] They indicated that Al2O3 also penetrated into grain boundaries and pinholes of CZTSSe film to passivate these regions.[95] As the i-ZnO layer were replaced by Al2O3, the sputtering damage to CdS surface during TCO preparation was reduced and consequently the VOC were improved ca. 5%.[95] By introducing the aforementioned type of Al2O3 layer, the CZTSSe device performances were improved significantly and the highest conversion efficiency is 11.5%.[95] It is noticeable that the Al2O3 film can be etched by NH4OH during CdS deposition, which is different from the stable of TiO2 film in the CBD environment.[94] Therefore, the series resistance (RS) is almost the same with the thickness increase of Al2O3 layer between CZTS and CdS, whereas the RS obviously increases with the thickness increase of Al2O3 layer between CdS and ITO.[95] Erkan et al. also introduced ALD- Al2O3 layer between CdS and CZTSSe and found that the density of acceptor-like defects is reduced and the depletion width is widened. As a result, VOC are improved.[96] Kim et al. indicated that there is a strong accumulation of Cu and some deficiency of Cd near the interface of CdS/CZTSe, i.e. the intermixing of Cu and Cd is severe between CdS and CZTSe during the CdS deposition (shown in Fig. 14(a)).[88] In CIGS solar cells, the Cd doping and the formation of CdCu donor defects near the CdS/CIGS interface will cause a large band bending and convert the CIGS surface to n type which is beneficial to the separation of photo- generated carries.[97-98] In kesterite solar cells, however, it is disclosed that neutral defect cluster (CdCu + VCu) is most easily formed during CdS deposition on CZTS(e) film, and a small amount of donor defect CdCu and neutral defect CdZn are formed.[99] Thus, the surface of CZTS(e) film is difficult to convert to n-type, and thereby the benefits of intermixing of Cu and Cd near CdS/CZTS(e) interface is ambiguous. Kim et al. introduced a 5 nm Al2O3 layer between CdS and CZTSe, which effectively blocked the inter-diffusion of Cu and Cd (shown in Fig. 14(b)).[88] Besides, the intensity increase of PL spectra indicated that the non-radiative recombination is suppressed for the sample with Al2O3 passivation layer (shown in Fig. 14(d)). Finally, VOC and FF were significantly improved, which indicates that the server intermixing between CdS and CZTSe is detrimental to photovoltaic performances of CZTSe solar cells.[88] The density of interface defects can be reduced by epitaxial growth of buffer layer on CZTS(e) absorber. Liu et al. achieved the epitaxial growth of CdS on CZTS via sulfurization in combined SnS & S atmosphere, or sulfurization in only S atmosphere with surface defects but passivated by air annealing.[100] Their results indicated that the epitaxial growth of CdS leads to a low interfacial defects density and thereby low recombination rate. So, the minority lifetime is increased and the device efficiency of CZTS solar cells are improved to 8.76%.[100] Although the density of interface defects can be reduced by epitaxial growth of CdS on CZTS, the lattice mismatch between CdS and CZTS is much larger (ca. 7%) than that between CdS and CIGS (ca. 1.5%) or CZTSe (ca. 2.4%), which is detrimental to the epitaxial growth CdS on CZTS.[101] CeO2 has a nearly perfect lattice match with CZTS (ca. 0.4%) and the interface defects can be decreased significantly. Besides, the conduction band offset (CBO) between CeO2 and CZTS is -0.12 ± 0.20 eV, which is only slightly below the optimal CBO range, and it is more favorable than the CdS/CZTS CBO. Thus, Crovetto et al. proposed that the interface recombination can be reduced by the introduction of a CeO2 layer between CdS and CZTS.[101] As a result, the VOC is improved. However, the electron effective mass of CeO2 is very high, and thereby the electron mobility is very low.[101] As a result, the photocurrent transportation is suppressed and the JSC is reduced. Thus, the thickness of CeO2 should be ultrathin and controlled precisely. It is better that the ultrathin CeO2 layer is prepared via atomic layer deposition. In summary, the interfacial defects will increase the interfacial recombination, so the passivation of interface is a key point for high efficient CZTS(e) solar cells. The interface can be passivated by an ultrathin dielectric layer, including Al2O3, TiO2, and SiO2 etc., and the epitaxial growth of buffer on CZTS(e) absorber in some extent. AHWF material, including MoO3 and WO3 electrostatic field at back contact, and thereby promote the transportation of charge electrons. It , etc., can be inserted between CZTSSe absorber and Mo back contact to form an needs to further explore the passivation for the interface to improve the transportation of charge carriers. 5. Etching of secondary phases CuInSe It is calculated that the phase stable regions of CZTS and CZTSe are much smaller than that of 2 (CISe) because of the much more competing secondary phases, as shown in Fig. 15.[102] Compared with CISe, although the stable chemical potential range of µ Cu (-0.4 to 0 eV) in kesteriteis slightly smaller than that of CISe (-0.5 to 0), the range of µ Zn (ca. 0.2 eV) and µ Sn (ca. 0.6 eV) are much smaller than that of µ In (ca. 1.0 eV).[102] As a result, various secondary phases, such as CuS(e), Cu2S(e), Cu2SnS(e)3, ZnS(e), SnS(e), and SnS(e)2, are easily to coexist in CZTS(e) absorber. A deficit of Zn would lead to the segregationof Cu2SnS(e)3, SnS(e), and SnS(e)2.[102] The band gap of Cu2SnS(e)3 and SnS(e) are smaller than that of CZTS(e), which will increase the VOC deficit. The conduction band edge energy of SnS(e)2 is much lower than that of CZTS(e).[103-104] Therefore, the SnS(e)2 secondary phases tend to trap minority charge carriers, produce shunt paths, and thereby deteriorate the VOC and FF of CZTS(e) devices.[105-107] On the other hand, an excess of Zn will lead to the segregation of ZnS(e). ZnS(e) secondary phases are less detrimental to device performance as they are on the back of CZTS(e). However, the presence of ZnS(e) on the surface of CZTS(e) will inhibit the transportation of charge carriers because of the big spike CBO (1.32 eV) between ZnS(e) and CZTS(e), causing the decrease of JSC and the increase of RS.[108-112] Besides, the presence of CuxS(e) secondary phases can introduce shunt paths for its high conductivity.[107, 113-115] It is indicated that the secondary phases tend to accumulate near the top or bottom of CZTS(e) absorber, which will increase the interface recombination and thereby deteriorate the performance of CZTS(e) solar cells.[116-117] Therefore, the composition control, the avoidance of the formation of secondary phases, and the removal of secondary phases are crucial for high efficient CZTS(e) solar cells. The secondary phases at the bottom of CZTS(e) can be eliminated via suppressing the reaction between CZTS(e) and Mo by an intermediate layer, such as TiN, Al2O3, and ZnO, etc., or modifying the sulfuration/selenization process.[64, 78-80, 87, 118] As for the secondary phases on the front surface of kesterite film, chemical etching is an effective way to remove them. Cu xS(e) secondary phases are usually removed by potassium cyanide (KCN) etching.[119-122] The measurement of Kelvin probe force microscopy (KPFM) and Raman scattering spectroscopy suggested that the CuxS(e) can be removed completely by KCN etching.[119-121] KCN is toxic and hazardous, thus the usage content should be very small. Buffière adopted a 5 % KCN aqueous solution containing 0.5 wt % KOH for safety reasons.[122] The FF of CZTSe solar cell are improved after 30 s of etching. However, they found that elements Zn, Sn, and Se are preferentially removed from CZTSe surface by KOH, causing an increase of Cu content at CZTSe surface after long etching time.[122] The Cu-rich surface will increase the recombination rate at CdS/CZTSe interface and thereby deteriorate the FF of CZTSe solar cells. Therefore, the KCN etching time should be less than 30 s and/ or the concentration of KOH should be reduced to improve the performance of CZTSe solar cells.[122] Other than the removal of CuxS(e) phases, KCN etching can improve the performance of CZTS(e) solar cells in other ways. Bär et al. indicated that the surface composition will be changed after KCN etching. Therefore, the surface band gap will be increased from 1.53 ± 0.15 eV to 1.91 ± 0.15 eV, which will enhance the cliff- like conduction band offset at CdS/CZTS.[16] Then the energetic barrier for charge carrier recombination will be increased and thereby the VOC will be improved.[16] Besides, KCN etching can also remove the surface overlayer (including native oxides and airborne contamination) completely, which will reduce the density of contamination-induced surface defects and the interface recombination.[123] In addition, Pinto et al. reported the Cu2-x(S,Se) secondary phase can be removed completely by a safer chemical mixture of ethylenediamine and 2- mercaptoethanol.[124] It is indicated that H2O2 can also effectively remove CuxS phases following by[125] 𝐶𝐶𝐶𝐶2+𝑍𝑍2−+4𝐻𝐻2𝑂𝑂2→𝐶𝐶𝐶𝐶𝑍𝑍𝑂𝑂4+4𝐻𝐻2𝑂𝑂 (3) 𝐶𝐶𝐶𝐶2+𝑍𝑍2−+4𝐻𝐻2𝑂𝑂2→𝐶𝐶𝐶𝐶2𝑍𝑍𝑂𝑂4+4𝐻𝐻2𝑂𝑂 (4) ZnS secondary phases can be selectively removed by HCl solution.[126] Fairbrother et al. reported the efficiency of CZTS solar cells are improved from 2.7% to 5.2% after HCl etching.[126] As for ZnSe secondary phase, Mousel et al. adopted the HCl solution to etch the surface ZnSe secondary phases and improved the device efficiency from about 4% to above 5%.[127] However, ZnSe secondary phases are mostly removed via an oxidizing route using KMnO4/H2SO4 followed by a Na2S etching.[128] Firstly, ZnSe phases are oxided to Se0, and then the Se0 phases are removed by Na2S, as shown in Table 1. López-Marino used three different oxidizing agents, H2O2, KMnO4, and K2Cr2O7, to etch ZnSe and found that KMnO4 is more effective than others.[128] The device performances are imprvoed significantly after ZnSe etching. The improvement of JSC and series resistance can be attributed to minimizing the blocking of charge carriers transportationdue to the ZnSe secondary phases. Besides, the improvement of VOC, FF, and Rsh can be ascribed to a chemical passivation of the CZTSe absorber surface. SnS(e) and SnS(e) 2 can be removed from the CZTS(e) surface by (NH4)2S etching.[106, 109, 129- 131] The removal of Sn-Se phases were demonstrated by comparing the Sn content, Sn-Se XRD peaks, and Raman signals before and after etching.[129] Xie et al. indicated that the (NH4)2S solution may also passivate the surface of CZTSe absorber via removal of surface oxides and formation of S passivated species other than removal of Sn-Se scondary phases. As a result, all J- V parameters of CZTSe solar cells are improved and the conversion efficiency increases between 20 and 65% after (NH4)2S etching.[129] However, it is indicated that the longer etching time tend to deteriorate the device performance because of the preferred etching of Se, Sn, and Zn, causing the increase of Cu content at the CdS/CZTSe interface.[130] Wei et al. presented the removal of SnS secondary phases using Na2S solution, which were certified by Raman scattering mapping and PL spectra.[132] Ren et al. indicated that the SnS secondary phases on CZTS surface are detrimental to solar cells, whereas the SnS on CZTS rear can passivate the back contact, because of the spike like CBO between CZTS and SnS, limiting electron transportation toward the back contact, as shown in Fig. 16.[133] Therefore, the formation location of SnS secondary phase should be carefully controlled. As for Cu2SnS(e)3 secondary phases, they can be etched in Br in methanol (Br2-MeOH).[127, 134] In summary, the stable region of CZTS and CZTSe are too small, which causes the presence of secondary phases. The secondary phases are detrimental to the photovoltaic performances of CZTS(e) solar cells. Chemical etching is an effective way to remove the secondary phases on the surface of CZTS(e) absorbers. The photovoltaic performances of CZTS(e) solar cells can be improved after the removal of secondary phases. 6. Summary and perspective The efficiency of CZTSSe solar cells has reached at 12.6%, which is fabricated by solution method. However, the efficiency cannot be improved for several years because of high Voc deficit, which is related to many aspects. Interface is one of such aspects. In this review, we discussed the problems at interfaces of CZTS(e) solar cells, which are detrimental to the photovoltaic performances, and the corresponding solutions to overcome the problems and improve the photovoltaic performances. First, the conduction band alignment at CdS/CZTS(e) interface is not optimal. The CBO between CdS and CZTS or CZTSe is cliff like or a slightly bigger spike barrier, respectively, which is detrimental to VOC or JSC, respectively. The CBO can be optimized by modify the composition of CZTS(e) absorber or adopt an alternative buffer layer, including Zn(O,S), (Zn,Cd)S, Zn1-xSnxOy, Zn1-xMgxO, and In2S3,etc.. Secondary, over thick MoS(e)2 layer is easily to be formed during selenization/sulphuration process at high temperature under high Se/S pressure, so the thickness of MoS(e)2 layer between CZTS(e) absorber and Mo back contact should be tailored. The formation of MoS(e)2 can be suppressed by a dense temporary precursor, a MoS(e)2 layer with titled Se-Mo-Se sheets, or an intermediate barrier layer, including TiN, TiB2, and MoO2, etc.. Thirdly, the interfacial defects will increase the interfacial recombination, so interfacial passivation is necessary. A dielectric layer, including Al2O3, TiO2, and SiO2 etc., can be used to passivate the front and rear interface. An electrostatic field at back contact, created by depositing a high work function (HWF) material (such as MoO3, WO3, etc.) between CZTS(e) absorber and Mo back contact, can also prevent rear recombination. Besides, the epitaxial growth of buffer on CZTS(e) absorber will reduce the interfacial recombination in some extent. Finally, secondary phases are easily to be segregated due to the small stable regions of CZTS(e). Chemical etching is an effective way to remove the secondary phases on the surface of CZTS(e) absorbers. The photovoltaic performances of CZTS(e) solar cells can be improved after the removal of secondary phases. Although various methods have been adopted to improve the interfacial properties, the VOC- deficit and efficiency of CZTS(e) solar cells still have a big gap with their S-Q limit and that of CIGS solar cells. So, more research is necessary, including adopting a new preparation process (like MOCVD) of CZTS(e) absorber to improve the interfacial properties, adopting a new buffer layer to optimize the band alignment, and adopting a new passivation method and passivation material to passivate the defects at interfaces, etc..The main advantage of MOCVD over other growth techniques is the low growth temperature, thereby the decomposition of CZTS(e) absorber can be avoided, resulting in the high purity of the material and the absence of secondary phases. References [1] Jackson P, Wuerz R, Hariskos D, Lotter E, Witte W, and Powalla M 2016 Phys. Status Solidi (RRL) - Rapid Res. Lett. 10 583 [2] Candelise C, Winskel M, and Gross R 2012 Prog. Photovolt: Res. Appl. 20 816 [3] Phipps G, Mikolajczak C, and Guckes T 2008 Renewable Energy Focus 9 56 [4] Mitzi D B, Gunawan O, Todorov T K, and Barkhouse D A R 2013 Philos. Trans. R. Soc. A 371 20110432 [5] Bag S, Gunawan O, Gokmen T, Zhu Y, Todorov T K, and Mitzi D B 2012 Energy Environ. Sci. 5 7060 [6] Katagiri H, Sasaguchi N, Hando S, Ohashi J, and Yokota T 1997 Sol. Energy Mater. Sol. Cells 49 407 [7] Haight R, Barkhouse A, Gunawan O, Shin B, Copel M, Hopstaken M, and Mitzi D B 2011 Appl. Phys. Lett. 98 253502 [8] Wang W, Winkler M T, Gunawan O, Gokmen T, Todorov T K, Zhu Y, and Mitzi D B 2014 Adv. Energy Mater. 4 1301465 [9] Siebentritt S 2013 Thin Solid Films535 1 [10] Gunawan O, Todorov T K, and Mitzi D B 2010 Appl. Phys. Lett. 97 233506 [11] Nadenau V, Rau U, Jasenek A, and Schock H W 2000 J. Appl. Phys. 87 584 [12] Turcu M, Pakma O, and Rau U 2002 Appl. Phys. Lett. 80 2598 [13] Scheer R 2009 J. Appl. Phys.105 104505 [14] Huang T J, Yin X, Qi G, and Gong H 2014 physica status solidi (RRL) - Rapid Research Letters 08 735 [15] Minemotoa T, Matsuia T, Takakuraa H, Hamakawaa Y, Negamib T, Hashimotob Y, Uenoyamab T, and Kitagawab M 2001 Sol. Energy Mater. Sol. Cells 67 83 [16] Bär M, Schubert B A, Marsen B, Wilks R G, Pookpanratana S, Blum M, Krause S, Unold T, Yang W, Weinhardt L, Heske C, and Schock H W 2011 Appl. Phys. Lett. 99 222105 [17] Tajima S, Kataoka K, Takahashi N, Kimoto Y, Fukano T, Hasegawa M, and Hazama H 2013 Appl. Phys. Lett.103 243906 [18] Li J, Du Q, Liu W, Jiang G, Feng X, Zhang W, Zhu J, and Zhu C 2012 Electron. Mater. Lett. 8 365 [19] Li J, Wei M, Du Q, Liu W, Jiang G, and Zhu C 2013 Surf. Interface Anal. 45 682 [20] Santoni A, Biccari F, Malerba C, Valentini M, Chierchia R, and Mittiga A 2013 J. Phys. D: Appl. Phys. 46 175101 [21] Udaka Y, Takaki S i, Isowaki K, Nagai T, Kim K M, Kim S, Tampo H, Shibata H, Matsubara K, Niki S, Sakai N, Kato T, Sugimoto H, and Terada N 2017 Phys. Status Solidi C 14 1600178 [22] Bao W, and Ichimura M 2012 Jpn. J. Appl. Phys. 51 10NC31 [23] Palsgaard M L N, Crovetto A, Gunst T, Markussen T, Hansen O, Stokbro K, and Brandbyge M 2016 International Conference on. IEEE Simulation of Semiconductor Processes and Devices (SISPAD), September 6-8, 2016, Nuremberg, Germany, p. 377 [24] Kato T, Hiroi H, Sakai N, and Sugimoto H 2013 28th European Photovoltaic Solar Energy Conference, September 30, 2013, Paris, France, p. 2125 [25] Neuschitzer M, Sanchez Y, López-Marino S, Xie H, Fairbrother A, Placidi M, Haass S, Izquierdo-Roca V, Perez-Rodriguez A, and Saucedo E 2015 Prog. Photovolt: Res. Appl. 23 1660 [26] Chua R H, Li X, Walter T, Teh L K, Hahn T, Hergert F, Mhaisalkar S, and Wong L H 2016 Appl. Phys. Lett. 108 043505 [27] Meyer B K, Polity A, Farangis B, He Y, Hasselkamp D, Krämer T, and Wang C 2004 Appl. Phys. Lett. 85 4929 [28] Li J, Liu X, Liu W, Wu L, Ge B, Lin S, Gao S, Zhou Z, Liu F, Sun Y, Ao J, Zhu H, Mai Y, and Zhang Y 2017 Solar RRL 1 1700075 [29] Kamada R, Yagioka T, Adachi S, Handa A, Tai K F, Kato T, and Sugimoto H 2016 IEEE 43rd Photovoltaic Specialists Conference (PVSC), June, 2016, Portland, OR. USA, p. 5 [30] Barkhouse D A R, Haight R, Sakai N, Hiroi H, Sugimoto H, and Mitzi D B 2012 Appl. Phys. Lett. 100 193904 [31] Sharbati S, and Sites J R 2014 IEEE Journal of Photovoltaics 4 697 [32] Sharbati S, Keshmiri S H, McGoffin J T, and Geisthardt R 2014 Appl. Phys. A 118 1259 [33] Neuschitzer M, Lienau K, Guc M, Barrio L C, Haass S, Prieto J M, Sanchez Y, Espindola- Rodriguez M, Romanyuk Y, Perez-Rodriguez A, Izquierdo-Roca V, and Saucedo E 2016 J. Phys. D: Appl. Phys.49 125602 [34] Grenet L, Grondin P, Coumert K, Karst N, Emieux F, Roux F, Fillon R, Altamura G, Fournier H, Faucherand P, and Perraud S 2014 Thin Solid Films 564 375 [35] Sakai N, Hiroi H, and Sugimoto H 2011 37th IEEE Photovol. Special. Conf. (PVSC), June 19-24, 2011, Seattle, WA, USA, p. 003654 [36] Hiroi H, Sakai N, Muraoka S, Katou T, and Sugimoto H 2012 June 3-8, 2012, Austin, TX, USA, p. 001811 [37] Bao W, Sachuronggui, and Qiu F-Y 2016 Chin. Phys. B. 25 127102 [38] Sun K, Yan C, Liu F, Huang J, Zhou F, Stride J A, Green M, and Hao X 2016 Adv. Energy Mater. 6 1600046 [39] Messaoud K B, Buffière M, Brammertz G, Lenaers N, Boyen H-G, Sahayaraj S, Meuris M, Amlouk M, and Poortmans J 2017 J. Phys. D: Appl. Phys. 50 285501 [40] Platzer-Björkman C, Frisk C, Larsen J K, Ericson T, Li S Y, Scragg J J S, Keller J, Larsson F, and Törndahl T 2015 Appl. Phys. Lett. 107 243904 [41] Ericson T, Larsson F, Törndahl T, Frisk C, Larsen J, Kosyak V, Hägglund C, Li S, and Platzer-Björkman C 2017 Solar RRL 1 1700001 [42] Tajima S, Umehara M, and Mise T 2016 Jpn. J. Appl. Phys. 55 112302 [43] Hironiwa D, Matsuo N, Sakai N, Katou T, Sugimoto H, Chantana J, Tang Z, and Minemoto T 2014 Jpn. J. Appl. Phys. 53 106502 [44] Hironiwa D, Chantana J, Sakai N, Kato T, Sugimoto H, and Minemoto T 2015 Curr. Appl Phys.15 383 [45] Hironiwa D, Matsuo N, Chantana J, Sakai N, Kato T, Sugimoto H, and Minemoto T 2015 physica status solidi (a) 212 2766 [46] Sandoval-Paz M G , Sotelo-Lerma M , Valenzuela-Jáuregui J J , Flores-Acosta M , and Ramı(cid:127)rez-Bon R 2005 Thin Solid Films 472 5 [47] Yan C, Liu F, Song N, Ng B K, Stride J A, Tadich A, and Hao X 2014 Appl. Phys. Lett. 104 173901 [48] Yu J, Zheng Z, Dong L, Cheng S, Lai Y, Zheng Q, Zhou H, Jia H, and Zhang H 2017 Chin. Phys. B. 26 046802 [49] Buffière M, Barreau N, Brammertz G, Sahayaraj S, Meuris M, and Poortmans J 2015 June 14-19, 2015, New Orleans, LA, USA, p. 1 [50] Khadka D B, Kim S, and Kim J 2015 J. Phys. Chem. C 119 12226 [51] Kim J, Hiroi H, Todorov T K, Gunawan O, Kuwahara M, Gokmen T, Nair D, Hopstaken M, Shin B, Lee Y S, Wang W, Sugimoto H, and Mitzi D B 2014 Adv. Mater. 26 7427 [52] Hiroi H, Sakai N, Kato T, and Sugimoto H 2013 June 16-21, 2013, Tampa, FL, USA, p. 0863 [53] Yan C, Liu F, Sun K, Song N, Stride J A, Zhou F, Hao X, and Green M 2016 Sol. Energy Mater. Sol. Cells 144 700 [54] Su C-Y, Liao K-H, Pan C-T, and Peng P-W 2012 Thin Solid Films 520 5936 [55] Li Z-H, Cho E-S, and Kwon S J 2011 Appl. Surf. Sci. 257 9682 [56] Pethe S A, Takahashi E, Kaul A, and Dhere N G 2012 Sol. Energy Mater. Sol. Cells 100 1 [57] Salomé P M P, Malaquias J, Fernandes P A, and Cunha A F d 2010 J. Phys. D: Appl. Phys. 43 345501 [58] Wu H-M, Liang S-C, Lin Y-L, Ni C-Y, Bor H-Y, Tsai D-C, and Shieu F-S 2012 Vacuum 86 1916 [59] Spies J A, Schafer R, Wager J F, Hersh P, Platt H A S, Keszler D A, Schneider G, Kykyneshi R, Tate J, Liu X, Compaan A D, and Shafarman W N 2009 Sol. Energy Mater. Sol. Cells 93 1296 [60] Wada T, Kohara N, Nishiwaki S, and Negami T 2001 Thin Solid Films 387 118 [61] Zhang X, Kobayashi M, and Yamada A 2017 ACS Appl Mater Interfaces 9 16215 [62] Li J, Zhang Y, Zhao W, Nam D, Cheong H, Wu L, Zhou Z, and Sun Y 2015 Adv. Energy Mater. 5 1402178 [63] Scragg J J, Dale P J, Colombara D, and Peter L M 2012 Chemphyschem 13 3035 [64] Scragg J J, Kubart T, Wätjen J T, Ericson T, Linnarsson M K, and Platzer-Björkman C 2013 Chem. Mater. 25 3162 [65] Scragg J J, Watjen J T, Edoff M, Ericson T, Kubart T, and Platzer-Bjorkman C 2012 J. Am. Chem. Soc. 134 19330 [66] Shin B, Zhu Y, Bojarczuk N A, Jay Chey S, and Guha S 2012 Appl. Phys. Lett. 101 053903 [67] Yao L, Ao J, Jeng M-J, Bi J, Gao S, Sun G, He Q, Zhou Z, Sun Y, and Chang L-B 2017 Sol. Energy Mater. Sol. Cells 159 318 [68] Shin S W, Gurav K V, Hong C W, Gwak J, Choi H R, Vanalakar S A, Yun J H, Lee J Y, Moon J H, and Kim J H 2015 Sol. Energy Mater. Sol. Cells 143 480 [69] Shin B, Bojarczuk N A, and Guha S 2013 Appl. Phys. Lett. 102 091907 [70] Xiao Z Y, Yao B, Li Y F, Ding Z H, Gao Z M, Zhao H F, Zhang L G, Zhang Z Z, Sui Y R, and Wang G 2016 ACS Appl. Mater. Interfaces 8 17334 [71] Liu F, Sun K, Li W, Yan C, Cui H, Jiang L, Hao X, and Green M A 2014 Appl. Phys. Lett. 104 051105 [72] Schnabel T, and Ahlswede E 2017 Sol. Energy Mater. Sol. Cells 159 290 [73] Duchatelet A, Savidand G, Vannier R N, and Lincot D 2013 Thin Solid Films 545 94 [74] Nina K, Kimura Y, Yokoyama K, Kido O, Binyo G, and Kaito C 2008 Physica E 40 2995 [75] Werfel F, and Minni E 1983 J. Phys. C: Solid State Phys. 16 6091 [76] Lopez-Marino S, Espíndola-Rodríguez M, Sánchez Y, Alcobé X, Oliva F, Xie H, Neuschitzer M, Giraldo S, Placidi M, Caballero R, Izquierdo-Roca V, Pérez-Rodríguez A, and Saucedo E 2016 Nano Energy 26 708 [77] Würz R, Fuertes Marrón D, Meeder A, Rumberg A, Babu S M, Schedel-Niedrig T, Bloeck U, Schubert-Bischoff P, and Lux-Steiner M C 2003 Thin Solid Films 431-432 398 [78] López-Marino S, Placidi M, Pérez-Tomás A, Llobet J, Izquierdo-Roca V, Fontané X, Fairbrother A, Espíndola-Rodríguez M, Sylla D, Pérez-Rodríguez A, and Saucedo E 2013 J. Mater. Chem. A 1 8338 [79] Li W, Chen J, Cui H, Liu F, and Hao X 2014 Mater. Lett. 130 87 [80] Liu X, Cui H, Li W, Song N, Liu F, Conibeer G, and Hao X 2014 physica status solidi (RRL) - Rapid Research Letters 8 966 [81] Gao S, Zhang Y, Ao J, Lin S, Zhang Z, Li X, Wang D, Zhou Z, Sun G, Liu F, and Sun Y 2017 Sol. Energy Mater. Sol. Cells (doi.org/10.1016/j.solmat.2017.10.011) [82] Zhao J, Wang A, and Green M A 1999 Prog. Photovolt: Res. Appl. 7 471 [83] Saint-Cast P, Benick J, Kania D, Weiss L, Hofmann M, Rentsch J, Preu R, and Glunz S W 2010 IEEE Electron Device Lett. 31 695 [84] Martı(cid:127)n I, Vetter M, Orpella A, Puigdollers J, Cuevas A, and Alcubilla R 2001 Appl. Phys. Lett. 79 2199 [85] Schmidt J, Kerr M, and Cuevas A 2001 Semicond. Sci. Technol. 16 164 [86] Dullweber T, Gatz S, Hannebauer H, Falcon T, Hesse R, Schmidt J, and Brendel R 2012 Prog. Photovolt: Res. Appl. 20 630 [87] Vermang B, Ren Y, Donzel-Gargand O, Frisk C, Joel J, Salome P, Borme J, Sadewasser S, Platzer-Bjorkman C, and Edoff M 2016 IEEE J. Photovolt. 6 332 [88] Kim J, Park S, Ryu S, Oh J, and Shin B 2017 Prog. Photovolt: Res. Appl. 25 308 [89] Liu F, Huang J, Sun K, Yan C, Shen Y, Park J, Pu A, Zhou F, Liu X, Stride J A, Green M A, and Hao X 2017 NPG Asia Mater. 9 e401 [90] Antunez P D, Bishop D M, Lee Y S, Gokmen T, Gunawan O, Gershon T S, Todorov T K, Singh S, and Haight R 2017 Adv. Energy Mater. 7 1602585 [91] Ranjbar S, Brammertz G, Vermang B, Hadipour A, Cong S, Suganuma K, Schnabel T, Meuris M, da Cunha A F, and Poortmans J 2017 physica status solidi (a) 214 1600534 [92] Zhou F, Zeng F, Liu X, Liu F, Song N, Yan C, Pu A, Park J, Sun K, and Hao X 2015 ACS Appl. Mater. Interfaces 7 22868 [93] Wu W, Cao Y, Caspar J V, Guo Q, Johnson L K, McLean R S, Malajovich I, and Choudhury K R 2014 Appl. Phys. Lett. 105 042108 [94] Ranjbar S, Hadipour A, Vermang B, Batuk M, Hadermann J, Garud S, Sahayaraj S, Meuris M, Brammertz G, da Cunha A F, and Poortmans J 2017 IEEE J. Photovolt. 7 1130 [95] Lee Y S, Gershon T, Todorov T K, Wang W, Winkler M T, Hopstaken M, Gunawan O, and Kim J 2016 Adv. Energy Mater. 6 1600198 [96] Erkan M E, Chawla V, and Scarpulla M A 2016 J. Appl. Phys. 119 194504 [97] Liao D, and Rockett A 2003 J. Appl. Phys. 93 9380 [98] Liao D, and Rockett A 2003 Appl. Phys. Lett. 82 2829 [99] Maeda T, Nakamura S, and Wada T 2012 Jpn. J. Appl. Phys. 51 10NC11 [100] Liu F, Yan C, Huang J, Sun K, Zhou F, Stride J A, Green M A, and Hao X 2016 Adv. Energy Mater. 6 1600706 [101] Crovetto A, Yan C, Iandolo B, Zhou F, Stride J, Schou J, Hao X, and Hansen O 2016 Appl. Phys. Lett. 109 233904 [102] Chen S, Walsh A, Gong X G, and Wei S H 2013 Adv. Mater. 25 1522 [103] Schlaf R, Pettenkofer C, and Jaegermann W 1999 J. Appl. Phys. 85 6550 [104] Chen S, Walsh A, Yang J-H, Gong X G, Sun L, Yang P-X, Chu J-H, and Wei S-H 2011 Phys. Rev. B 83 125201 [105] Temgoua S, Bodeux R, Naghavi N, and Delbos S 2015 Thin Solid Films 582 215 [106] Vauche L, Risch L, Arasimowicz M, Sánchez Y, Saucedo E, Pasquinelli M, Goislard de Monsabert T, Grand P P, and Jaime-Ferrer S 2016 J. Renewable Sustainable Energy 8 033502 [107] Vora N, Blackburn J, Repins I, Beall C, To B, Pankow J, Teeter G, Young M, and Noufi R 2012 J. Vac. Sci. Technol. A 30 051201 [108] Schwarz T, Cojocaru-Mirédin O, Choi P, Mousel M, Redinger A, Siebentritt S, and Raabe D 2013 Appl. Phys. Lett. 102 042101 [109] Xie H, Dimitrievska M, Fontané X, Sánchez Y, López-Marino S, Izquierdo-Roca V, Bermúdez V, Pérez-Rodríguez A, and Saucedo E 2015 Sol. Energy Mater. Sol. Cells 140 289 [110] Timo Wätjen J, Engman J, Edoff M, and Platzer-Björkman C 2012 Appl. Phys. Lett. 100 173510 [111] Hsu W-C, Repins I, Beall C, DeHart C, Teeter G, To B, Yang Y, and Noufi R 2013 Sol. Energy Mater. Sol. Cells 113 160 [112] Nagoya A, Asahi R, and Kresse G 2011 J Phys Condens Matter 23 404203 [113] Abdullaev G B, Aliyarova Z A, and Asadov G A 1967 physica status solidi (b) 21 461 [114] Fuertes Marrón D, Glatzel T, Meeder A, Schedel-Niedrig T, Sadewasser S, and Lux- Steiner M C 2004 Appl. Phys. Lett. 85 3755 [115] Hsieh T-P, Chuang C-C, Wu C-S, Chang J-C, Guo J-W, and Chen W-C 2011 Solid-State Electron. 56 175 [116] Fairbrother A, Fontané X, Izquierdo-Roca V, Espíndola-Rodríguez M, López-Marino S, Placidi M, Calvo-Barrio L, Pérez-Rodríguez A, and Saucedo E 2013 Sol. Energy Mater. Sol. Cells 112 97 [117] Yin X, Tang C, Sun L, Shen Z, and Gong H 2014 Chem. Mater. 26 2005 [118] Nam D, Cho S, Sim J-H, Yang K-J, Son D-H, Kim D-H, Kang J-K, Kwon M-S, Jeon C-W, and Cheong H 2016 Sol. Energy Mater. Sol. Cells 149 226 [119] Tanaka T, Sueishi T, Saito K, Guo Q, Nishio M, Yu K M, and Walukiewicz W 2012 J. Appl. Phys. 111 053522 [120] Kim G Y, Kim J R, Jo W, Lee K D, Kim J Y, Nguyen T T T, and Yoon S 2014 Curr. Appl Phys. 14 1665 [121] Nguyen T T T, Shin H-Y, Kim G Y, Kim J R, Jo W, Yoon S, Lee K D, and Kim J Y 2015 J. Korean Phys. Soc. 66 117 [122] Buffière M, Brammertz G, Sahayaraj S, Batuk M, Khelifi S, Mangin D, El Mel A-A, Arzel L, Hadermann J, Meuris M, and Poortmans J 2015 ACS Appl. Mat. Interfaces 7 14690 [123] Erkan M E, Chawla V, Repins I, and Scarpulla M A 2015 Sol. Energy Mater. Sol. Cells 136 78 [124] Pinto A H, Shin S W, Aydil E S, and Penn R L 2016 Green Chem. 18 5814 [125] Chavda A, Patel M, Mukhopadhyay I, and Ray A 2016 ACS Sustainable Chem. & Engin. 4 2302 [126] Fairbrother A, Garcia-Hemme E, Izquierdo-Roca V, Fontane X, Pulgarin-Agudelo F A, Vigil-Galan O, Perez-Rodriguez A, and Saucedo E 2012 J. Am. Chem. Soc. 134 8018 [127] Mousel M, Redinger A, Djemour R, Arasimowicz M, Valle N, Dale P, and Siebentritt S 2013 Thin Solid Films 535 83 [128] López-Marino S, Sánchez Y, Placidi M, Fairbrother A, Espindola-Rodríguez M, Fontané X, Izquierdo-Roca V, López-García J, Calvo-Barrio L, Pérez-Rodríguez A, and Saucedo E 2013 Chem.-A Eur. J 19 14814 [129] Xie H, Sánchez Y, López-Marino S, Espíndola-Rodríguez M, Neuschitzer M, Sylla D, Fairbrother A, Izquierdo-Roca V, Pérez-Rodríguez A, and Saucedo E 2014 ACS Appl. Mat. Interfaces 6 12744 [130] Buffière M, Brammertz G, El Mel A-A, Barreau N, Meuris M, and Poortmans J 2017 Thin Solid Films 633 135 [131] Buffière M, Mel A-A E, Lenaers N, Brammertz G, Zaghi A E, Meuris M, and Poortmans J 2015 Adv. Energy Mater. 5 1401689 [132] Wei Z, Newman M J, Tsoi W C, and Watson T M 2016 Appl. Phys. Lett. 109 123902 [133] Ren Y, Richter M, Keller J, Redinger A, Unold T, Donzel-Gargand O, Scragg J J S, and Platzer Björkman C 2017 ACS Energy Letters 2 976 [134] Timmo K, Altosaar M, Raudoja J, Grossberg M, Danilson M, Volobujeva O, and Mellikov E 2010 35th IEEE Photovoltaic Specialists Conference (PVSC), June 20-25, 2010, Honolulu, HI, USA, p. 001982 [135] Lee Y S, Gershon T, Gunawan O, Todorov T K, Gokmen T, Virgus Y, and Guha S 2015 Adv. Energy Mater. 5 1401372 [136] Tajima S, Umehara M, Hasegawa M, Mise T, and Itoh T 2017 Prog. Photovolt: Res. Appl. 25 14 Figure captions Fig. 1.(color online) Temperature dependence of VOCfor CZTSSe and CIGSSe solar cells. Reproduced with permission from Ref. [10]. Fig. 2.(color online) Schema of band alignment at buffer/absorber interface: (a) cliff like, (b) and spike like. Spike and cliff are barriers for photo-generated and injection electrons, respectively. Reproduced with permission from Ref. [15]. Fig. 3.(color online) Schematic of the band alignment between CZTSSe and (a) ZnS and (b) ZnO, respectively. Reproduced with permission from Ref. [30]. Fig. 4.(color online) Band energy diagram of Zn(O,S) as a function of O/(S+O). Reproduced with permission from Ref. [31-32]. Fig. 5.(color online) Light and dark J-V curves of CZTSe solar cells with Zn(O,S) buffer layer prepared with different TU concentration during the CBD process and CdS as a reference buffer layer. Reproduced with permission from Ref. [33]. Fig. 6.(color online) Light J-V curves of CZTSe solar cells with Zn(O,S) buffer layer prepared with different TU concentration during the CBD process before and after light soaking for 260 min. Reproduced with permission from Ref. [33]. Fig. 7.(color online) The schematic diagrams of band alignments of the Zn(O,S)/CZTSe interface for (a) the Untreated sample and (b) the Etched&annealed sample. Reproduced with permission from Ref. [28]. Fig. 8.(color online) Schematic diagram of the band structure of the CIGS solar cells: (a) without MoSe2 layer, and (b) with MoSe2 layer. Fig. 9.(color online) Schematic of excess Se concentration profile across CZTSe/MoSe2 during annealing in a steady-state. Reproduced with permission from Ref. [69]. /Mo Fig. 10.(color online) Schematic drawings of the different back contact con figurations. Reproduced with permission from Ref. [72]. Fig. 11. (color online) Crystal structures of MoSe2 along different direction, which is created with CrystalMaker software: (a) (100) preferred orientation with Se-Mo-Se sheets perpendicular to the substrate, (b) (110) preferred orientation with Se-Mo-Se sheets perpendicular to the substrate, (c) (002) preferred orientation with Se-Mo-Se sheets parallel to the substrate, and (d) (103) preferred orientation with Se-Mo-Se sheets tilted to the substrate. Reproduced with permission from Ref. [81]. Fig. 12.(color online) (a) Schematic drawings of passivation layer with nano-patterned local contacts. (b) SEM images of contact holes on Mo after nanosphere lithography. (c) Cross- sectional TEM image of a bottom Al2O3 passivation layer (white dotted square) with a local contact between CZTSe and MoSe2. Insets are Fourier transform images taken by the area of near the surface (yellow dotted square) and the bulk (red dotted square), respectively. (d) J-V characteristics for the CZTSe solar cells with and wihout a (d) Al2O3 and (e) SiO2 passivation layer at the CZTSe/Mo interface. (f) 10 K photoluminescence spectra of CZTSe solar cells with and without an Al2O3 passivation layer at the CZTSe/Mo interface. Reproduced with permission from Ref. [88]. Fig. 13.(color online) Schematic band structure schematic of a CZTSSe solar cell with a high work function material between CZTSSe absorber and Mo back contact. Fig. 14.(color online) HAADF-STEM images and corresponding EDS mapping of Cu, Cd, and S of CZTSe solar cells (a) without and (b) with an Al2O3 passivation layer at the CZTSe/CdS interface. (c) Changes in Voc and efficiency as a function of Al2O3 thickness (solid lines are guides to eyes and a dashed line indicates a respective reference). (d) 10 K PL spectra of CZTSe solar cells with (purple) and without (green) a top Al2O3 passivation layer at the CZTSe/CdS interface. Reproduced with permission from Ref. [88]. Fig. 15.(color online) The calculated phase stable region of CISe (left), CZTS (center), and CZTSe (right). Reproduced with permission from Ref. [102]. Fig. 16.(color online) (a) Band edges at the interface of CZTS/SnS and (b) charge carrier transport paths close to CZTS/SnS interface. Reproduced with permission from Ref. [133]. Fig. 1.(color online) Temperature dependence of VOCfor CZTSSe and CIGSSe solar cells. Reproduced with permission from Ref. [10]. Fig. 2.(color online) Schema of band alignment at buffer/absorber interface: (a) cliff like, and (b) spike like, respectively. Spike and cliff are barriers for photo-generated and injection electrons, respectively. Reproduced with permission from Ref.[15]. Fig. 3.(color online) Schematic of the band alignment between CZTSSe and (a) ZnS and (b) ZnO, respectively. Reproduced with permission from Ref. [30]. Fig. 4.(color online) Band energy diagram of Zn(O,S) as a function of O/(S+O). Reproduced with permission from Ref. [31-32]. Fig. 5.(color online) Light and dark J-V curves of CZTSe solar cells with Zn(O,S) buffer layer prepared with different TU concentration during the CBD process and CdS as a reference buffer layer. Reproduced with permission from Ref. [33]. Fig. 6.(color online) Light J-V curves of CZTSe solar cells with Zn(O,S) buffer layer prepared with different TU concentration during the CBD process before and after light soaking for 260 min. Reproduced with permission from Ref. [33]. Fig. 7.(color online) The schematic diagrams of band alignments of the Zn(O,S)/CZTSe interface for (a) the Untreated sample and (b) the Etched & annealed sample. Reproduced with permission from Ref. [28]. Fig. 8.(color online) Schematic diagram of the band structure of the CIGS solar cells: (a) without MoSe2 layer, and (b) with MoSe2 layer. Fig. 9.(color online) Schematic of excess Se concentration profile across CZTSe/MoSe2/Mo during annealing in a steady-state. Reproduced with permission from Ref. [69]. Fig. 10.(color online) Schematic drawings of the different back contact configurations. Reproduced with permission from Ref. [72]. Fig. 11. (color online) Crystal structures of MoSe2 along different direction, which is created with CrystalMaker software: (a) (100) preferred orientation with Se-Mo-Se sheets perpendicular to the substrate, (b) (110) preferred orientation with Se-Mo-Se sheets perpendicular to the substrate, (c) (002) preferred orientation with Se-Mo-Se sheets parallel to the substrate, and (d) (103) preferred orientation with Se-Mo-Se sheets tilted to the substrate. Reproduced with permission from Ref. [81]. Fig. 12.(color online) (a) Schematic drawings of passivation layer with nano-patterned local contacts. (b) SEM images of contact holes on Mo after nanosphere lithography. (c) Cross- sectional TEM image of a bottom Al2O3 passivation layer (white dotted square) with a local contact between CZTSe and MoSe2. Insets are Fourier transform images taken by the area of near the surface (yellow dotted square) and the bulk (red dotted square), respectively. (d) J-V characteristics for the CZTSe solar cells with and wihout a (d) Al2O3 and (e) SiO2 passivation layer at the CZTSe/Mo interface. (f) 10 K photoluminescence spectra of CZTSe solar cells with and without an Al2O3 passivation layer at the CZTSe/Mo interface. Reproduced with permission from Ref. [88]. Fig. 13.(color online) Schematic band structure schematic of a CZTSSe solar cell with a high work function material between CZTSSe absorber and Mo back contact. Fig. 14.(color online) HAADF-STEM images and corresponding EDS mapping of Cu, Cd, and S of CZTSe solar cells (a) without and (b) with an Al2O3 passivation layer at the CZTSe/CdS interface. (c) Changes in Voc and efficiency as a function of Al2O3 thickness (solid lines are guides to eyes and a dashed line indicates a respective reference). (d) 10 K PL spectra of CZTSe solar cells with (purple) and without (green) a top Al2O3 passivation layer at the CZTSe/CdS interface. Reproduced with permission from Ref. [88]. Fig. 15.(color online) The calculated phase stable region of CISe (left), CZTS (center), and CZTSe (right). Reproduced with permission from Ref.[102]. Fig. 16.(color online) (a) Band edges at the interface of CZTS/SnS, and (b) charge carrier transport paths close to CZTS/SnS interface. Reproduced with permission from Ref. [133]. Table 1 A comparison of photovoltaic performance and VOC solar cells with their theoretical performance and with CIGS. SQL-CZTSSe refers to the deficit for record CZTS(e)-based theoretical performance of the CZTSSe solar cell. SQL- CZTSSe CIGS CZTSe g E [eV] 1.15 1.10 1.01 V OC [mV] 887 741 423 CZTSSe 1.13 513.4 CZTS 1.50 700 J SC [mA/cm2 42 37.8 40.6 35.2 21.3 ] FF [%] 89 Eff. [%] 32.8 80.6 22.6 67.3 11.6 69.8 12.6 63.0 9.4 VOC deficit [mV] 263 359 578 617 800 Ref. [9] [1] [135] [8] [136] Table 2 Summary of conduction band offset (CBO) at CdS/CZTS or CdS/CZTSe interface, respectively. Interface CBO [eV] -0.33 0.41 0.0 -0.06 -0.3 -0.15 0.48 0.34 0.55 CdS/CZTS CdS/CZTSe Ref. [16] [7] [17] [18] [20] [21] [7] [19] [21]
1912.13319	1	1912	2019-11-02T11:54:33	Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications	[ "physics.app-ph" ]	Transmission lines are essential components in various signal and power distribution systems. In addition to their main use as connecting elements, transmission lines can also be employed as continuous sensors for the measurement and detection of external influences such as mechanical strains and deformations. The measuring principle is based on deformation-induced changes of the characteristic impedance. Reflections of an injected test signal at resulting impedance mismatches can be used to infer applied deformations. To determine the effect of deformations on the characteristic impedance, we develop a numerical framework that allows us to solve Maxwell's equations for any desired transmission-line geometry over a wide frequency range. The proposed framework utilizes a staggered finite-difference Yee method on non-uniform grids to efficiently solve a set of decoupled partial differential equations that we derive from the frequency domain Maxwell equations. To test our framework, we compare simulation results with analytical predictions and corresponding experimental data. Our results suggest that the proposed numerical framework is able to capture experimentally observed deformation effects and may therefore be used in transmission-line-based deformation and strain sensing applications. Furthermore, our framework can also be utilized to simulate and study the electromagnetic properties of complex arrangements of conductor, insulator, and shielding materials.	physics.app-ph	physics	Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications Stefan H. Strub Institute for Theoretical Physics, ETH Zurich, 8093, Zurich, Switzerland LEONI Studer AG, Hohlstrasse 190, 8004, Zurich, Switzerland Lucas Bottcher Institute for Theoretical Physics, ETH Zurich, 8093, Zurich, Switzerland Center of Economic Research, ETH Zurich, 8092, Zurich, Switzerland LEONI Studer AG, Hohlstrasse 190, 8004, Zurich, Switzerland E-mail: [email protected] Abstract. Transmission lines are essential components in various signal and power distribution systems. In addition to their main use as connecting elements, transmission lines can also be employed as continuous sensors for the measurement and detection of external influences such as mechanical strains and deformations. The measuring principle is based on deformation-induced changes of the characteristic impedance. Reflections of an injected test signal at resulting impedance mismatches can be used to infer applied deformations. To determine the effect of deformations on the characteristic impedance, we develop a numerical framework that allows us to solve Maxwell's equations for any desired transmission-line geometry over a wide frequency range. The proposed framework utilizes a staggered finite-difference Yee method on non-uniform grids to efficiently solve a set of decoupled partial differential equations that we derive from the frequency domain Maxwell equations. To test our framework, we compare simulation results with analytical predictions and corresponding experimental data. Our results suggest that the proposed numerical framework is able to capture experimentally observed deformation effects and may therefore be used in transmission-line-based deformation and strain sensing applications. Furthermore, our framework can also be utilized to simulate and study the electromagnetic properties of complex arrangements of conductor, insulator, and shielding materials. Keywords: transmission line, numerical analysis, finite-difference method, skin effect, Bessel functions, Maxwell's equations, distributed sensing Submitted to: Meas. Sci. Technol. 9 1 0 2 v o N 2 ] h p - p p a . s c i s y h p [ 1 v 9 1 3 3 1 . 2 1 9 1 : v i X r a Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 2 1. Introduction it The transmission of electromagnetic waves is at the heart of many signal and power distribution applications. For wavelengths that are small compared to the conductor length, transmission lines such as coaxial cables or twisted pairs are used to limit radiative losses and reflections. In addition to their application in signal and power transmission systems, is also possible to use transmission lines as continuous sensors for the measurement and detection of different environmental influences [1, 2]. One important application is the measurement of mechanical strains and deformations to monitor the condition of civil engineering structures such as houses, bridges, and damns [1]. In contrast to discrete sensors (e.g., strain and displacement gauges), the advantage of continuous transmission-line sensors is the possibility to perform measurements at any location along a sensing line. The measuring principle in electrical transmission lines is based on two mechanisms: (i) Localized mechanical strains lead to transmission-line deformations and corresponding changes of the characteristic impedance. (ii) An injected test signal gets reflected at deformation- induced impedance mismatches and the information contained in signal reflections can be used to infer applied deformations. Electrical transmission lines are, however, not the only possibility to realize continuous strain and deformation sensors in practice. Another approach is based on optical fibres and Bragg grating and Brillouin scattering methods. The idea behind the Bragg grating method is to use optical gratings within optical fibres and measure wave-length changes that correspond to certain deformations. The disadvantage of this method is that it is only quasi-continuous and requires a large number of gratings to achieve good resolution [3]. In the Brillouin scattering technique, frequency shifts of the scattered light relative to the incident light are used to infer deformations. Despite the fact that this method does not require discrete gratings, the possible resolution is limited to about 10 cm [4, 5]. In addition to the low spatial resolution, another drawback of optical methods is that they may not be suitable for cost-sensitive applications. Electrical transmission lines, on the other hand, can be fabricated more cost- efficiently and allow for resolutions of greater than 1 cm [1]. To measure deformations with electrical transmis- sion lines, it is necessary to determine the effect of de- formations on the characteristic impedance. However, existing models only capture concentric transmission- line deformations and thus neglect any further shape- specific information [1]. To accurately describe the effect of arbitrary deformations on the characteristic impedance, we develop a numerical framework to solve Maxwell's equations in the frequency domain for any desired transmission-line geometry over a wide fre- quency range. Our framework is based on a staggered finite-difference Yee method on non-uniform grids and solves a set of decoupled partial differential equations (PDEs) that we derive from the frequency domain Maxwell equations. We test the ability of the pro- posed framework to characterize deformation effects in transmission lines by comparing simulation results with analytical predictions and corresponding experi- mental data. 2. Transmission-line theory Before focusing on the development of our numerical framework for the study of deformed transmission lines, we summarize some concepts from transmission- line theory. In most transmission-line applications, the transversal electromagnetic (TEM) mode can be regarded as the dominant signal propagation mode [6]. In this mode, electric and magnetic fields lie in the transverse (x, y) plane orthogonal to the direction of propagation z of the incident electromagnetic wave. Based on the TEM assumption, it is possible to derive the telegrapher's equations from Maxwell's theory and describe the TEM wave propagation along a transmission line [7, 6]. In the frequency domain, these equations describe the transmission-line voltage V (z) and current I(z) at position z and are given by dV (z) dz dI (z) dz = − (R + iωL) I (z) , = − (G + iωC) V (z) , (1a) (1b) where ω = 2πf and f is the corresponding frequency. The transmission-line parameters are R = R(z, ω), L = L(z, ω), C = C(z, ω), and G = G(z, ω) and denote resistance, inductance, capacitance, and conductance per unit length, respectively. Some important factors that influence these parameters are transmission-line Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 3 Figure 1. Transmission line model. The transmission line model is based on equivalent circuit representations of individual line segments where R, L, C, and G are resistance, inductance, capacitance, and conductance per unit length, respectively. The direction of propagation of the incident electromagnetic wave is z. geometry, conductivity and dielectric properties of the used materials, and frequency of the incident wave. The telegrapher's equations may also be inter- preted in terms of a sequence of equivalent circuits as we show in figure 1. Each segment is specified by the corresponding local transmission-line parame- ters and when put together they describe the whole transmission line. Instead of solving Maxwell's equa- tions for a given transmission line and incident wave in three dimensions, the telegrapher's equations pro- vide a way to describe the same problem by solv- ing eqs. (1a) and (1b). This approach permits a sub- stantial reduction of computational effort. However, it also relies on the assumption of a TEM mode wave propagation that is only valid for uniform transmis- sion lines consisting of perfect conductors surrounded by a homogeneous medium [6]. Still, even in the cases of lossy conductors, inhomogeneous surrounding me- dia, and nonuniform cross-sections, the telegrapher's equations are used under the assumption that such ef- fects only lead to negligible deviations from the TEM mode description (quasi-TEM assumption) [6]. We therefore simulate deformed transmission lines accord- ing to eqs. (1a) and (1b). In a transmission line, electromagnetic waves propagate with a phase velocity of v = 1/ LC. If the surrounding medium is homogeneous with permeability µ and permittivity = (cid:48) − i (cid:48)(cid:48) = √ 0 ((cid:48)r − i (cid:48)(cid:48)r ), the phase velocity is v = 1/ [6]. The imaginary part of the permittivity accounts for dielectric losses and 0 is the vacuum permittivity. In most transmission line applications, the permeability µ = µ0µr is equal to the vacuum permeability µ0, because common conductors (e.g., copper and aluminum) exhibit a relative permeability of µr ≈ 1. For each transmission-line segment, the character- µ(cid:48) √ istic impedance is [6] Z0 (ω) = (cid:114) R + iωL G + iωC (2) and therefore a function of the transmission-line parameters. At an interface between a line segment with impedance Z1 and another one with impedance Z2, the corresponding reflection coefficient [6] Γ12 = Z2 − Z1 Z2 + Z1 (3) describes the ratio of the reflected and the incident wave. No reflections occur for transmission lines with equal impedances along the line (matched transmission lines). To describe the influence of transmission-line deformations on characteristic impedance and signal propagation, we have to determine the corresponding transmission-line parameters. Only in certain cases such as for coaxial geometries, is it possible to find closed analytical solutions. For general deformations, it is necessary to utilize numerical methods. Under the quasi-TEM assumption, we first determine the electric and magnetic fields E (z, ω) and B (z, ω) for a transmission-line segment at position z. Based on the obtained electromagnetic fields, we then compute the transmission-line parameters for the considered segment. Specifically, let Ω ⊆ R2 be the cross section of a certain transmission-line segment at position z. The cross section Ω shall include the surrounding medium up to a certain distance where the magnetic field is sufficiently small. We denote the current density in z direction by Jz (z, ω) and describe resistive losses by [8] P (z, ω) = dA , (4) where σ denotes the electric conductivity. Based on eq. (4), the resistance per unit length is R (z, ω) = P (z, ω) I 2 , (5) is the current where I through the transmission line. Next, we compute the inductance with the help of the total magnetic energy averaged over one cycle [8] that flows ¯Wm (z, ω) = 1 4 H (z, ω) · B∗ (z, ω) dA , (6) (cid:90) Ω (cid:90) Jz (z, ω)2 Ω σ Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 4 where H (z, ω) = B (z, ω) /µ. The inductance per unit length is [8] general case of a non-homogeneous dielectric medium, the conductance per unit length is [6] L (z, ω) = 4 ¯Wm (z, ω) I 2 . (7) G (z, ω) = It (z, ω) V , (14) (cid:90) Ω Similarly, we determine the capacitance by considering the total electric energy averaged over one cycle [8] ¯We (z, ω) = 1 4 E (z, ω) · D∗ (z, ω) dA , (8) where D (z, ω) = E (z, ω) /(cid:48). We obtain the capacitance per unit length through [8] C (z, ω) = 4 ¯We (z, ω) V 2 , (9) where V is the voltage difference between inner and outer conductor. We note that electric field and capacitance are frequency-independent until currents that flow through the dielectric material become relevant. This is only the case for frequencies that are not captured by the quasi-TEM approximation. Another possibility to compute the capacitance is to consider the total charge Q. Therefore, let Ω(cid:48) ⊆ R2 be the area that covers all conductors of the same voltage level. The total charge is Q (z, ω) = (cid:48) = 0 ρ dA (cid:48)rE · n(cid:48) dl , Ω(cid:48) ∂Ω(cid:48) (10) where ρ is the charge density. The relative permittivity (cid:48)r may be position-dependent. The second step law, where n(cid:48) in eq. (10) follows from Gauss' is perpendicular to the boundary ∂Ω(cid:48) of Ω(cid:48). The capacity per unit length is (cid:90) (cid:90) For a homogeneous dielectric medium, the conductance per unit length is [6] G (z, ω) = ω tan (δ) C (z, ω) , (12) where tan (δ) = ω(cid:48)(cid:48) − σdielectric ω(cid:48) (13) denotes the loss tangent. In practice, dielectric losses dominate and tan (δ) = (cid:48)(cid:48)/(cid:48). To fully describe it is necessary to determine the dielectric material, the frequency dependence of tan (δ). For typical dielectric materials that are used in transmission line applications, the frequency dependence is almost constant over a certain frequency range [6]. In the more (cid:90) where It is the total transverse conduction current per unit line length between the conductors. Integrating the current density J = (σ − ω(cid:48)(cid:48)) E + Js over ∂Ω(cid:48) yields the total transverse conduction current and the conductance per unit length G (z, ω) = 1 V ∂Ω(cid:48) J · n(cid:48) dl , (15) where Js in J is the externally applied source current. 3. Potential formulation of Maxwell's equations After having outlined the basic strategy of how to describe transmission lines of arbitrary cross sections in section 2, we now determine the electromagnetic fields of a given transmission-line segment [9, 10]. Based on eqs. (5), (7), (9) and (15), it is then possible to obtain the corresponding transmission-line parameters. We begin with a brief summary of the potential formulation of Maxwell's equations as proposed in [9, 10]. Maxwell's equations in the frequency domain are ∇ × E − iωµH = 0 , ∇ × H − (σ − iω) E = Js , ∇ · (E) = ρ , ∇ · (µH) = 0 . (16) (17) (18) (19) We divide eq. (16) by µ, take the curl, and substitute eq. (17) into the resulting expression to obtain ∇ ×(cid:0)µ−1∇ × E(cid:1) − iωσE = iωJs , (20) (21) (23) (24) is the generalized conductivity. Moreover, a decompo- sition of E into the vector potential A and the scalar potential Φ yields E = A + ∇Φ . (22) The vector potential satisfies the Coulomb gauge condition ∇ · A = 0 . Substituting eq. (22) into eq. (20) leads to ∇ × ∇ × A − iωµσ (A + ∇Φ) = iωµJs , where we assumed a constant permeability µ. Using the Coulomb gauge condition of eq. (23), simplifies eq. (24) to ∇2A + iωµσ (A + ∇Φ) = −iωµJs (25) C (z, ω) = Q (z, ω) V . where (11) σ := σ − iω Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 5 and taking the divergence of eq. (25) yields ∇ · [σ (A + ∇Φ)] = −∇ · Js . (26) x , (27a) y , (27b) z , (27c) (27d) As described in section 2, under the quasi-TEM as- sumption we consider the transmission line to be com- posed of individual segments whose electromagnetic fields lie in the transverse (x, y) plane. Therefore, the solution of eqs. (25) and (26) can be obtained with a substantial reduction in computational effort. Under the quasi-TEM assumption, the resulting PDEs are (cid:0)∂2 (cid:0)∂2 x + ∂2 y x + ∂2 y (cid:1) Ax + iωµσ (Ax + ∂xΦ) = −iωµJ s (cid:1) Ay + iωµσ (Ay + ∂yΦ) = −iωµJ s (cid:1) Az + iωµσAz = −iωµJ s (cid:0)∂2 x + ∂2 y ∂x[σ (Ax + ∂xΦ)] + ∂y[σ (Ay + ∂yΦ)] = −∂xJ s x − ∂yJ s y . We note that Φ dropped out of eq. (27c). As a result Az is decoupled from Ax, Ay, and Φ. To determine capacitance and conductance for different frequencies, we have to solve eqs. (27a) to (27d) only once since the capacitance as defined in eq. (9) exhibits no frequency dependence and the conductance scales linearly with ω according to eq. (15) [6]. In section 4.2, we outline that it is sufficient to solve eq. (27c) to determine the frequency dependence of the magnetic field, and thus of the resistance and inductance of a certain transmission-line segment. This again leads to a substantial reduction of computational effort compared to the case in which we have to solve the whole set of PDEs given by eqs. (27a) to (27d). According to [9], possible boundary conditions for the solution of eq. (25) are (∇ × A) × n∂Ω = 0 , A · n∂Ω = 0 , ∇Φ∂Ω · n = 0 , (cid:90) ΦdA = 0 , Ω (28a) (28b) (28c) (28d) where n is the unit normal vector on the boundary ∂Ω. In the case of transmission lines, the inner and outer conductors can be assumed to exhibit potentials of the same absolute value relative to the respective area, but with opposite signs. Therefore, the boundary conditions given by eqs. (28c) and (28d) are already satisfied. 4. Discretization To solve eqs. (27a) to (27d) for arbitrary transmission- line cross sections, we employ a staggered finite- difference scheme on non-uniform grids [11, 10, 9]. Figure 2. Electromagnetic field discretization. Position of the discretized electromagnetic field components according to [11]. (a) Electromagnetic field com- ponents in the plane. (b) Potential components in the plane. Figure 3. Reduction to a two-dimensional problem. Under the quasi-TEM assumption, it is possible to consider a two-dimensional version of the Yee grid. Specifically, we use a method that is based on central differences at midpoints. That is, we compute derivatives between two adjacent grid points using a forward-difference scheme and consider the resulting derivative values to be located at the corresponding midpoints. In this way, boundaries between different conductors and dielectric materials can be resolved well. We discretize the electromagnetic fields according to the Yee method (see figure 2) [11]. Here and in the subsequent sections, we use F i,j as a shorthand x notation for the the x-component Fx (xi, yj) of a vector field F at position (xi, yj), and similarly for the y and z components of F. Instead of solving the three- dimensional problem (see figure 2), we consider a two-dimensional representation of the Yee grid (see figure 3). This reduction to two dimensions is a consequence of the quasi-TEM assumption that we described in section 3. We show the discretization of the electromagnetic potentials and corresponding xyzEi+12,jxBi+12,jyEi,j+12yBi,j+12xBi+12,j+12zEi,jzEi,j+1zxyzAi+12,jxJi+12,jxAi,j+12yJi,j+12yAi,jzJi,jzΦi,jσi,jModeling Deformed Transmission Lines for Continuous Strain Sensing Applications 6 (cid:0)∂2 xf(cid:1)i,j of a function f . We eliminate ∂xf i,j to obtain hi−1 x x + hi x 2f i−1,j (cid:0)hi−1 (cid:0)hi−1 2f i+1,j x + hi x (cid:1) − 2f i,j (cid:1) + O(cid:0)hi−1 hi−1 x hi x x − hi x hi x = + (cid:1) (35) and similarly for a vector potential component (cid:0)∂2 (cid:1)i,j+ 1 xAy 2 = hi−1 x 2 i−1,j+ 1 y x + hi x 2A (cid:0)hi−1 (cid:0)hi−1 i+1,j+ 1 2 y x + hi x 2A + hi x Since the vector potentials are located between grid points, we replace hi,j and, for the second derivative of the vector potential, i+ 1 x by h x x + hi+1,j 2 ,j x (cid:1) . (cid:1) /2 (36) i,j+ 1 2 y hi−1 x hi x x − hi x (cid:1) − 2A (cid:1) + O(cid:0)hi−1 = (cid:0)hi,j (cid:17) − 2A (cid:17) . 2 ,j i+ 1 x i− 1 i+ 1 h x h 2 x 2 (37) obtain(cid:0)∂2 xAx (cid:1)i+ 1 2 ,j (cid:16) (cid:16) 2A 2 ,j i− 1 x i− 1 i+ 1 h x + h 2 x 2 2A 2 ,j i+ 3 x i− 1 x + h h 2 i+ 1 2 x = + i− 1 h x 2 i+ 1 h 2 x parameters in figure 3b. 4.1. Non-uniform grid One possibility to numerically solve eqs. (27a) to (27d), is to consider equal grid spacings in a two-dimensional Yee grid (see figure 3). For perfectly coaxial geometries, such an approach is suitable to resolve the electromagnetic fields since they vanish outside the transmission line. However, for arbitrary geometries, it is important to resolve far field contributions of the magnetic field to correctly compute the inductance according to eqs. (6) and (7). We therefore employ a non-uniform Yee discretization of the computational domain Ω ⊆ R2 (see figure 4) and denote the difference x (hi between grid points xi (yi) and xi+1 (yi+1) by hi y). We first focus on the non-uniform Yee discretiza- tion of all derivatives occurring in eqs. (27a) to (27d) and then present the fully discretized version of the considered PDEs. We only describe the discretization along the x-axis bearing in mind that the same steps also apply in y-direction. For the discretization of ∂xΦ, we consider central differences at midpoints and thus compute the derivative between xi and xi+1 at position xi+ 1 2 (see figure 4). We obtain (∂xΦ)i+ 1 2 ,j = Φi+1,j − Φi,j hi x . (29) In the next step, we discretize σAx and consider σi,j to be located at the same position as Φi,j and J i,j z (see figure 3b). We have to determine σi+ 1 2 ,j since the 2 . To do vector field component A so, we compute the harmonic mean is located at xi+ 1 i+ 1 2 x (cid:18) 1 (cid:18) 1 σi+ 1 2 ,j = 2 1 σi,j + σi+1,j and obtain σ1+ 1 2 ,jA 2 ,j i+ 1 x = 2 and σi,j + 1 σi+1,j (cid:19)−1 (cid:19)−1 A (30) 2 ,j i+ 1 x (31) Furthermore, we compute the term ∂x(σ∂xΦ) in eq. (27d) by considering central differences at mid- points: ∂x(σ∂xΦ) = σi− 1 − σi+ 1 2 ,j(cid:0)Φi−1,j − Φi,j(cid:1) 2 ,j(cid:0)Φi,j − Φi+1,j(cid:1) i− 1 x hi−1 h x 2 i− 1 x hi x 2 h (38) . Finally, we can formulate eqs. (27a) to (27d) in terms of a staggered finite-difference Yee discretization on a non-uniform grid: (cid:16) (cid:16) 2A 2 ,j i+ 3 x i− 1 i+ 1 h x + h 2 x 2 2 ,j i+ 1 x i− 1 i+ 1 h x h 2 x 2 (cid:17) − 2A (cid:17) + (cid:16) 2A hj−1 y (cid:17) (cid:16) 2A 2 ,j+1 i+ 1 x hj−1 y + hj y + i+ 1 h 2 x − 2A 2 ,j i+ 1 x hj−1 y hj y + hj y (cid:18) + iωµσi+ 1 2 ,j 2 ,j i+ 1 x A + Φi,j + Φi+1,j hi x (cid:17) i+ 1 2 ,j−1 x hj−1 y + hj y (cid:19) = −iωµJ 2 ,j i+ 1 s,x , (39) [∂x (σAx)]i,j = σi+ 1 2 ,jA 2 ,j i+ 1 x − σi− 1 i− 1 h x 2 2 ,jA 2 ,j i− 1 x . (32) i− 1 h x 2 2A 2 ,j i− 1 x i− 1 i+ 1 h x + h 2 x 2 For computing the second derivatives of the electro- magnetic potentials, we consider the Taylor expansions f i−1,j = f i,j−hi−1 x ∂xf i,j+ 1 2! x (cid:0)hi−1 (cid:1)2 (cid:0)hi x ∂2 (cid:1)3(cid:17) xf i,j−O(cid:16)(cid:0)hi−1 (cid:1)2 (cid:1)3(cid:17) xf i,j + O(cid:16)(cid:0)hi x (33) ∂2 and f i+1,j = f i,j + hi x∂xf i,j + 1 2! x (34) Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 7 Figure 4. Non-uniform grid. An illustration of a non-uniform discretization along the x-axis. where J(cid:48) = −iωµ (Js,x, Js,y, Js,z,∇Js/(iωµ)), A(cid:48) = (Ax, Ay, Az, Φ), and ∆ + iωµσ 0 0 ∂x σ D = 0 ∆ + iωµσ 0 0 0 ∂y σ ∆ + iωµσ 0 iωµσ∂x iωµσ∂y 0 ∇σ∇  . We solve eq. (43) with a sparse linear system solver and consider Js,z to be the only non-vanishing source current component. 4.2. Computing electromagnetic fields After having determined the electromagnetic poten- tials by solving eq. (43), we now compute the corre- sponding electromagnetic fields. The electric field E as defined by eq. (22) is  . hi x Φi,j+1−Φi,j  Φi+1,j−Φi,j  hj y 0 1 iω ∂yEz−∂xEz ∂xEy − ∂yEx (44)  (45) (46)  . (cid:17) 2 i,j− 1 2A y j− 1 j+ 1 h y + h 2 y 2 2 (cid:0)hi−1 (cid:0)hi−1 i−1,j+ 1 y x + hi x i+1,j+ 1 2 y x + hi x 2A (cid:1) − 2A (cid:1) + 2A hi−1 x + hi x i,j+ 1 2 y hi−1 x hi x (cid:16) i,j+ 1 − 2A 2 y j− 1 h y j+ 1 h 2 y 2 + (cid:32) + iωµσi,j+ 1 2 A j− 1 h y 2 (cid:16) i,j+ 3 2A 2 y j+ 1 j− 1 h y + h 2 y 2 j+ 1 h 2 y (cid:17) i,j+ 1 2 y + Φi,j + Φi,j+1 hj y (cid:33) = −iωµJ i,j+ 1 2 s,y , (40) + 2Ai−1,j (cid:0)hi−1 (cid:16) z hi−1 x hi x (cid:1) − 2Ai,j z (cid:17) − 2Ai,j x + hi x 2Ai,j−1 z hj−1 y + hj y z = −iωµJ i,j s,z , z hj−1 y hj y hi−1 x + hj−1 y + iωµσi,jAi,j (cid:1) hi x x + hi x 2Ai+1,j z (cid:0)hi−1 (cid:16) hj y + 2Ai,j+1 z hj−1 y + hj y (cid:17) and σi+ 1 2 ,jA 2 2 ,j i+ 1 x 2 ,jA − σi− 1 i− 1 h x 2 ,j(cid:0)Φi+1,j − Φi,j(cid:1) 2(cid:0)Φi,j+1 − Φi,j(cid:1) i− 1 x hi h x 2 σi+ 1 σi,j+ 1 2 ,j i− 1 x + i,j+ 1 2 y σi,j+ 1 2 A 2 ,j(cid:0)Φi,j − Φi−1,j(cid:1) 2(cid:0)Φi,j − Φi,j−1(cid:1) i− 1 x hi−1 h x 2 − σi− 1 − σi,j− 1 + + 2 hj y i+ 1 2 ,j s,x j− 1 h y s,x − J i− 1 i− 1 h x 2 ,j 2 J = + 2 s,y − J i,j− 1 J j− 1 h y 2 2 j− 1 h y i,j+ 1 2 s,y . Our approach extends [9] by considering a non-uniform discretization of the potential formulation of Maxwell's equations as defined by eqs. (27a) to (27d). (42) To numerically solve eqs. (39) to (42), we rewrite the discretized PDEs in terms of a linear system of equations DA(cid:48) = J(cid:48) , (43) (41) Ei,j = Ai,j + Based on eq. (16), the magnetic field is − σi,j− 1 2 A j− 1 h y 2 i,j− 1 y 2 B = µH = 1 iω ∇ × E = and thus hj−1 y Bi,j = 1 iω  z z −Ei,j Ei,j+1 hj y − Ei+1,j −Ei,j hi x − Ei,j+1 Ei+1,j −Ei,j −Ei,j hj hi x y x x y y z z We only have to compute the electric field for one frequency, because the capacitance is frequency- independent and the conductance scales linearly with ω according to eq. (15). For all other frequencies, it is sufficient to numerically solve (∆ + iωµσ) Az = −iωµJs,z (47) to compute resistance and inductance as defined in eqs. (5) and (7). The vector potential component Az xi−1xixi+1xi+2xi−12xi+12xi+32hi−1xhixhi+1xhi−12x=hi−1x+hix2hi+12x=hix+hi+1x2Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 8 fully determines the magnetic of a transmission-line segment field since 1 Bx = iω By = − 1 iω ∂yEz = ∂yEz = − 1 iω 1 iω ∂yAz , ∂xAz . (48a) (48b) We note that the outlined numerical framework can be used to simulate the electromagnetic fields (see eqs. (44) and (46)) and transmission-line parameters (9) and (15)) of arbitrary (see arrangements of conductor, insulator, and shielding materials. eqs. (7), (5), 5. Results Based on the numerical framework that we introduced in sections 3 and 4, we now examine the effect of deformations on transmission-line parameters. We first study the convergence characteristics of the proposed framework by considering a single copper strand and comparing our numerical results with the corresponding analytical theory. After this initial test, we simulate the frequency-dependence of transmission- line parameters for undeformed coaxial geometries and draw a comparison to the deformed case. Finally, we compare our simulations of deformed transmission lines with corresponding experimental data. 5.1. Single copper strand To test the proposed numerical framework, we consider a single copper strand of radius r1 and compare the numerically obtained values of the magnetic field with the ones predicted by analytic theory. For low frequencies, the current density within the wire is uniform and using Amp`ere's law yields r ≤ r1 , r > r1 , (cid:40) µIr B(r) = (49) , 2πr2 1 µI 2πr , where r is the distance to the center of the strand, I is the current flowing through the strand, and µ = µ0. For high frequencies, the current density is not uniformly distributed in the wire, but exhibits a higher density on its surface (skin effect). In this case, the electric field is [15] and by applying Amp`ere's law to eq. (51) we obtain (cid:40) µI1(kr) 2πr1I1(kr1) I , µI 2πr , B(r) = r ≤ r1 , r > r1 . (52) (cid:88) We now compare our simulation results with the analytical predictions of eq. (52) for a single copper strand of radius r1 = 0.48 mm that is surrounded by air. We show the results in figures 5a and 5c, and find good agreement between our simulations and analytical theory. For a frequency of f = 1 MHz, we see the influence of the skin effect on the magnetic field. To study the convergence characteristics of our method for different numbers of grid points along one axis N , we define the error 1 M (i,j)∈S Bi,j ∆ (N ) = simulation (N ) − Bi,j analytical (N ) , (cid:113) (cid:9) and ri,j = where S =(cid:8)(i, j)ri,j ≤ rc (53) (xi)2 + (yj)2. The cut-off radius is denoted by rc, and M is the number of considered points. We use the convention that the value of ri,j = 0 corresponds to the center of the cooper strand. In our subsequent analysis, we set rc = 2 mm. The analytical magnetic field values Bi,j analytical correspond to the ones of eq. (52). According to eqs. (36) and (46), the global error is expected to be of order O (h), where h is the distance between the nodes of the grid. We find that the numerically determined error of our method agrees well with the expected scaling (see figures 5b and 5d). 5.2. Undeformed coaxial transmission line To examine the effect of deformations on transmission- line parameters, we first simulate an undeformed coaxial cable (see figure 6a). In this way, we can later compare our results on deformed transmission lines with those obtained for an undeformed reference. We denote the radius of the inner conductor by r1 and r2 is the inner radius of the outer conductor of thickness t. Therefore, the outer radius of the outer conductor is r3 = r2 + t. Similarly to the magnetic field of a single copper strand (see eq. (52)), we find for an undeformed coaxial transmission line  µI1(kr) 2πr1I1(kr1) I , µI 2πr , µI1(k(r3−r)) 2πr2I1(k(r3−r2)) I , 0 , r ≤ r1 , r1 ≤ r ≤ r2 , r2 ≤ r ≤ r3 , r > r3 . (54) Ez (r) = kI0 (kr) 2πσr1I1 (kr1) I , (50) B(r) = √ where k = iωµσ, and I0 and I1 are the modified Bessel functions of first order and first and second kind, respectively. The current density is thus Jz (r) = kI0 (kr) 2πr1I1 (kr1) I (51) The expressions for r ≤ r2 and r ≥ r3 are exact. The magnetic field vanishes for r ≥ r3, because the currents in the inner and outer conductors are oriented in the opposite directions. For the magnetic field in Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 9 (a) f = 1 Hz (b) f = 1 Hz (c) f = 1 MHz (d) f = 1 MHz Figure 5. Magnetic field of a single copper strand at low and high frequencies. In the left panels, we show the absolute value of the magnetic field of a single copper strand surrounded by air for frequencies f = 1 Hz and f = 1 MHz. The radius of the strand is r1 = 0.48 mm. The current is set to I = 1 A and the number of grid points along one axis is N = 256. The convergence characteristics of our method is shown in the corresponding right panels. The parameters were set to (cid:48)r,copper = 1 [12], (cid:48)(cid:48)r,copper = 0, σcopper = 5.98 × 107 S m [13], (cid:48)r,air = 1.00059 [14], σair = 0. the outer conductor, we use an approximation that assumes that the current density distribution is the same as for a wire of radius t. In figures 7a and 7c, we see that the simulated magnetic field values agree well with the predictions of eq. (54). For large enough frequencies, we recognize the influence of the skin effect (see figure 7c). As for the single copper strand, we examine the convergence properties of our method for the coaxial geometry (see figures 7b and 7d). Due to the skin effect, a larger number of grid cells is required for higher frequencies than for lower ones to resolve the electromagnetic fields. Next, we determine the transmission-line pa- rameters of the undeformed coaxial cable according to eqs. (5), (7), (9) and (15), and compare the obtained values with the analytical low and high frequency ap- proximations. The analytical expressions for capaci- tance and conductance per unit length are [6] (cid:16) r2 2π0(cid:48)r ln C = (cid:17) , r1 G = ω tan (δ) C . (55) (56) We set (cid:48)r = 2.25 and tan (δ) = 10−3. According to eqs. (55) and (56), the capacitance exhibits no frequency dependence and the conductance scales linearly with ω. This behavior is captured by our simulations (see figure 8). Instead of setting tan (δ) equal to a constant, it is also possible to use empirically determined frequency dependencies of tan (δ). The current densities in the conductors are frequency-dependent. For low frequencies, the current density is uniformly distributed within the conductor (DC approximation), and for high frequencies the current density is concentrated at the conductor surface (HF approximation). The skin effect is approximated by describing the conductors as hollow 21012Distance to core [mm]0.00.10.20.30.4B(r) [mT]SimulationAnalytic26272829210N105104103102(N)[mT]-121012Distance to core [mm]0.00.10.20.30.4B(r) [mT]SimulationAnalytic26272829210N104103102(N)[mT]-1Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 10 (a) Undeformed coaxial cable (b) Deformed coaxial cable Figure 6. Cross sections of an undeformed and a deformed coaxial cable. The left panel shows an undeformed coaxial cable with an inner conductor of radius r1. The outer conductor begins at radius r2 away from the center and has thickness t. The outer radius of the outer conductor is thus r3 = r2 + t. An example of a deformed coaxial cable is shown in the right panel. The radius of the upper and lower half circles is r4. The center of these half circles is located at distance h away from the center of the inner conductor. cylinders with their thickness being determined by the corresponding skin depth. In the low frequency regime, the resistance and conductance per unit length are [16] the resistance according to the scaling of R (ω) ∼ √ ω for ω → ∞. If one is only interested in the it is also possible to neglect resistance- impedance, related deviations since (cid:114) Z0 (ω) = for ω → ∞ . L C (61) 5.3. Deformed coaxial transmission line To study the influence of deformations on transmission- line parameters, we now compress coaxial transmission line between two rigid parallel plates (see figure 6a). The circumference is invariant under the applied deformation, so the 2πr2 = 2πr4 + 4h , (62) where r2 is the inner radius of the outer conductor of the undeformed cable, r4 is the radius of the upper and lower half circles of the deformed cable, and h is the distance of the centers of the half circles from the center of the inner conductor (see figure 6b). We consider a deformation with r4 = 0.8 mm and h = 1.02 mm, and compare the transmission- line parameters of this deformed coaxial cross-section with those of the undeformed reference. We show the corresponding results in figure 9. We find that resistance and conductance are almost unaffected by the deformation. However, capacitance and inductance differ significantly when compared to the undeformed reference. In the high frequency regime, the impedance (cid:19) , + (cid:18) 1 (cid:20) (cid:18) r2 (cid:19) 1 3 − r2 r2 2 (cid:18) 1 + 4 r2 1 ln r1 1 3 − r2 2) (cid:114) µω 2σ ln µ 2π (cid:18) 1 (cid:19) (cid:18) r2 1 2π r1 + r1 1 σπ µ 2π R = L = + 4 (r2 R = L = and for high frequencies (cid:19)(cid:19)(cid:21) (cid:18) r3 r2 ln 2 − 3r2 r2 3 + 4r2 3 3 − r2 r2 2 (57) (58) (59) (cid:19) + 1 r2 µ 4π , 2 µωσ (cid:19) (cid:18) 1 r1 + 1 r2 . (60) We show the analytical approximations of R and L and the corresponding simulation results in figure 8. We find that the simulated transmission-line parameters are in good agreement with the analytic DC and HF approximations. In addition, our numerical results also describe the analytically inaccessible transition region between the DC and HF regime. For large frequencies, deviations of the simulated resistance values from analytic theory occur. The reason is that very small grid spacings are necessary to resolve the current density and compute the resistance per unit length according to eq. (4). Another more computationally efficient possibility is to extrapolate r1r2tr1hr4tModeling Deformed Transmission Lines for Continuous Strain Sensing Applications 11 (a) f = 1 Hz (b) f = 1 Hz (c) f = 1 MHz (d) f = 1 MHz Figure 7. Magnetic field of a coaxial cable at low and high frequencies. In the left panels, we show the absolute value of the magnetic field of a coaxial cable with copper conductors surrounded by air for frequencies f = 1 Hz and f = 1 MHz. The current is set to I = 1 A and the current in the inner conductor flows in the opposite direction as the one in the outer conductor. The number of grid points along one axis is N = 1024. The convergence characteristics of our method is shown in the corresponding right panels. The remaining parameters are r1 = 0.48 mm, r2 = 1.45 mm, r3 = 1.6 mm, (cid:48)r,dielectric = 2.25, σdielectric = 0, tan (δ) = 10−3, (cid:48)r,copper = 1 [12], σcopper = 5.98 × 107 S m [13], (cid:48)r,air = 1.00059 [14], σair = 0. of the undeformed coaxial cable is about 50 % larger than the impedance of the deformed one. According to eq. (3), this corresponds to a reflection coefficient of Γ = −0.2 at the interface between the undeformed and deformed segments. We outline in section 5.4, that such reflections are detectable with a time-domain reflectometer (TDR). In addition to variations in the cross section, it is also possible to incorporate spatial permittivity variations that result from applied mechanical strains. 5.4. Comparison with experimental data show that simulations Our the impedance of a deformed coaxial transmission line is significantly smaller as compared to the undeformed case (see figure 9). To test our numerically obtained transmission-line parameters and impedance values, we experimentally analyze the effect of deformations on the TDR profile of a deformation-sensitive coaxial cable. TDR measurements are based on the injection of a test signal (e.g., a step function) into a transmission line. Reflections of the injected signal occur at impedance discontinuities. Each reflection in the transmission line leads to a wave which travels back to the TDR where the resulting voltage differences and corresponding time stamps are monitored. These monitored quantities contain information about the spatial distribution of reflection coefficients and impedances along the transmission line. For our experiments, we use a Campbell Scientific TDR100. The considered coaxial transmission line has a length of 9.70(1) m and its dielectric material is a thermoplastic elastomer with a relative permittivity of (cid:48)r,dielectric = 2.14. The number in parentheses is the uncertainty in the last digit. With a vise of length 6.0(1) cm, we apply a deformation 6.98(1) m away from the TDR. We denote the distance between 21012Distance to core [mm]0.00.10.20.30.4B(r) [mT]SimulationAnalytic26272829210N105104103102(N)[mT]-121012Distance to core [mm]0.00.10.20.30.4B(r) [mT]SimulationAnalytic26272829210N104103102(N)[mT]-1Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 12 Figure 8. Transmission-line parameters of an undeformed coaxial cable. We show the transmission-line parameters and the characteristic impedance (simulation and analytic approximations) as defined by eqs. (2), (5), (7), (9) and (15) for an undeformed coaxial cable. The number of grid points along one axis is N = 512 and the remaining parameters are r1 = 0.48 mm, r2 = 1.45 mm, r3 = 1.6 mm, (cid:48)r,dielectric = 2.25, σdielectric = 0, tan (δ) = 10−3, (cid:48)r,copper = 1 [12], σcopper = 5.98 × 107 S m [13], (cid:48)r,air = 1.00059 [14], σair = 0. the two parallel plates of the vise by d. We show an illustration of the deformed transmission lines that we considered in our simulations and experiments the considered transmission line, we did not observe any traces of plastic-deformation effects in the TDR curves. However, depending on the used dielectric and cable jacket materials, plastic deformations may in figures 10a and 10b. For occur and affect measurements. Under the quasi- TEM assumption, the transmission line is composed of undeformed and deformed segments. We have to compute the frequency dependence of the transmission- line parameters for both cross sections only once to simulate the corresponding TDR profile according to the telegraphers equations as defined in eqs. (1a) and (1b). To solve eqs. (1a) and (1b), we employ 100102104106108Frequency [Hz]102101100Resistance [m]SimulationDC approximationHF approximation100102104106108Frequency [Hz]0.200.220.240.260.280.30Inductance [Hm]SimulationDC approximationHF approximation100102104106108Frequency [Hz]100105110115120Capacitance [pFm]SimulationAnalytic100102104106108Frequency [Hz]10131011109107105103Conductance [Sm]SimulationAnalytic100102104106108Frequency [Hz]101102103104Impedance []SimulationDC approximationHF approximation104105106107108Frequency [Hz]40455055606570Impedance []SimulationDC approximationHF approximationModeling Deformed Transmission Lines for Continuous Strain Sensing Applications 13 Figure 9. Transmission-line parameters of a deformed coaxial cable. We show the transmission-line parameters and the characteristic impedance as defined by eqs. (2), (5), (7), (9) and (15) for a deformed coaxial cable. The analytic approximation describes the undeformed case. The number of grid points along one axis is N = 512 and the remaining parameters are r1 = 0.48 mm, r4 = 0.8 mm, h = 1.02 mm, t = 0.15 mm, (cid:48)r,dielectric = 2.25, σdielectric = 0, tan (δ) = 10−3, (cid:48)r,copper = 1 [12], σcopper = 5.98 × 107 S m [13], (cid:48)r,air = 1.00059 [14], σair = 0. 100102104106108Frequency [Hz]102101100Resistance [m]SimulationDC approximation (undef.)HF approximation (undef.)100102104106108Frequency [Hz]0.100.150.200.250.30Inductance [Hm]SimulationDC approximation (undef.)HF approximation (undef.)100102104106108Frequency [Hz]100120140160180Capacitance [pFm]SimulationAnalytic (undef.)100102104106108Frequency [Hz]10131011109107105103Conductance [Sm]SimulationAnalytic (undef.)100102104106108Frequency [Hz]101102103104Impedance []SimulationDC approximation (undef.)HF approximation (undef.)104105106107108Frequency [Hz]203040506070Impedance []SimulationDC approximation (undef.)HF approximation (undef.)Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 14 (a) Deformed transmission line (simulation). (b) Deformed transmission line (microscope). (c) TDR measurement and simulation. (d) TDR measurement and simulation (zoomed). Figure 10. Comparison of simulations with experimental data. The upper panels show the deformed coaxial transmission- line cross sections (simulation and experiment). The inner conductor consists of seven copper wires of radius r1 = 0.16 mm, the shield is a braiding of tinned copper wires with a thickness of t = 0.15 mm, and the dielectric material is a thermoplastic elastomer with a relative permittivity of (cid:48)r,dielectric = 2.14 and (cid:48)(cid:48)r,dielectric = 4.28 × 10−3 and a radius of r2 = 1.45 mm. The cable jacket has a thickness of 0.40 mm. The simulation neglects that the shield is made of multiple tinned copper wires but approximates the cable by a non-braided copper shield of thickness t = 0.12 mm. In the lower panels, we show the corresponding TDR measurements for the described cable of length 9.70(1) m. We apply the shown deformation 6.98(1) m away from the TDR input with a vise of length 6 cm. We denote the distance between the two parallel plates of the vise by d. d [mm] 2.48(3) 2.22(3) 1.94(3) r4 [mm] 0.95(1) 0.85(1) 0.75(1) h [mm] 0.79(1) 0.93(1) 1.10(1) Table 1. For each measurement and simulation in figures 10c and 10d, we summarize the corresponding cable geometry parameters. Parameters r4 and h determine the simulated deformation according to figure 6b and d is the distance between the two parallel plates of the used vise. a method as described in [17]. We again note that we only consider geometric effects in our simulations, and that we do not account for strain-induced spatial variations of the permittivity within the dielectric material. To test the validity of our simulation approach, we compare the simulated TDR profiles with experimentally obtained ones. We summarize the parameters that we use in our simulations in table 1. In figures 10c and 10d, we show the comparison between our simulations and corresponding experiments. These results clearly show that our framework is able to reproduce the observed TDR profiles. Determining the dependence of the voltage minimum on the plate distance d in figure 10d requires to numerically solve eqs. (1a) and (1b) for a given set of transmission- line parameters/impedances. A closed-form analytical expression can only be obtained for the impedance of concentrically-deformed transmission lines (i.e., coaxial In the case of concentric deformations, we can obtain the characteristic impedance from eqs. (55), (56), (58) and (61): lines of different diameter) [1]. Z0(ω) = 1 2π for ω → ∞ . (63) (cid:114) µ0µr 0r ln (cid:18) r2 (cid:19) r1 To describe the effect of general deformations on Z, it is necessary to use a numerical framework such as the one we propose in this work. In particular, 0255075100125150175t [ns]0.00.20.40.60.81.0V(t) [V]Measurement d = 1.94 mmMeasurement d = 2.22 mmMeasurement d = 2.48 mmSimulation92949698100t [ns]0.420.440.460.480.500.52V(t) [V]Measurement d = 1.94 mmMeasurement d = 2.22 mmMeasurement d = 2.48 mmSimulationModeling Deformed Transmission Lines for Continuous Strain Sensing Applications 15 we find that the simulations are able to capture the functional dependence of voltage minimum and plate distance d in the deformed regions. The experimentally observed TDR profiles are therefore consistent with our numerical framework for the simulation of arbitrarily- shaped transmission-line cross sections. 6. Conclusion We have introduced a numerical framework to simu- late arbitrarily-shaped transmission lines over a wide frequency range. Our method is based on solving a sparse system of linear equations and thus is suit- able for fast computations. The numerically computed transmission-line parameters and electromagnetic field values agree well with corresponding analytical predic- tions. We considered the effect of geometric deforma- tions, but our framework can be extended to also ac- count for strain-induced permittivity variations. We tested the proposed framework by comparing simula- tion results with corresponding experimental data and found good agreement between the model predictions and the experimentally observed behavior. Our study may improve continuous sensing ap- plications for the measurement of deformations and mechanical strains based on strain-sensitive transmis- sion lines. Instead of mapping observed impedance values to concentric deformations and thus neglecting any further structural information as suggested in [1], our method also captures changes in the cross-section geometry. Furthermore, our framework is also useful to determine analytically inaccessible transmission-line parameters and impedances for complex arrangements of conductor, insulator, and shielding materials. A possible extension of our work is to combine our framework with an inversion algorithm and determine estimates of the most likely deformations and mechan- ical strains given a certain TDR observation. This can contribute to improvements in corrosion detection [18] and monitoring fracture propagation in concrete struc- tures [1, 19]. When embedding transmission-line sen- sors in concrete structures, it is important to examine strain transfer characteristics at the interface between cable jacket and surrounding material. Similar to fibre- optic sensors [20], the mechanical and physicochemi- cal properties of the materials at the interface between sensor line and surrounding material (e.g., cable jacket material) may substantially influence the sensor perfor- mance. Recent advances in strain-transfer theory [21] may provide a possibility to correctly interpret mea- surement data, even in the presence of strain-transfer loss at the interface. To enhance the long-term sensing performance, it is important to limit material degradation by using materials with physicochemical properties that are suitable for a given application [20]. For instance, photochemical and oxidative degradation (i.e., chemical aging) can be limited by lowering the influence of air and light on the sensor. In addition, one should also consider the effect of the change of physical properties over time (i.e., physical aging) that may occur in the used polymer blends [22]. If transmission-line sensors are tailored to a specific application, previous studies suggest that they can stay in service for decades [1]. Depending on the used dielectric material, it may also be important to account for the influence of temperature variations on the measurement results if the permittivity temperature coefficient is large enough. The dielectric material of the transmission line we consider has a temperature coefficient of about 10−4 K−1 [23], so the effect of temperature variations on the impedance is negligible. In addition to permittivity changes, temperature variations may also affect the coating material and strain transfer characteristics [24]. Acknowledgements We thank Dani Or and Daniel Breitenstein for the possibility to use their Campbell Scientific TDR100 time-domain reflectometer and Joshua LeClair for helpful comments and discussions. We also thank the LEONI AG for providing a sample of a pressure- sensitive coaxial transmission line. References [1] Lin M W, Thaduri J and Abatan A O 2005 Measurement Science and Technology 16 1495 -- 1505 [2] Zhu C, Zhuang Y, Chen Y and Huang J 2019 IEEE Transactions on Instrumentation and Measurement [3] Sirkis J 1998 Using Bragg grating sensor systems in con- struction materials and bridges: perspectives and chal- lenges FIBER OPTICS SENSORS FOR CONSTRUC- TION MATERIALS AND BRIDGES-PROCEEDINGS OF THE INTERNATIONAL WORKSHOP HELD MAY 1998 [4] Brown A W, Smith J P, Bao X, Demerchant M D and Bremner T 1999 Journal of intelligent material systems and structures 10 340 -- 349 [5] Murayama H, Kageyama K, Naruse H and Shimada A 2004 Journal of Intelligent Material Systems and Structures 15 17 -- 25 [6] Paul C R 2008 Analysis of multiconductor transmission lines (John Wiley & Sons) [7] Chicone C 2016 An Invitation to Applied Mathematics: Differential Equations, Modeling, and Computation (Academic Press) [8] Jackson J D 1999 Classical electrodynamics (AAPT) [9] Haber E, Ascher U M, Aruliah D A and Oldenburg D W 2000 Journal of Computational Physics 163 150 -- 171 [10] Haber E and Ascher U M 2001 SIAM Journal on Scientific Computing 22 1943 -- 1961 [11] Yee K S 1966 IEEE Transactions on Antennas and Propagation 14 302 -- 307 [12] M Lourtioz J, Benisty H and Berger V 2006 Photonic Crystals, Towards Nanoscale Photonic Devices vol 59 Modeling Deformed Transmission Lines for Continuous Strain Sensing Applications 16 [13] Johnson H and Graham M 2003 High-speed Signal Prop- agation: Advanced Black Magic Prentice Hall Modern Semicondu (Prentice Hall PTR) ISBN 9780130844088 [14] Hector L G and Schultz H L 1936 Physics 7 133 -- 136 [15] Schelkunoff S A 1934 Bell System Technical Journal 13 532 -- 579 [16] Hayt W H 2012 Engineering electromagnetics 8th ed (1221 Avenue of the Americas, New York, NY 10020: McGraw- Hill) [17] D'Aucelli G M, Giaquinto N, Piuzzi E, Cataldo A, De Benedetto E and Cannazza G 2017 1 -- 6 [18] Liu W, Hunsperger R G, Chajes M J, Folliard K J and Kunz E 2002 Journal of Materials in Civil Engineering 14 217 -- 223 [19] Chen G, Mu H, Drewniak J L and Pommerenke D J 2003 Continuous coaxial cable sensors for monitoring of rc structures with electrical time domain reflectometry Smart Structures and Materials 2003: Smart Systems and Nondestructive Evaluation for Civil Infrastructures vol 5057 (International Society for Optics and Photonics) pp 410 -- 422 [20] Santos J L and Farahi F 2014 Handbook of optical sensors (Crc Press) [21] Wang H, Xiang P and Jiang L 2018 Sensors and Actuators A: Physical [22] Hodge I M 1995 Science 267 1945 -- 1947 [23] Harrop P 1969 Journal of Materials Science 4 370 -- 374 [24] Wang H and Dai J G 2019 Composites Part B: Engineering 162 303 -- 313
1811.11498	3	1811	2019-07-03T19:27:17	Zero-field magnetometry based on nitrogen-vacancy ensembles in diamond	[ "physics.app-ph", "quant-ph" ]	Ensembles of nitrogen-vacancy (NV) centers in diamonds are widely utilized for magnetometry, magnetic-field imaging and magnetic-resonance detection. They have not been used for magnetometry at zero ambient field because Zeeman sublevels lose first-order sensitivity to magnetic fields as they are mixed due to crystal strain or electric fields. In this work, we realize a zero-field (ZF) magnetometer using polarization-selective microwave excitation in a 12C-enriched HPHT crystal sample. We employ circularly polarized microwaves to address specific transitions in the optically detected magnetic resonance and perform magnetometry with a noise floor of 250 pT/Hz^(1/2). This technique opens the door to practical applications of NV sensors for ZF magnetic sensing, such as ZF nuclear magnetic resonance, and investigation of magnetic fields in biological systems.	physics.app-ph	physics	Zero-field magnetometry based on nitrogen-vacancy ensembles in diamond Huijie Zheng,1, ∗ Jingyan Xu,2 Geoffrey Z. Iwata,1 Till Lenz,1 Julia Michl,3 Boris Yavkin,3 Kazuo Nakamura,4 Hitoshi Sumiya,5 Takeshi Ohshima,6 Junichi Isoya,7 Jorg Wrachtrup,3 Arne Wickenbrock,1 and Dmitry Budker1, 8, 9, 10 1Johannes Gutenberg-Universitat Mainz, 55128 Mainz, Germany 2Chinese Academy of Sciences, Key Lab of Quantum Information, University of Science and Technology of China, Hefei 230026, People's Republic of China 3Institute of Physics, University of Stuttgart and Institute for Quantum Science and Technology IQST, 70174 Stuttgart, Germany 4Application Technology Research Institute, Tokyo Gas Company, Ltd., Yokohama, 230-0045 Japan 5Advanced Materials Laboratory, Sumitomo Electric Industries, Ltd., Itami, 664-0016 Japan 6Takasaki Advanced Radiation Research Institute, National Institutes for Quantum and Radiological Science and Technology, Takasaki, 370-1292, Japan 7Faculty of Pure and Applied Sciences, University of Tsukuba, Tsukuba, 305-8573 Japan 8Helmholtz Institut Mainz, 55099 Mainz, Germany 9Department of Physics, University of California, Berkeley, CA 94720-7300, USA 10Nuclear Science Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA (Dated: July 5, 2019) Ensembles of nitrogen-vacancy (NV) centers in diamonds are widely utilized for magnetometry, magnetic-field imaging and magnetic-resonance detection. At zero ambient field, Zeeman sublevels in the NV centers lose first-order sensitivity to magnetic fields as they are mixed due to crystal strain or electric fields. In this work, we realize a zero-field (ZF) magnetometer using polarization-selective microwave excitation in a 13C-depleted crystal sample. We employ circularly polarized microwaves to √ address specific transitions in the optically detected magnetic resonance and perform magnetometry with a noise floor of 250 pT/ Hz. This technique opens the door to practical applications of NV sensors for ZF magnetic sensing, such as ZF nuclear magnetic resonance, and investigation of magnetic fields in biological systems. I. INTRODUCTION Negatively charged nitrogen-vacancy (NV) centers in diamond have garnered wide interest as magnetome- ters [1 -- 6], with diverse applications ranging from elec- tron spin resonance (ESR) and biophysics to materials science [7 -- 13]. However, typical operation of an NV mag- netometer requires an applied bias magnetic field to non- ambiguously resolve magnetically sensitive features in the level structure. Due to the Zeeman effect, the bias field lifts degeneracy among magnetic sublevels in the NV- center ground state, allowing microwave transitions be- tween spin states to be addressed individually [14]. Such a bias field can be undesirable for applications where it could perturb the system to be measured, such as in mag- netic susceptometry [15] and measurements in magneti- cally shielded environments, or, for example, can create challenging cross talk within sensor arrays [16]. Elimination the need for a bias field would extend the dynamic range of NV magnetometers to zero field. Zero- field, NV-based magnetometry opens up new application avenues, and makes these versatile, solid-state sensors competitive with other magnetic field sensors such as su- perconducting quantum interference devices (SQUIDs) and alkali-vapor magnetometers [17, 18], because, despite the lower sensitivity of NVs, they offer additional benefits due to their small size, high spatial resolution, capability ∗ [email protected] of operation over large temperature and pressure ranges, and wide bandwidth [2]. The relative simplicity of NVs operated at zero field can readily complement existing sensors in applications such as zero- and ultralow-field nuclear magnetic resonance (ZULF-NMR) [19, 20], track- ing field fluctuations in experimental searches for electric dipole moments [21], and magnetoencelography or mag- netocardiography [22, 23]. Magnetically sensitive microwave transitions within NV centers can be probed using the optically detected magnetic resonance (ODMR) technique, which relies on detecting changes in photoluminescence (PL) while ap- plying microwave fields to optically pumped NVs [24]. At zero field, these transitions overlap, and shift equally with opposite sign in response to magnetic fields. Therefore, NV ensembles have been considered unusable as zero-field magnetometers [2], except in certain cases, for detecting ac fields in the presence of applied microwaves [25]. We overcome these complications at zero field by selec- tively driving resolved hyperfine transitions in NV centers in a 13C-depleted diamond with frequency-modulated, circularly polarized microwaves. This results in ODMR fluorescence with a linear response to small magnetic fields. We present such a zero-field NV magnetometer with a demonstrated noise floor of 250 pT/ Hz. √ 9 1 0 2 l u J 3 ] h p - p p a . s c i s y h p [ 3 v 8 9 4 1 1 . 1 1 8 1 : v i X r a 2 and carbon powder prepared by pyrolysis of 99.97% 12C- enriched methane as a carbon source were used [27]. It was irradiated with 2 MeV electrons from a Cockcroft- Walton accelerator to a total fluence of 1.8 × 1018cm−2 at room temperature, and annealed at 800oC for 5 hours. This source diamond was reported to have 3-ppm initial nitrogen and 0.9-ppm NV− after conversion, measured by electron-paramagnetic-resonance techniques [1]. To study the ODMR signal around zero field, fluo- rescence spectra are taken by sweeping the center fre- quency of linearly polarized MWs for a range of magnetic field values around zero. Figure 2 (a) shows the resulting data, which show hyperfine resolved transitions originat- ing from all crystal orientations. The narrow linewidths in this diamond allow for a clear distinction between transitions occurring along different crystal axes. These transitions are shown in detail at specific field values in Fig. 2 (b), including at zero field, where 12 transitions overlap and merge into four distinct features. The NV spin Hamiltonian that describes the energy spectrum, and includes interaction with an applied mag- netic field B and hyperfine interaction with the intrinsic 14N is written as follows H = DS2 x − S2 y ) + ge µB B · S z + E (S2 + S · A · I − gn µN B · I + QI2 z, (1) where D = 2870 MHz is the zero-field-splitting param- eter, E (≈ 0.15 MHz for the employed sample) is the off-diagonal strain- or electric field-splitting tensor [28], S and I are the electron- and nuclear-spin operators, ge and gN are the electron and nuclear-spin g-factors, µB and µN are the Bohr and nuclear magnetons, and Q = −5 MHz is the nuclear-quadrupole-splitting param- eter. The electron spin S and nuclear spin I are coupled via the diagonal hyperfine tensor A⊥ 0 0 0 A⊥ 0 0 A(cid:107) 0  , A = (2) where A⊥=-2.7 MHz and A(cid:107) =-2.16 MHz [29]. To deter- mine the parameter E for the sample, the experimentally measured spectra are fitted to the Hamiltonian with E as a variable and other parameters fixed to values reported in the literature. At low fields, the Hamiltonian is dominated by the 2870-MHz zero-field splitting, D, between the ms = 0(cid:105) and the degenerate ms = ±1(cid:105) states. The hyperfine interaction with the nuclear spin of the NVs' intrinsic 14N results in three hyperfine projections for each electron spin state, and so the MW transitions between the ms states are split threefold [Fig. 1 (b)]. These groupings of ms = ±1(cid:105) states separate in energy with increasing magnetic field due to the Zeeman effect. At fields where the Zeeman shift results in degenerate hyperfine levels of the ms = ±1(cid:105) states, anticrossings between hyperfine states with the same nuclear spin projection occur due to the tensor E [28]. FIG. 1. (a) The experimental setup for a zero-field NV mag- netometer when modulating the MW frequency. PD: photo- diode; MW: microwave; PCB: printed circuit board. (b) The ground-state-level diagram for NV in diamond. The label, ms refers to the electron spin projection, while mI refers to the nuclear spin projection. II. EXPERIMENT A schematic of the experimental apparatus is shown in Fig. 1 (a). Intensity stabilized (<0.5% power fluctuations at 110 mW) green laser light at 532 nm is used to opti- cally pump the NV centers in diamond into a single spin projection (ms = 0) in the ground state. Microwaves (MWs) drive transitions from this ground state into the magnetically sensitive spin states (ms = ±1), reducing the fluorescence resulting from the pumping cycle [26]. The diamond is glued to a parabolic light concentrator to collect fluorescence, which is focused through a filter and onto a photodetector (PD), which registers ≈ 1.2 mW. The parabolic-concentrator arrangement has been demonstrated to have over 60% collection efficiency in Ref. [1]. The signal from the PD is fed into a lock-in am- plifier (LIA), which is referenced to the frequency mod- ulation of the MWs. Three pairs of Helmholtz coils are wound onto a three-dimensionally (3D) printed mount around the diamond sample. To zero the ambient mag- netic field, the coils are driven by three independent sta- ble current supplies. The sensor is a 99.97% 12C, (111)-cut diamond single crystal, with dimensions 0.71 mm× 0.69 mm and a thick- ness of 0.43 mm. It is laser-cut from a 12C-enriched di- amond single crystal grown by the temperature-gradient method at high pressure (6.1 GPa) and high temperature (1430oC). A metal solvent containing a nitrogen getter 3 (a) ODMR spectra with linearly polarized MWs as a function of the axial magnetic field, with transitions originating FIG. 2. from all crystal-axis orientations. Those transitions corresponding to NVs oriented along the direction of the applied field are labeled (ms, mI(cid:105)) and overlaid with the simulation according to the Hamiltonian of Eq. 1. The unlabeled transitions are due to NV centers that are not oriented along the (cid:104)111(cid:105) direction. These features from other orientations overlap because they share a common relative angle with the applied field. (b) Continuous-wave ODMR spectra for selected values of Bz. At zero field only the central transitions are split. The lower-energy peak at ≈ 2868 MHz corresponds to the transitions ms = 0, mI = +1(cid:105) → ms = +1, mI = +1(cid:105) and ms = 0, mI = −1(cid:105) → ms = −1, mI = −1(cid:105). The higher-energy peak at ≈ 2872 MHz corresponds to the transitions ms = 0, mI = −1(cid:105) → ms = +1, mI = −1(cid:105) and ms = 0, mI = +1(cid:105) → ms = −1, mI = +1(cid:105). These anticrossings between hyperfine states are ap- parent in the ODMR Zeeman spectra of Fig. 2 (a). The anticrossings for the transition energies at ≈ ±0.08 mT correspond to interaction between the states ms = −1, mI = ±1(cid:105) and ms = +1, mI = ±1(cid:105). The bot- tom spectrum in Fig. 2 (b) shows strain- and electric field- splitting of the transitions ms = 0, mI = 0(cid:105) → ms = ±1, mI = 0(cid:105) due to the interaction E between the up- per states, while the transitions ms = 0, mI = +1(cid:105) → ms = ±1, mI = +1(cid:105), and ms = 0, mI = −1(cid:105) → ms = ∓1, mI = −1(cid:105) merge in the ODMR spectrum. In many diamond samples these features are indiscernible, be- cause the transverse zero-field splitting E is larger than the intrinsic hyperfine splitting of the NV center. The well-resolved hyperfine structure in this diamond sample allows us to selectively address only the overlapping tran- sitions that occur at ±2 MHz from the central feature. Note that here we use the transitions that overlap at ei- ther ∼2868 MHz or ∼2872 MHz for zero-field magnetom- etry, since the splitting, E, from the strain and/or electric field is suppressed as a result of being at least an order of magnitude smaller than the energy separation between the corresponding branches, ms = ±1, mI = +1(cid:105) or ms = ±1, mI = −1(cid:105), which are split by the hyperfine- coupling term. These features are also used to calibrate the zero-field value of currents in our coils, by maximizing their fluorescence contrast, which occurs when the mag- netic sublevels become degenerate as a result of the exter- nal magnetic field being zeroed. The estimated residual field is less than 3 µT in all directions which corresponds to the linewidth of these transitions. We use circularly polarized MWs to drive a single elec- tron spin transition out of the feature composed of over- lapping resonances. This single transition has a linear dependence on the magnetic field. The circularly po- larized MWs are created using a printed circuit board (PCB) that follows the design of Ref. [30]. The board consists of two, 200-µm wires separated by a distance d = 4.5 mm and placed on a plane that is d/2 away from the diamond sample. Each wire carries a MW signal split from the same source, with one passed through a variable phase shifter. This arrangement results in orthogonal os- cillating magnetic fields at the diamond sample, verified with the high contrast between ODMR traces at fixed field with either left or right circularly polarized MWs [Fig. 3 (a)]. The MW field can be continuously varied between linearly and circularly polarized [Fig. 3 (b)]. The efficacy of our circular MW polarizing scheme is shown in Fig. 3. In particular, Fig. 3 (b,c) demonstrate a relative suppression of σ+ and σ− transitions to below 1% of the maximum contrast, respectively. Previously overlapping transitions are thus isolated, removing the symmetric dependence as a function of the field to be measured. III. MAGNETOMETRY METHOD In typical ODMR magnetometry, linearly polarized MW fields drive transitions between the ms = 0(cid:105) and 4 ms = ±1(cid:105) states, decreasing the detected fluorescence with a resonant response with respect to the MW fre- quency. At fields where the transitions to ms = ±1(cid:105) states are well resolved, modulation is applied to the MW frequency, and the resulting PL signal is detected on a PD and demodulated on a LIA at the modulation frequency. The first-order harmonic output exhibits a linear response of the PL to the magnetic field. How- ever, at zero field, the ms = ±1(cid:105) states are degenerate, and transitions to these states, including those from all crystal orientations, overlap, causing the LIA output to no longer exhibit a linear dependence with the magnetic field, as described below. In our experimental setup, we measure a field that is applied along the (cid:104)111(cid:105) direction. While the angle be- tween the (cid:104)111(cid:105) oriented NVs and the applied field is 0o, the other three NV orientations are all oriented with a 70o angle to the applied field. Therefore, the applied field is projected onto the three other crystal orientations equally, but a single-axis description is sufficient to un- derstand how magnetic field sensitivity arises in the (cid:104)111(cid:105) oriented NVs. For a background PL of P0, the detected signal from applying MWs at frequency ω to a given transition can be modeled as follows A(cid:0) Γ (cid:1)2 (ω − ω0)2 +(cid:0) Γ 2 (cid:1)2 , 2 P = P0 − (3) where P is the PL signal, and ω0, Γ, and A are the cen- ter frequency, the linewidth and the amplitude of this Lorentzian profile. At zero field, the PL signal is convolved with fea- tures due to strain and electric field, however, we can still approximate the signal with the Lorentzian form in Eq. 3. Focusing on the lowest-frequency peak of those shown in the zero-field trace of Fig. 2 (b), two transitions, ms = 0, mI = +1(cid:105) → ms = +1, mI = +1(cid:105) with ampli- tude A+ and central frequency ω0+, and ms = 0, mI = −1(cid:105) → ms = −1, mI = −1(cid:105) with amplitude A− and cen- tral frequency ω0−, overlap such that ω0+ = ω0− = ω0. If the field Bz along a single diamond axis is increased, the central frequencies change by an amount, ω0± = ω0 ± ge µB Bz . (4) The effect of a small modulation of the MW frequency is described by the first-order expansion of P for each transition around ω0 with respect to ω. The result is the linear dependence on the magnetic field, and is a sum of the contributions from each transition. Using the relation in Eq. 4, we find that for small values of Bz, ∆P ∆ω = (A+ − A−)KBz + c, (5) where we group various terms into the parameters K = 8geµB/Γ2 and c = −8(ω − ω0)(A+ + A−)/Γ2. When linearly polarized microwaves are applied, the transition FIG. 3. (a) ODMR traces at fixed field with circular MWs (the polarization of the applied MW is indicated at the right bottom corner of each subfigure: top σ− and bottom σ+). Here the peaks A and D (B and C) correspond to the transi- tions from ms = 0(cid:105) to ms = −1(cid:105) and ms = 0(cid:105) to ms = 1(cid:105) of on-axis (off-axis) NVs, respectively. (b) Fitted amplitudes of A and D in (a), as a function of the relative phase between two applied microwave fields. Blue crosses (amber circles) in- dicate the amplitudes of A (D). The error bars on the data (1 standard error of the mean) are smaller than the symbol size. (c) ODMR spectra under linear, σ−, σ+ MW as a function of magnetic field. The color scale indicates the peak depth, in percent, relative to the off-resonant case. 5 FIG. 4. (a) A detail of the LIA output X around zero field (black line) with a dispersive curve fitting and a linear fit (red) to the data while modulating the frequency of the applied circularly polarized microwave at 3 kHz with a depth of √ 45 kHz. (b) The magnetic-field noise spectrum. The blue line indicates magnetically sensitive noise, the amber line indicates magnetically insensitive at a MW frequency of ≈2900 MHz (average noise between 1 and 1000 Hz is 250 pT/ √ Hz), and the green line indicates electronic noise (average noise between 1 and 1000 Hz is 70 pT/ Hz). The photon shot-noise limit of the Hz. The decrease in signal for frequencies above 1 kHz is due to the filtering of the LIA. magnetometer is indicated at 4 pT/ √ probabilities for each ms state, and therefore the values of A±, are equal. As a result, the linear change in PL is zero for small changes in magnetic field. However, circularly polarized microwaves can be applied instead, resulting in different transition probabilities so that the first term in Eq. 5 does not cancel out, resulting in magnetically sensitive changes in the PL. Therefore, to perform high- sensitivity magnetometry, we apply circularly polarized MWs and modulate the central frequency. Note here that the shape of the magnetometry signal is sensitive to the imperfection of circular MW polarization, which has varying effects for other crystal orientations, and to small detunings of the central MW frequency, which may arise from drifts in the diamond temperature. These effects can explain the asymmetry in Fig. 4 (a). The contributions of the NVs oriented along other crys- tallographic axes also overlap in the lineshape but they have a weaker dependence on the magnetic field (applied at an angle to these axes) and are suppressed when circu- larly polarized MWs are applied. As a result, the linear dependence described above is the dominant behavior for small fields. IV. ALTERNATIVE MAGNETOMETRY METHOD As an aside, we mention that it is also possible to apply an oscillating magnetic field along the (cid:104)111(cid:105) direction in order to perform magnetometry with linearly polarized MWs. This oscillating field modulates the central fre- quencies of the two transitions as follows ω0± = ω0 ± [ge µB Bz + ωmod ] , (6) where ωmod = geµBη sin(νt) and η and ν are the ampli- tude and frequency of the oscillating field, respectively. The resulting PL is the sum of contributions from the rel- evant transitions, each described by the first-order expan- sion of Eq. 3 in ωmod around ωmod = 0. The linear mag- netic dependence for small values of Bz, and η (cid:28) Γ/geµB is, ∆P ∆ωmod = (A+ + A−) KBz + c(cid:48), (7) where K is as defined in Eq. 5 and c(cid:48) = 8(ω − ω0)(A− − A+)/Γ2. The signal is detected on a LIA referenced to the field modulation frequency, ν. Since A± do not can- cel out in Eq. 7, there is linear magnetic sensitivity in the LIA output signal, which, for A− = A+, is insensi- tive to the detuning of the central MW frequency within the linear regime. The magnetic dependence, ∆P/∆Bz, reaches a maximum when η = Γ/2. This method can be simpler to implement in some ap- plications for dc measurements of small fields, since com- pensation coils, for instance, can be used to apply the modulation, and there is no need for circularly polarized microwaves. In certain dc and low-frequency applica- tions, such as for biomagnetic signals, this field modula- tion can be averaged out. Furthermore, the employment of an oscillating bias field relaxes the constraints on bias stability, a concern for precision sensors. V. MAGNETIC FIELD SENSITIVITY To demonstrate the magnetic sensitivity of the zero- field magnetometer, we scan the magnetic field through zero while modulating the frequency of the σ+-polarized MWs, which are centered at ∼ 2872 MHz. The derivative fluorescence signal as detected in a properly phased LIA output depends linearly on the field between ≈ ±1 µT, which determines the dynamic range of the magnetome- ter (which can be extended by applying magnetic bias). The calibration signal is shown in Fig. 4 (a) with a mod- ulation frequency of 3 kHz, and a modulation depth of 45 kHz. The data near zero field are fitted to a straight line, and the slope of this line is used to translate the LIA output signal to the magnetic field. The noise in the LIA output signal voltage therefore conveys the sen- sitivity of the magnetometer. For noise measurements, the LIA output is recorded for 1 s while the background magnetic field is set to zero. The data are passed through a fast Fourier transform and displayed in Fig. 4 (b), from which we can establish our noise floor and thus the sen- sitivity, for a given bandwidth. For noise frequencies between dc-30 Hz we observe a 1/f -behavior of the magnetic noise that we at- tribute to ambient noise, primarily arising from the √ compensation-coil current stability. While the noise floor √ at 250 pT/ Hz can be attributed to laser-intensity noise, a photon shot-noise limit of 4 pT/ Hz is achievable -- a value calculated from the number of incident photons at the photodetector and the contrast of the magnetically sensitive feature in the MW spectral response. The mag- netically insensitive noise spectrum is obtained by oper- ating the setup at an off-resonant microwave frequency of 2900 MHz, where there are no magnetically sensitive features in the NV spectrum. Since noise peaks at 50 Hz and harmonics are absent in this magnetically in-sensitive √ spectrum, we attribute them to magnetic noise in the lab- oratory. The electronic noise floor (≈ 70 pT/ Hz) was measured by turning off the green excitation light and acquiring the output of the LIA. 6 VI. CONCLUSION √ We demonstrate a NV-based zero-field magnetometer with a 250 pT/ Hz noise floor. The device employs a diamond sample with a well-resolved ground-state hy- perfine structure of the NV center, and uses circularly polarized microwaves to selectively excite magnetically sensitive transitions that, in the absence of such selectiv- ity, yield ODMR signals that are first-order insensitive to near-zero magnetic fields. This device can be use- ful in applications where a bias field is undesirable and extends over the dynamic range of NV magnetometry to cover existing zero-field technologies such as SQUIDs and alkali-vapor magnetometers. ZF magnetometry with sin- gle NVs will be presented in a later publication. Improve- ments in the present technique will result in sensitivities that are useful for ZULF-NMR and, with further minia- turization, these zero-field diamond sensors can find use in biomagnetic applications such as magnetoencelogra- phy and magnetocardiography. VII. ACKNOWLEDGEMENTS We gratefully acknowledge W. Gawlik for provid- ing an early version of the circuit board for produc- ing circularly polarized microwaves. We thank G. Balasubramanian, Philipp Neumann, A. Jarmola, G. Chatzidrosos and J. W. Blanchard for informative dis- cussions and fruitful advice. This work was supported by the EU FET-OPEN Flagship Project ASTERIQS (Ac- tion 820394) and the German Federal Ministry of Educa- tion and Research (BMBF) within the Quantumtechnolo- gien program (FKZ 13N14439) and the Japan Society of the Promotion of Science (JSPS) KAKENHI (Grant No.17H02751). [1] T. Wolf, P. Neumann, K. Nakamura, H. Sumiya, T. Ohshima, J. Isoya, and J. Wrachtrup, Subpicotesla diamond magnetometry, Phys. Rev. X 5, 041001 (2015). [2] L. Rondin, J.-P. Tetienne, T. Hingant, J.-F. Roch, P. Maletinsky, and V. Jacques, Magnetometry with nitrogen-vacancy defects in diamond, Reports on Progress in Physics 77, 056503 (2014). [3] A. Wickenbrock, H. Zheng, L. Bougas, N. Leefer, S. Afach, A. Jarmola, V. M. Acosta, and D. Budker, Microwave-free magnetometry with nitrogen-vacancy centers in diamond, Applied Physics Letters 109, 053505 (2016). [4] H. Zheng, G. Chatzidrosos, A. Wickenbrock, L. Bougas, R. Lazda, A. Berzins, F. H. Gahbauer, M. Auzinsh, R. Ferbe, and D. Budker, Level anti-crossing magne- tometry with color centers in diamond, Proc. of SPIE , 101190X (2017). [5] A. Blank, G. Shapiro, R. Fischer, P. London, and D. Ger- shoni, Optically detected magnetic resonance imaging, Applied Physics Letters 106, 034102 (2015). enhanced room-temperature magnetometry using ab- sorption by nitrogen-vacancy centers in diamond, Phys. Rev. Lett. 112, 160802 (2014). [7] F. Jelezko and J. Wrachtrup, Single defect centres in di- amond: A review, physica status solidi (a) 203, 3207 (2006). [8] F. Kong, P. Zhao, X. Ye, Z. Wang, Z. Qin, P. Yu, J. Su, F. Shi, and J. Du, Nanoscale zero-field electron spin res- onance spectroscopy, Nature Communications 9, 1563 (2018). [9] J. F. Barry, M. J. Turner, J. M. Schloss, D. R. Glenn, Y. Song, M. D. Lukin, H. Park, and R. L. Walsworth, Optical magnetic detection of single-neuron action po- tentials using quantum defects in diamond, Proceedings of the National Academy of Sciences 113, 14133 (2016). [10] G. Balasubramanian, I. Chan, R. Kolesov, M. Al-Hmoud, J. Tisler, C. Shin, C. Kim, A. Wojcik, P. R. Hemmer, A. Krueger, et al., Nanoscale imaging magnetometry with diamond spins under ambient conditions, Nature 455, 648 (2008). [6] K. Jensen, N. Leefer, A. Jarmola, Y. Dumeige, V. M. Acosta, P. Kehayias, B. Patton, and D. Budker, Cavity- [11] S. Steinert, F. Dolde, P. Neumann, A. Aird, B. Nayde- nov, G. Balasubramanian, F. Jelezko, and J. Wrachtrup, High sensitivity magnetic imaging using an array of spins in diamond, Review of Scientific Instruments 81, 043705 (2010). [12] C. L. Degen, Scanning magnetic field microscope with a diamond single-spin sensor, Applied Physics Letters 92, 243111 (2008). [13] P. London, J. Scheuer, J.-M. Cai, I. Schwarz, A. Ret- zker, M. B. Plenio, M. Katagiri, T. Teraji, S. Koizumi, J. Isoya, R. Fischer, L. P. McGuinness, B. Naydenov, and F. Jelezko, Detecting and polarizing nuclear spins with double resonance on a single electron spin, Phys. Rev. Lett. 111, 067601 (2013). [14] J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, P. Neumann, D. Budker, P. R. Hemmer, A. Yacoby, R. Walsworth, and M. D. Lukin, High-sensitivity dia- mond magnetometer with nanoscale resolution, Nature Physics 4, 810 (2008). [15] D. Eberbeck, A. P. Astalan, K. Petersson, F. Wiekhorst, C. Bergemann, C. Johansson, U. Steinhoff, H. Richter, A. Krozer, and L. Trahms, in 4th European Conference of the International Federation for Medical and Biological Engineering, edited by J. Vander Sloten, P. Verdonck, M. Nyssen, and J. Haueisen (Springer Berlin Heidelberg, Berlin, Heidelberg, 2009) pp. 2317 -- 2321. [16] R. Jimenez-Martnez, S. Knappe, and J. Kitching, An op- tically modulated zero-field atomic magnetometer with suppressed spin-exchange broadening, Review of Scien- tific Instruments 85, 045124 (2014). [17] G. Bao, A. Wickenbrock, S. Rochester, W. Zhang, and D. Budker, Suppression of the nonlinear zeeman effect and heading error in earth-field-range alkali-vapor mag- netometers, Phys. Rev. Lett. 120, 033202 (2018). [18] W. Quan, K. Wei, and H. Li, Precision measurement of magnetic field based on the transient process in a k-rb- 21ne co-magnetometer, Opt. Express 25, 8470 (2017). [19] M. Jiang, T. Wu, J. W. Blanchard, G. Feng, X. Peng, and D. Budker, Experimental benchmarking of quantum control in zero-field nuclear magnetic resonance, Science Advances 4 (2018). [20] M. P. Ledbetter, I. M. Savukov, D. Budker, V. Shah, S. Knappe, J. Kitching, D. Michalak, S. Xu, and A. Pines, Zero-field remote detection of nmr with a mi- crofabricated atomic magnetometer, Proceedings of the National Academy of Sciences of the United States of America 105, 2286 (2008). [21] G. Bison, V. Bondar, P. Schmidt-Wellenburg, A. Schn- abel, and J. Voigt, Sensitive and stable vector magne- 7 tometer for operation in zero and finite fields, Opt. Ex- press 26, 17350 (2018). [22] A. I. Ahonen, M. S. Hamalainen, M. J. Kajola, J. E. T. Knuutila, P. P. Laine, O. V. Lounasmaa, L. T. Parkko- nen, J. T. Simola, and C. D. Tesche, 122-channel squid instrument for investigating the magnetic signals from the human brain, Physica Scripta 1993, 198 (1993). [23] R. Fenici, D. Brisinda, and A. M. Meloni, Clinical appli- cation of magnetocardiography, Expert Review of Molec- ular Diagnostics 5, 291 (2005). [24] E. van Oort, P. Stroomer, and M. Glasbeek, Low-field op- tically detected magnetic resonance of a coupled triplet- doublet defect pair in diamond, Phys. Rev. B 42, 8605 (1990). [25] S. Saijo, Y. Matsuzaki, S. Saito, T. Yamaguchi, I. Hanano, H. Watanabe, N. Mizuochi, and J. Ishi- Hayase, Ac magnetic field sensing using continuous- wave optically detected magnetic resonance of nitrogen- vacancy centers in diamond, Applied Physics Letters 113, 082405 (2018). [26] K. Jensen, P. Kehayias, and D. Budker, Magnetometry with nitrogen-vacancy centers in diamond, in High Sen- sitivity Magnetometers, edited by A. Grosz, M. J. Haji- Sheikh, and S. C. Mukhopadhyay (Springer International Publishing, Cham, 2017) pp. 553 -- 576. [27] K. Nakamura, S. Yamashita, T. Tojo, M. Mitsuishi, K. Kataoka, and M. Yoshimoto, Single-crystal synthe- sis of highly thermal conductive 12c-enriched diamond from pyrolytic carbon powder by the high-pressure, high- temperature method, Diamond and Related Materials 16, 1765 (2007). [28] T. Mittiga, S. Hsieh, C. Zu, B. Kobrin, F. Machado, P. Bhattacharyya, N. Z. Rui, A. Jarmola, S. Choi, D. Budker, and N. Y. Yao, Imaging the local charge en- vironment of nitrogen-vacancy centers in diamond, Phys. Rev. Lett. 121, 246402 (2018). [29] S. Felton, A. M. Edmonds, M. E. Newton, P. M. Mar- tineau, D. Fisher, D. J. Twitchen, and J. M. Baker, Hy- perfine interaction in the ground state of the negatively charged nitrogen vacancy center in diamond, Phys. Rev. B 79, 075203 (2009). [30] M. Mrozek, J. Mlynarczyk, D. S. Rudnicki, and W. Gaw- lik, Circularly polarized microwaves for magnetic reso- nance study in the ghz range: Application to nitrogen- vacancy in diamonds, Applied Physics Letters 107, 013505 (2015).
1712.01165	2	1712	2018-09-26T13:11:18	Magnetic fields: a tool for the study of organic solar cells	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Charge transfer in polymer devices represents a crucial, though highly inaccessible stage of photocurrent generation. In this article we propose studying the properties and behaviour of organic solar cells through the modification of photocurrent generation when an external magnetic field is applied. By allowing the parameters of our theoretical model not to be constrained to any specific material, we are able to show that not only a modest external magnetic field leads to a significant increase in photocurrent intensity, but also how such magnetic field can be used to study in detail the energy levels and transition rates within the polymer compound. Systematic exploration of key properties in organic composites thus can lead to highly optimised devices in which a magnetic field produces an enhancement in the efficiency of polymer solar cells.	physics.app-ph	physics	Magnetic fields: a tool for the study of organic solar cells S. Oviedo-Casado1∗ and A. Urbina2, and J. Prior1,3 1 Departamento de F´ısica Aplicada, Universidad Polit´ecnica de Cartagena, Cartagena 30202, Spain 2 Departamento de Electr´onica, Universidad Polit´ecnica de Cartagena, Cartagena 30202, Spain 3Instituto Carlos I de F´ısica Te´orica y Computacional, Universidad de Granada, Granada 18071, Spain 8 1 0 2 p e S 6 2 ] h p - p p a . s c i s y h p [ 2 v 5 6 1 1 0 . 2 1 7 1 : v i X r a Charge transfer in polymer devices represents a crucial, though highly inaccessible stage of photocurrent generation. In this article we propose studying the properties and behaviour of organic solar cells through the modification of photocurrent generation when an external magnetic field is applied. By allowing the parameters of our theoretical model not to be constrained to any specific material, we are able to show that not only a mod- est external magnetic field leads to a significant increase in photocurrent intensity, but also how such magnetic field can be used to study in detail the energy levels and transition rates within the polymer compound. System- atic exploration of key properties in organic composites thus can lead to highly optimised devices in which a magnetic field produces an enhancement in the efficiency of polymer solar cells. I. INTRODUCTION The ubiquitous goal in photovoltaic research is to improve the performance of existing materials and con- figurations, as nowadays the best polymeric photovoltaic (OPV) cells present an efficiency not surpassing 11% [1 -- 4]. Typical configurations for the active layer of these devices are conjugated polymers blended with fullerene derivatives [5 -- 7], whose quantum efficiency, namely the ratio of absorbed photon to created exciton, raises in these compounds above 0.9 [8, 9]. This means that there exists a necessity to bridge the gap existing between the initial high efficiency and the final photocurrent generated, and for that purpose the position of the energy levels -- and in particular the charge transfer states -- has been proposed to be critical [10]. In addition, photocurrent generation is highly dependent on the recombination rates of the different energy levels, hence they are strongly responsible for the final efficiency of OPVs [11, 12]. Although the link between the relative position of intermediate energy level bands and global recombination rates is not yet well understood, it is clear that unravelling the inner dynamics of polymer compounds holds the key to fabricate improved polymer solar cells. Intermediate stages in the process of photocurrent generation on polymer devices are mediated by electron- hole (e-h) pairs which, contrary to what happens in other photovoltaic systems such as photosynthetic com- plexes, involve transport both of charge and energy [13]. Moreover, OPVs require the electron to be com- pletely separated from the hole in order to generate current. In between the formation of an electron-hole pair and its eventual dissociation (or recombination), lies a series of ladder-like steps in which the electron progressively separates from the hole. These are denominated charge transfer (CT) states, and minimising recombination during the time the electron-hole pair is in this stage is crucial to maximise photocurrent gener- ation [7, 12, 14]. Studying charge transfer states is however difficult, since they are nearly dark and therefore not sensitive enough for spectroscopic techniques [7]. On the other hand, charge transfer states involve spin 1/2 particles, susceptible to the influence of a magnetic field, which can be used to alter the properties of charge transfer states and thus modify the generated photocurrent -- an easily measured outcome -- hence providing in- formation about the different properties of the OPV, and allowing to systematically search for optimal designs. In addition, applying a magnetic field could also lead to more efficient solar cells, leading polymer devices to be eventually commercially advantageous over solid state, semiconductor solar cells. ∗ [email protected] 2 In this article, we complement and expand the results presented in [15], where although a magnetic field was shown to be able to influence the photocurrent generation process by enhancing triplet charge transfer state formation via an increased inter-system crossing, it was found that the required magnetic fields were too strong to be of commercial interest. Thus potential experimental applications were mentioned. In this article, not only we refine the model to show that a significant increase in photocurrent intensity can be obtained at lower magnetic fields and for realistic polymer materials, but we primarily focus on demonstrating with numerous examples the experimental possibilities offered by employing a magnetic field as a tool to elucidate the different and optimal characteristics of real polymer compounds. This article is organised as follows: Section 2 introduces the OPV theoretical model that we use. Section 3 is devoted to present the results for an OPV in which the magnetic field produces appreciable changes in photocurrent intensity. In Section 4 the different spectroscopic possibilities and applications of an external magnetic field are demonstrated in detail. We finish the article with some conclusions in Section 5. II. MODEL In order to fully understand the process of photogeneration in polymer solar cells, microscopic theoretical models are required; from photoexcitation, through charge separation, until the electron reaches the positive plate, as is schematically shown in the upper panel of Figure 1, where we already glimpse the difficulty of coming up with an exact model, owing to the high number of degrees of freedom composing even the most simple polymer:fullerene blends, and the typically disordered structure that such materials present [16]. We propose here an heuristic approach to the problem, in which we identify the dynamically relevant independent degrees of freedom, and associate a quantum state to each of them, thus having an ansatz capable of qualita- tively reproducing the behaviour of realistic OPVs which at the same time is able to provide information about the key processes occurring within the blend. The model that we propose is depicted in the lower panel of Fig. 1. An incoming photon generates an electron-hole pair (exciton e, S(cid:105)) in the polymer, acting here as a donor. The electron rapidly migrates to the interface between donor and acceptor (fullerene in 1(a)) where, being loosely bound to the original hole, it forms a charge transfer state (CT, S(cid:105)). As this process occurs with near unity quantum efficiency, and in a much faster time-scale than all subsequent evolution (i.e ≈ 100 fs), we can regard it as instantaneous throughout the present study [9] and therefore, we will consider initial excitation to take place directly to a charge transfer state. Subsequent evolution sees the electron either dissociating completely, or recombining with the same hole (geminate) or a different one (non-geminate) into the ground state g, S(cid:105). If dissociation occurs photocurrent is generated and we describe it as a free charge (FC(cid:105)) state. It is the delicate balance between recombination and free charge generation what sets the global efficiency of the OPV. The initially generated e-h pairs conserve the spin of the ground state, i.e. they are in a singlet state and we have labelled them accordingly. However, the possibility of having triplet charge transfer states (CT, T(cid:105)) is real, and leads to a different, complementary pathway of generating free charges. Moreover, as the ground state has spin zero, direct recombination of triplet CT states is forbidden [10] and has to happen through an auxiliary triplet exciton state (e, T(cid:105)), hence the dissociation of free charges is in principle more efficient if it happens from triplet CT states.The necessary conversion from the initial singlet CT to the more convenient triplet CT -- a process called inter-system crossing (ISC) -- occurs either via hyperfine and spin-orbit coupling or due to wavefunction overlap, all of them too small to be relevant in OPVs. The alternative is to enhance spin flips employing an external magnetic field [17 -- 20]. That an external magnetic field is able to enhance inter-system crossing has been controversial [12, 20], owing mainly to the time-scales involved. For an external magnetic field to be able to induce spin flips the precession frequency induced by the magnetic field has to be faster than a) the time it takes for the spins to align in the direction of the magnetic field and b) the recombination rates for singlet CT states. Up to now, the time-scales studied were in the range of µs, where we are in the scenario where the magnetic field only aligns 3 Figure 1. (a) shows an schematic organic solar cell structure, with a donor-acceptor layer composed of P3HT:PCBM, where incoming light induces electronic transitions to a higher state of a π orbital in the donor. The Coulombically bound electron-hole pair rapidly migrates to the donor:acceptor interface where charge separation occurs and a net photocurrent is created, leading ultimately to electricity generation. On (b), our model solar cell and photocurrent generation process proposal is presented. Incoming photons create singlet exciton states that evolve into singlet CT states in hundreds of fs. Inter-system crossing converts coherently back and forth between singlet and triplet CTs, the later of which are split due to Zeeman effect caused by the interaction with an external magnetic field. CT states can either recombine to the ground state or dissociate to form free charges. Both processes are incoherent, represented by black, solid arrows whose relative width represents the characteristic rates for the corresponding transitions. An impedance between the free charge and the ground state allows to study the I-V characteristics of the model. spins and potential ISC is either neutral (i.e. not observable) or negative, as it converts already existing triplet CTs into singlets. Crucially, our model follows those of Friend et al. and considers evolution of CT states in the nanosecond time-scale, where precession of electron spins and differences existing in the coupling of the electron and the hole to the external magnetic field (mainly due to a different g-factor) make the external magnetic field not only an important source but the main source of ISC. A combination of this effect with a slower triplet CT recombination pathway (w.r.t singlet CT) generates a situation in which the population of the triplet CT state is enhanced, thus leading to faster FC generation. (a)(b)The Hamiltonian describing the model presented in Fig. 1(b) is HS = E0 g, S(cid:105)(cid:104)g, S + Es CT, S(cid:105)(cid:104)CT, S + Et CT, T(cid:105)(cid:104)CT, T + Eet e, T(cid:105)(cid:104)e, T + EFC FC(cid:105)(cid:104)FC + Hm, 4 (1) including all the relevant energy levels and where interaction with an external magnetic field is accounted for by Hm = 2µBgT BCT, T(cid:105)(cid:104)CT, T + ΩIS C(B)(CT, S(cid:105)(cid:104)CT, T + CT, T(cid:105)(cid:104)CT, S) , (2) (cid:17) (cid:88) (cid:16) T =0,±1 with µB the Bohr magneton, g is the g-factor, T runs over the different triplet spin values and B is the magnetic field. The factor ΩIS C(B) includes all possible sources of inter-system crossing, both internal (mainly hyperfine coupling) as well as the external magnetic field. Incoherent transitions between energy levels, including recombination and charge dissociation are ac- counted for via Lindbladian terms of the form Lα = γα (cid:34) (cid:110) (cid:111)(cid:35) σαρ(t)σ α − 1 † 2 † ασα, ρ(t) σ , (3) where each γα represents a different transition rate (see Fig. 1(b)), and each operator σα describes the hopping between corresponding energy levels. In this manuscript we are interested in obtaining a measurement of the performance of the corresponding solar cell. With this purpose, the intensity of the generated photocurrent at the steady state becomes the ideal quantity to be analysed and eventually optimised. Such intensity is gauged as the amount of free charges that decay to the ground state via an artificial impedance simulated as an infinite particle reservoir with cou- pling rate Γ. With this definition, the photocurrent intensity is I = eρFC,FCΓ and the voltage difference is V = EFC − Eg + 1 with kB the Boltzmann constant and TΓ the corresponding impedance (here simulated as an environment) temperature, which is room temperature throughout this manuscript. Here, the voltage arises from the energy released in equilibrating the system with an environment at temperature TΓ, and is calculated through the derivative of the logarithm of the correspond- ing partition function. The density matrix elements in these definitions are obtained from the steady state solution of the full Lindblad equation , with e the electron charge, and β = 1 kBTΓ β ln ρFC,FC ρgg = −i(cid:2)HS , ρ(t)(cid:3) + dρ dt (cid:2)ρ(t)(cid:3) , Lα (cid:88) α (4) whose initial state is the ground state and where solar-like excitation to the singlet CT is simulated as the interaction with a particle reservoir of high temperature photons with upwards transition rate of γsun = 0.1. III. EFFICIENT SOLAR CELL IN A MAGNETIC FIELD In this section we present a state-of-the-art polymer solar cell subject to the influence of a magnetic field, demonstrating that with realistic parameters, a meaningful increase in both intensity and power outcome with relatively small magnetic fields can be achieved. Figure 2 shows the I-V characteristic for our solar cell model in the presence of a magnetic field ranging from zero to two Tesla, together with the generated power corresponding to such intensity, defined as P = I·V. Both characteristics (I-V and P-V), which are representative of the OPV performance, present a significant increase in the presence of a magnetic field. In [15] we employed the same model but with restringing parameters 5 that matched those of a well known but scantily efficient polymer compound, such as the P3HT can be. Here instead, we allow for much more freedom in order to search for the ideal polymer material that maximises the enhancing effect of a magnetic field in photocurrent. Thus we choose a typical singlet CT recombination rate of one nanosecond [13], while allowing for the triplet CT recombination rate to be one order of magnitude slower, hence favouring the protection of population in such triplet CT state, thus facilitating dissociation of charges into free charge states and photocurrent against recombination and losses. Note that even though we choose parameters in line with the desired effect we seek, these remain realistic, as it is now possible in frontier materials to achieve the effective freezing of the triplet CT states [10, 22], consequently turning off their recombination pathway. One can easily imagine how in such materials, employing a magnetic field should be -- as we propose -- ideal to enhance the performance of polymer solar cells. In addition, the inset in Figure 2 shows the percentage increase in photocurrent intensity and power whose origin is the external magnetic field. Such percentage, calculated at the point of maximum power, represents an alternative, complementary measurement of the OPV performance. The magnitude of the intensity and power shown in Fig. 2 depends on the solar-like illumination γsun, which is an statistical measure of the number of incoming solar photons and provides numerical values for the different parameters that are in such a way normalised to a standard "size" unit of the solar cell. We would like to emphasise that despite not having the more usual intensity per unit area typical from experimental results in solar cells, the percentage increase is independent of this solar-like illumination, thus the 6% increase at two Tesla represent a significant step forward in the quest of obtaining polymer solar cells that can outcompete commercially the classical semiconductor solar cells. The quality of a solar cell can be measured by its filling factor (FF), which is defined through the maximum power attainable as Pmp = Imp · Vmp = FF · IS C · VOC, where IS C is the current at short circuit (i.e zero voltage) and VOC is the voltage at open circuit (i.e no current). We observe that for the model example presented in Fig. 2 the filling factor is 0.9, and it increases slightly with the magnetic field, providing an independent measure of the beneficial effect that can be obtained from applying an external magnetic field. Such number is above the best FF found nowadays in real OPVs, which reach up to around 0.8 [21]. The difference is easily explained in the simplicity of our model, which does not take directly into account factors such as electron mobility or structural disorder, which in real experiments always lower the quality of an OPV, and which are inevitable. IV. MAGNETS AS LABORATORY INSTRUMENTS FOR SOLAR CELL INVESTIGATION After demonstrating that we can directly interfere in the photocurrent generation process of an OPV, let us in this section introduce a different, complementary, and potentially very useful application of an external mag- netic field. Namely its capacity as an experimental tool to access information about energy levels disposition and transport rates. Through the modification of the generated photocurrent, we will show in the subsequent paragraphs that, based on a model such as we presented in section II it is possible to infer certain properties of the OPV subject to the magnetic field. Information about transition rates and energy levels is usually obtained via spectroscopic techniques, such as pump-probe or 2D spectroscopy. These methods rely on the capacity of the energy states to absorb and emit radiation and thus, whenever the amount of states is too big or they lack optical strength being therefore dark, getting information via spectroscopy is challenging. On the other hand, employing a magnetic field to affect the photocurrent intensity provides an indirect measure that does not depend so heavily on the above mentioned properties of the states, thus being much more reliable for the case of OPVs. 6 Figure 2. Simulated I-V characteristic (continuous) and power outcome (dashed) of the model solar cell with singlet CT recombination rate of 1 ns while the triplet CT recombination rate is one order of magnitude slower. The different values of the external magnetic field (ranging here from zero to two Tesla) show how a moderate increase leads to enhancement in the performance of the OPV. The inset shows the percentage increase in intensity (dots) and power (squares) calculated at the point of maximum power. Geminate recombination rate γgem is in this case 0.1 ns, with non-geminate recombination rate γnon−gem being 10 ns, free charge generation rate γ f c is 1 ns and triplet exciton recombination rate γrelax is 10 ns. The asterisk in the intensity and power units means that we work with units normalised to a simulated solar-like illumination (see main text). A. Rates variations Firstly, we study the behaviour of the current intensity curves when the OPV is subject to a magnetic field between 0 and 2 Tesla, and test the differences appearing when the various transition rates in our model (see Fig. 1(b)) change. Figure 3 shows the resulting percentage variation on the intensity of photocurrent for different values of each transition rate. It is clear that for all except for the non-geminate recombination rate, it is possible to extract information from the variation of the intensity in the presence of a magnetic field. This latter case does not show an appreciable variation due to its relative independence from the ISC as it affects directly and in the same way both singlet and triplet CT, hence not depending on the population present in the triplet CT state, but rather only on the global population present in the FC state. It is clear from this analysis where the emphasis in design should be put. Both the free charge generation and geminate recombination hold the key to not only improved designs but also to obtain astounding increases in efficiency due to the presence of an external magnetic field (up to 200 % in the case of the geminate recom- bination rate). Moreover, we find that the increment tends to saturate, thus stating that there is no troublesome infinite energy generation possibility, which is a double-check to the validity of our model. For the remaining 0 2 4 6 8 10 12 14 16 18 0 0.5 1 1.5 2 0 2 4 6 8 10 12 14 16 18I (mA)P = I·V (mW)V (Volts)B = 0B = 0.5B = 1B = 1.5B = 2 0 1 2 3 4 5 6 7 0 5 10 15 20 0 1 2 3 4 5 6(I(B)-I(0))/I(0) (%)(P(B) - P(0))/P(0) (%)B (T)7 Figure 3. From left to right and top to bottom, percentage variation with respect to 0 Tesla external magnetic field, on the intensity of the generated photocurrent for different incoherent transitions rates, labelled as in Fig. 1(b). (a) shows variations of the triplet CT recombination rate. (c) presents the triplet recombination rate variation. (d) presents the variation of the non-geminate recombination rate, though no difference at all is appreciable, (e) displays the curves from different geminate recombination rates while (f) is intended to amplify the last four curves in (e). In all cases the rates which are kept constant while one of them is varied, acquire values in line with those presented in Fig. 2. (b) is the variation of the FC generation rate. rates, the variations range from moderate to non-existent in the case of the non-geminate recombination rate. Where no more values have been calculated is because the resulting curves where indistinguishable from the presented ones. The fact that each of the rates display different influence in the photocurrent intensity in the presence of an external magnetic field means that a multivariate analysis of experimental results from several different samples, should be able to discern and infer the different values of the incoherent rates for each of the samples. Thus it is demonstrated the power of an external magnetic field to easily characterise different materials and designs. 012345600.511.520123456(a)(I(B)−I(0))/I(0) (%)B (Tesla)1 μs10 ns100 ns02040608010012014016000.511.52020406080100120140160(b)(I(B)−I(0))/I(0) (%)B (Tesla)100 ns10 ns1 ns100 ps10 ps1 ps012345600.511.520123456(c)(I(B)−I(0))/I(0) (%)B (Tesla)10 μs1 μs100 ns10 ns1 ns012345600.511.520123456(d)(I(B)−I(0))/I(0) (%)B (Tesla)1 ns1 μs05010015020000.511.52050100150200(e)(I(B)−I(0))/I(0) (%)B (Tesla)100 ns10 ns1 ns100 ps10 ps1 ps012345600.511.520123456(f)(I(B)−I(0))/I(0) (%)B (Tesla)100 ns10 ns1 ns100 ps8 Figure 4. Percentage variation on the intensity of the generated photocurrent, for different magnetic fields and changing the energies of different states. In (a) and (b), changing the energy of the free charge state and the triplet exciton state leads to no differences. In (c), changing the energy (in eV) of the triplet charge transfer state leads to a huge increase in the generated photocurrent whenever the triplet CT is close to resonance with the singlet CT (upper, yellow curve), while the effect is smaller if it is further from resonance or if it lies above the singlet CT. B. Energy level variations Only loosely bound e-h pairs are significantly affected by a magnetic field. This means that gaining in- formation about exciton states using an external magnetic field is generally not viable. Moreover, Fig. 4(top) shows that varying the energy of either the FC or the triplet exciton state has no influence on the intensity curve at any magnetic field. Thus in principle information of these states can only be accessed via modelling and through the variations in the corresponding transition rates, as we saw in the previous subsection. Furthermore it tells us that there is not much to gain from changing the relative position of these energy levels via clever engineering of OPVs. Unless the triplet exciton lays above the triplet CT state, in whose case decay of the triplet CT is strongly forbidden, which would lead to a great increase in intensity in the presence of a magnetic field, as can be concluded from the previous subsection. There is however much to be learned from the study of the variations that a magnetic field produces in intensity when the triplet CT state energy changes. We observe in Fig. 4 that a magnetic field is able to discern between different triplet CT energy configurations and, in addition, that the closer the triplet CT is to the singlet CT in energy -- while being below -- the greater the enhancement in photocurrent intensity. However notice that complete resonance is not desired, as Zeeman splitting, while small enough, would render triplet states to lie above the singlet CT, thus making ISC transitions increasingly hindered. 0 1 2 3 4 5 6 0 0.5 1 1.5 2 0 1 2 3 4 5 6(a)FC state(I(B)-I(0))/I(0) (%)B (Tesla) 0 1 2 3 4 5 6 0 0.5 1 1.5 2 0 1 2 3 4 5 6(b)Triplet exciton(I(B)-I(0))/I(0) (%)B (Tesla) 0 10 20 30 40 50 60 70 80 90 0 0.5 1 1.5 2 0 10 20 30 40 50 60 70 80 90(c)Triplet CT(I(B)-I(0))/I(0) (%)B (Tesla)1.0001.2501.4751.525V. CONCLUSIONS 9 Although historically an external magnetic field has been discredited as a potential useful tool in organic photovoltaics [12, 20], we have demonstrated that its benefit is twofold. On the one hand, clever engineering of materials can lead to great increases in generated photocurrent and efficiency when the OPV is in the presence of a magnetic field. An external magnetic field is able to alter the balance existing between recombination and generation of free charges, inclining it more towards a greater generation, as can be seen in Fig. 3. And on the other hand, magnetic fields have the potential to become an essential part of the laboratory toolkit that analyses the properties and behaviour of different organic materials. Employing moderate magnetic fields can allow for access to information about each transition rate between the different energy levels, as well as the position of some of this energy levels, thus permitting an almost complete characterisation of the material subject to study. This means that the cost of designing new, improved materials from the existent ones can be greatly reduced, as now the information about the ideal properties the sample must have is readily available. VI. ACKNOWLEDGEMENT This research was undertaken with financial support from MINECO (SPAIN), including FEDER funds: FIS2015-69512-R and No. ENE2016-79282-C5-5-R together with Fundaci´on S´eneca (Murcia, Spain) Project No. 19882/GERM/15. VII. AUTHOR CONTRIBUTIONS S.O.C. did the numerical simulations, J.P. and S.O.C. developed the theory. AU proposed the model and its application to PV technology. The manuscript was written by S.O.C. with significant contributions from all the other authors. [1] Martin A. Green, Keith Emery, Yoshihiro Hishikawa, Wilhelm Warta, and Ewan D. Dunlop. Solar cell efficiency tables (version 45). Progress in Photovoltaics: Research and Applications, 23(1):1 -- 9, 2015. PIP-14-274. [2] Stoichko D. Dimitrov and James R. Durrant. Materials design considerations for charge generation in organic solar cells. Chemistry of Materials, 26(1):616 -- 630, 2014. [3] Zhicai He, Chengmei Zhong, Shijian Su, Miao Xu, Hongbin Wu, and Yong Cao. Enhanced power-conversion efficiency in polymer solar cells using an inverted device structure. Nature Photonics, 6:591 -- 595, 2012. [4] M.C. Scharber, D. Mhlbacher, M. Koppe, P. Denk, C. Waldauf, A.J. Heeger, and C.J. Brabec. Design rules for donors in bulk-heterojunction solar cellstowards 10% energy-conversion efficiency. Advanced Materials, 18(6):789 -- 794, 2006. [5] C. X. Sheng and Z. V. Vardeny. Optical Studies of Photoexcitations in Polymer/Fullerene Blends for Organic Photo- voltaic Applications, pages 3 -- 41. Springer Berlin Heidelberg, Berlin, Heidelberg, 2015. [6] Sung Heum Park, Anshuman Roy, Serge Beaupre, Shinuk Cho, Nelson Coates, Ji Sun Moon, Daniel Moses, Mario Leclerc, Kwanghee Lee, and Alan J Heeger. Bulk heterojunction solar cells with internal quantum efficiency ap- proaching 100%. Nature Photonics, 3(5):297 -- 303, 2009. [7] Koen Vandewal, Kristofer Tvingstedt, Abay Gadisa, Olle Inganas, and Jean V Manca. On the origin of the open- circuit voltage of polymerfullerene solar cells. Nature Materials, 8(11):904 -- 909, 2009. [8] A A Bakulin, A Rao, V G Pavelyev, P H M van Loosdrecht, M S Pshenichnikov, D Niedzialek, J Cornil, D Beljonne, and R H Friend. The Role of Driving Energy and Delocalized States for Charge Separation in Organic Semiconduc- tors. Science, 335(6074):1340 -- 1344, 2012. 10 [9] Simon G´elinas, Akshay Rao, Abhishek Kumar, Samuel L Smith, Alex W Chin, Jenny Clark, Tom S van der Poll, Guillermo C Bazan, and Richard H Friend. Ultrafast long-range charge separation in organic semiconductor photo- voltaic diodes. Science (New York, N.Y.), 343(6170):512 -- 6, 2014. [10] Akshay Rao, Philip C Y Chow, Simon G´elinas, Cody W Schlenker, Chang-Zhi Li, Hin-Lap Yip, Alex K-Y Jen, David S Ginger, and Richard H Friend. The role of spin in the kinetic control of recombination in organic photo- voltaics. Nature, 500(7463):435 -- 9, 2013. [11] Carsten Deibel, Thomas Strobel, and Vladimir Dyakonov. Role of the charge transfer state in organic donoracceptor solar cells. Advanced Materials, 22(37):4097 -- 4111, 2010. [12] P. B. Deotare, W Chang, E Hontz, D. N. Congreve, L Shi, P. D. Reusswig, B Modtland, M. E. Bahlke, C. K. Lee, A. P. Willard, V Bulovi´c, T Van Voorhis, and M. A. Baldo. Nanoscale transport of charge-transfer states in organic donoracceptor blends. Nature Materials, 14(11):1130 -- 1134, 2015. [13] Dirk Veldman, Stefan C J Meskers, and Ren´e A J Janssen. The energy of charge-transfer states in electron donor- acceptor blends: insight into the energy losses in organic solar cells. Advanced Functional Materials, 19(12):1939 -- 1948, 2009. [14] William Shockley and Hans J. Queisser. Detailed balance limit of efficiency of p-n junction solar cells. Journal of [15] S Oviedo-Casado, Antonio Urbina, J Prior. Magnetic field enhancement of organic photovoltaic solar cells. Scientific Applied Physics, 32(3):510 -- 519, 1961. Reports, 7, article 4297, 2017. [16] Rodrigo Noriega, Jonathan Rivnay, Koen Vandewal, Felix P V Koch, Natalie Stingelin, Paul Smith, Michael F Toney, and Alberto Salleo. A general relationship between disorder, aggregation and charge transport in conjugated polymers. Nature Materials, 12(11):1038 -- 1044, 2013. [17] Klaus Schulten and Peter G. Wolynes. Semiclassical description of electron spin motion in radicals including the effect of electron hopping. The Journal of Chemical Physics, 68(7):3292, 1978. [18] E. L. Frankevich, A. A. Lymarev, and I. A. Sokolik. CT-excitons and magnetic field effect in polydiacetylene crystals. Chemical Physics, 162(1):1 -- 6, 1992. 11092, 2009. [19] Adam E. Cohen. Nanomagnetic control of intersystem crossing. Journal of Physical Chemistry A, 113(41):11084 -- [20] Eric Hontz, Wendi Chang, Daniel N. Congreve, Vladimir Bulovi´c, Marc A. Baldo, and Troy Van Voorhis. The Role of Electron-Hole Separation in Thermally Activated Delayed Fluorescence in Donor-Acceptor Blends. Journal of Physical Chemistry C, 119(45):25591 -- 25597, 2015. [21] Xugang Guo, Nanjia Zhou, Sylvia J. Lou, Jeremy Smith, Daniel B. Tice, Jonathan W. Hennek, Roco Ortiz Ponce, Juan T. Lopez Navarrete, Shuyou Li, Joseph Strzalka, Lin X. Chen, Robert P. H. Chang, Antonio Fachetti, Tobin J. Marks. Polymer solar cells with enhanced fill factors. Nature Photonics, 7, 825833, 2013. [22] Yifei Wang, Kevser Sahin-Tiras, Nicholas J. Harmon, Markus Wohlgenannt, and Michael E. Flatt´e. Immense mag- netic response of exciplex light emission due to correlated spin-charge dynamics. Phys. Rev. X, 6:011011, Feb 2016.
1905.00584	2	1905	2019-10-04T18:21:29	Demonstration of Sputtering Atomic Layer Augmented Deposition (SALAD): Aluminum Oxide-Copper Dielectric-Metal Nanocomposite	[ "physics.app-ph" ]	Designing a thin film structure often begins with choosing a film deposition method that employs a specific primary process by which chemical species are formed and transported. In other words, a film deposition system in which two complementary deposition methods are hybridized should lead to new ways of designing thin film structures with structural complexity tailored at a level that has never been envisioned.This premise inspired us to uniquely combine atomic layer deposition (ALD) and magnetron sputtering (SPU) within a single deposition chamber; the combined film deposition system is referred to as sputtering atomic layer augmented deposition (SALAD). SALAD allows us to take full advantages of both ALDs precise and accurate precursor delivery and SPUs versatility in choices of chemical elements. A SALAD system was designed based on knowledge obtained by computational fluid dynamics with the goal of conceiving a film deposition system that satisfied deposition conditions distinctive for both ALD and SPU, and a prototype SALAD system was assembled by employing off-the-shelf vacuum components.A s a demonstration, the SALAD system was utilized to deposit a unique nanocomposite made of a stack of aluminum oxide thin films by ALD and copper thin films by SPU, AlOx-Cu dielectric-metal (ACDM) nanocomposite thin films, on Si and glass substrates. Spectroscopic reflectivity collected on ACDM nanocomposite thin films shows unique dispersion features to which conventional effective medium theories used for describing optical properties of composites made of a dielectric host that contains metallic inclusions do not seem to simply apply.	physics.app-ph	physics	Demonstration of Sputtering Atomic Layer Augmented Deposition (SALAD): Aluminum Oxide-Copper Dielectric-Metal Nanocomposite Thin Films Brian Giraldo1, David M. Fryauf1, Brandon Cheney2, and Nobuhiko P. Kobayashi1 1. Nanostructured Energy Conversion Technology and Research (NECTAR), Department of Electrical and Computer Engineering, Baskin School of Engineering, University of California Santa Cruz, California 95064, United States 2. Department of Earth and Planetary Sciences, University of California Santa Cruz, California 95064, United States Keywords: magnetron, sputtering, atomic layer deposition, ALD, copper, aluminum oxide, optical properties, reflectivity, ellipsometry Abstract: Designing a thin film structure often begins with choosing a film deposition method that employs a specific primary process by which chemical species are formed and transported. In other words, a film deposition system in which two complementary deposition methods are hybridized should lead to new ways of designing thin film structures with structural complexity tailored at a level that has never been envisioned. This premise inspired us to uniquely combine atomic layer deposition (ALD) and magnetron sputtering (SPU) within a single deposition chamber; the combined film deposition system is referred to as sputtering atomic layer augmented deposition (SALAD). SALAD allows us to take full advantages of both ALD's precise and accurate precursor delivery and SPU's versatility in choices of chemical elements. A SALAD system was designed based on knowledge obtained by computational fluid dynamics with the goal of conceiving a film deposition system that satisfied deposition conditions distinctive for both ALD and SPU, and a prototype SALAD system was assembled by employing off-the-shelf vacuum components. As a demonstration, the SALAD system was utilized to deposit a unique nanocomposite made of a stack of aluminum oxide thin films by ALD and copper thin films by SPU -- AlOx-Cu dielectric-metal (ACDM) nanocomposite thin films -- on Si and glass substrates. Spectroscopic reflectivity collected on ACDM nanocomposite thin films shows unique dispersion features to which conventional effective medium theories used for describing optical properties of composites made of a dielectric host that contains metallic inclusions do not seem to simply apply. 1. Introduction Searching for answers to questions at the boundary between different disciplines (e.g., magnetism and superconductivity) has repeatedly yielded discoveries of exotic phenomena related to quantum nature of materials. In such exploratory searches, the current approaches focus on uncovering hidden phenomena emerging from materials prepared by rather conventional synthesis techniques, in that, choices of constituent atomic elements set the level of distinctiveness of materials. However, in our view, engineering of synthesis techniques is overlooked, and we strongly believe that engineering breakthroughs in synthesis techniques will lead to materials possessing structural complexities that have never been envisioned, ultimately benefiting the science of materials -- science enabled by engineering. For instance, sputtering (SPU) and atomic layer deposition (ALD) -- two prevalent technologies for the deposition of thin films -- have critical shortcomings. While SPU is capable of depositing thin films using flexible choices from a variety of chemical elements mixed at various compositions, it largely lacks accurate and precise control on delivering consistent amounts of chemical elements. In contrast, ALD allows deposition of thin films with accurate and precise control on thickness; however, it generally lacks explicit tunability of chemical composition. These shortcomings become noticeable when a thin film requires tight thickness control in the range of ~0.1 nm and chemical composition that needs to be varied continuously. In this paper, a new concept of hybrid thin film deposition system, based on SPU and ALD and referred to as Sputtering Atomic Layer Augmented Deposition (SALAD), is demonstrated. SALAD brings features not found in conventional stand-alone SPU or ALD, or in easily-imaginable cluster tools built upon SPU and ALD1, allowing SPU and ALD to operate 1 in a single environment in establishing, for instance, quasi-synchronous processes or sequential processes of the two deposition methods generally recognized as mutually incompatible from one another. SALAD allows us to perform "in-situ" tuning of surface composition at monolayer level without substantially altering a given or target thickness, which cannot be done by stand-alone SPU or ALD, or even by a cluster tool if the required total thickness is large (e.g., a few micrometers). A thin film structure that consists of aluminum oxide (AlOx) and copper (Cu) is presented as a case exemplifying unique capabilities of SALAD. Two specific materials, AlOx and Cu, were chosen to illustrate distinctive capabilities of SALAD by emphasizing that SALAD enables the combination of various thin film materials generally difficult to be prepared by a single deposition method of either SPU or ALD. 2. Design of SALAD to occur when Shown schematically in Fig. 1 is a cross sectional view of a SALAD system revealing its details with various critical dimensions. As shown in Fig. 1, the system consists of the following three main components: (1) SPU gun, (2) fast-acting spatial divider (FASD), and (3) precursor inlet/outlet for ALD. The three components are integrated on a single vacuum chamber with dimensions of the height hc and the diameter wc for a cylindrical chamber. The FASD plays a key role in uniquely establishing a deposition environment suitable for ALD within a deposition chamber optimized for SPU. The FASD also minimizes cross-contamination expected two deposition modes, ALD and SPU, run sequentially or concurrently. The FASD is inserted into or retracted from the deposition chamber within a time period of 1 s, without breaking the vacuum, to alternate two deposition environments -- one for ALD and the other for SPU. A SALAD system was designed by carrying out 2D computational fluid dynamics combined with to optimize, for instance, the width w, or diameter in case of a circular FASD, of the FASD in relation to the chamber width wc and the SPU throw distance h. Table 1 summarizes several key parameters used in the CFD-FEA. Since ALD and SPU co-exist in a single deposition chamber, it is critical to ensure that inserting the FASD creates a deposition environment suitable for ALD while h is optimized for SPU; with the FASD inserted in the chamber, a laminar flow that contains ALD precursors over the substrate is established and maintained even though h is optimized for SPU. In Fig. 1, dx and dy represent two orthogonal directions along which distance from their respective origin is measured; in other words, dx = 30 mm denotes the point located 30 mm from the left wall of the vacuum chamber while the FASD is located at dy = 30 mm illustrates that the FASD is placed 30 mm above the substrate. For the CFD-FEA design, a rather small SPU gun was used (wgun = 26 mm), thus, it was important to ensure that optimum uniformity in the spatial density of SPU species (spu) coming from the SPU gun was achieved across the substrate. Fig. 1: A prototype SALAD system used for the demonstration is schematically shown, revealing its details and basic concept. The system is equipped with a sputtering cathode "SPU gun" and a fast- acting spatial divider "FASD". A substrate is placed underneath the FASD. "Inlet" and "Outlet" direct carrier gases and precursors used in ALD. Several dimensional parameters used in the CFD- FEA analysis are also denoted. finite-element-analysis (CFD-FEA) Table 1: Summary of the key parameters evaluated in the CFD-FEA analysis. 2 Fig. 2(a) and (b) illustrate two maps showing the distribution of the normalized spu for two different h, (a) 10 mm and (b) 80 mm, indicating that h significantly influences the distribution of spu across the substrate, which is well known in designing SPU systems. Fig. 2(c) shows the dependence of nonuniformity defined as follows: (spu-max -- spu-min) / spu- min where spu-max and spu-min are the maximum and the minimum spu found across the substrate, clearly suggesting that h needs to be at least 60 mm to minimize non-uniformity, although h = 10 mm is expected to be more appropriate for ALD to maintain a laminar flow between the inlet and outlet. Fig. 3 shows the dependence of the distribution of the normalized density of gas molecules (ALD) used in ALD on w of the FASD. Gas molecules (i.e., ALD precursors) are injected into the chamber through the inlet and evacuated via the outlet assuming no reaction occurs over the substrate for three different w: (a) 7 mm, (b) 35 mm, and (c) 63 mm. The FASD is located at 10 mm above the substrate for all the three cases, indicating that a laminar flow is well established in Fig. 3(c). The establishment of a laminar flow is clearly seen in Fig. 4; it shows the dependence of the normalized ALD on the distance from the substrate along the dy direction, as indicated in Fig. 3(a), originated at the center of the substrate. As indicated by the red arrow denoted by "Development of a laminar flow" in Fig. 4, a laminar flow develops progressively as w increases and w needs to be at least 90% of the chamber width wc to obtain substantial contrast in the normalized ALD below and above the FASD for specific configurations used in the CFD-FEA analysis. Once the establishment of a laminar flow by inserting the FASD is visualized, it is worth evaluating the extent to which the FASD blocks SPU species moving towards the substrate. During a rather long ALD cycle (e.g., over 60 s), the SPU gun can be turned off in order to minimize the cross-contamination between SPU and ALD, however; for a short (e.g., ~3 s) ALD period, the SPU gun may need to remain turned on. Fig. 5 shows three maps displaying the distribution of normalized spu in (a) the case in which no FASD is inserted and in (b) and (c) in which the FASD is placed at two different locations, 10 mm and 30 mm, from the substrate along the dy direction, respectively. While, without the FASD being inserted, SPU species generated by the SPU gun directly reach the substrate as seen in (a), the insertion of the FASD blocks the SPU species from directly reaching the substrate as indicated in both (b) and (c). However, a substantial portion of SPU species leaking through the gap between the FASD and the wall of the vacuum chamber reach the substrate as evident in comparing Fig. 5(b) and (c). Species reaching the substrate in this fashion also depends on the position of the FASD along the dy direction as further illustrated in Fig. 6. Fig. 6(a) and (b) show normalized spu plotted as a function of distance along dx across the bottom of the vacuum chamber and along dx from the substrate as indicated in Fig. 5(c), respectively, indicating that the FASD with w = 80 mm placed at 10 mm from the substrate (i.e., distance along dy = 10 mm) would effectively block Fig. 2: (a)(b) Distribution maps of normalized spu in the chamber with two different h, the uniformity of spu, as well known, improves as h increases. (c) Non- uniformity of spu across the substrate plotted as a function of h, indicating h needs to be at least 60 mm to minimize the non-uniformity. indicating 4: the Fig. The variation in ALD along dy direction denoted in Fig. 3(a), indicating the progressive development of a laminar flow as w increases. Fig. 3: The gas flow pattern underneath FASD is shown for various w: (a) w = 7 mm, (b) w = 35 mm, and (c) w = 63 mm, respectively. 3 Fig. 5: Three maps display the distribution of spu in SPU for (a) No FASD and two different locations of the FASD that blocks the line-of-sight path of SPU species going towards the substrate as shown in (b) and (c). In (b) and (c), the FASD is located at 10 mm and 30 mm from the substrate along the dy direction, respectively. SPU species while it creates a laminar flow. When the FASD is raised to 30 mm (i.e., distance along dy = 30 mm), SPU species appears to leak through the gap between the FASD and the wall of the vacuum chamber and reach the substrate, which makes the situation as unacceptable as that without the FASD (i.e., No FASD). Although the CFD-FEA analysis was done on a SALAD system that accommodated a substrate with wsub = 20 mm as listed in Table 1, all the design parameters obtained in the analysis were scaled appropriately to assemble a SALAD system required for a series of experiments described below. The SALAD system used in the experiment was equipped with a SPU gun that accommodated a sputtering target with a diameter of 50 mm and was able to hold a substrate up to 100 mm in diameter. 3. Experimental Fig. 6: Fig. 6(a) and (b) show normalized spu plotted along the two different directions dx across the bottom of the chamber and dy above the substrate as indicated in Fig. 5(c), respectively. A series of deposition runs were carried out to distinctively demonstrate the capability of the SALAD system intended for unifying ALD and SPU in a single vacuum chamber. For the demonstration, a thin film structure made of aluminum oxide (AlOx), dielectric material, and copper (Cu), metallic material, was used. AlOx and Cu were deposited by ALD and SPU, being referred to as "ALD-AlOx" and "SPU-Cu", respectively. AlOx was chosen because of the level of its maturity as a dielectric material routinely deposited by ALD while Cu was chosen because of its free-electron density much higher than that of silver2,3 and gold4,5,6,7 -- the two metallic materials commonly used as inclusions embedded in dielectric materials in order to explicitly tailor their optical properties as a whole. Aluminum (Al) could be an alternative to Cu; however, Al was avoided as it was expected to have formed AlOx when Al was exposed to the AlOx ALD process which used water as an oxidizer. In other words, AlOx resulting from the oxidation of Al deposited by SPU would not be easily distinguished from ALD-AlOx. Cu chosen for the experiment also forms a variety of oxides including Cu2O, CuO, and Cu4O3 that would significantly modify dielectric properties of ALD-AlOx by behaving as a band insulator8, a charge transfer insulator9, and a mixed-valence compound consisting of Cu2O and CuO10, respectively. However, these oxides can be clearly distinguished from ALD-AlOx using analytical perspectives and are suitable for studying intermediate phases made by alternatively depositing ALD-AlOx and SPU-Cu being referred to as AlOx-Cu dielectric-metal (ACDM) nanocomposite thin films. Table 2 summarizes the deposition parameters set for ALD and SPU in depositing the ACDM nanocomposite thin films. These specific deposition parameters were explicitly obtained through separately calibrating ALD and SPU processes in the SALAD system with the goal of achieving an appropriate deposition rate required for designing a single SPU step that delivered a specific amount of Cu with a corresponding equivalent-thickness in reference to the self-limited deposition rate indefinitely provided by the ALD process. In other words, the ALD process offered a self-limited deposition rate ALD = 0.13 nm/cycle for ALD-AlOx; thus the deposition rate of the SPU process SPU for SPU-Cu was adjusted to be 0.04 nm/s so that the total amount of SPU-Cu was directly controlled by simply specifying a unique duration (tSPU) of 4 Table 2: Summary of the ALD and SPU conditions used for depositing the AlOx-Cu nanocomposite thin film structures. in Fig. 7: Time-progressive change the background pressure in the SALAD system shows three SALAD cycles recorded during the deposition of one of the AlOx-Cu nanocomposite thin film structures used in the study. Each SALAD cycle comprises: a TMA pulse denoted by the green arrows, a purge for TMA, a H2O pulse denoted by the blue arrows, a purge for H2O, and a SPU duration indicated by the red horizontal double-head arrows. 4000 spectroscopic deposited SPU-Cu. A FilmTek reflectometry/ellipsometry (SRE) equipped with a UV-Visible light source was used to obtain ALD and SPU. A SPU step with tSPU was inserted between two ALD cycles, which defined a single SALAD deposition cycle as illustrated in Fig. 7. In Fig. 7, the change in the background pressure in the SALAD system is plotted as a function of time revealing three SALAD cycles (note: only three out of the total 300 SALAD cycles are shown) recorded during the deposition of an ACDM nanocomposite thin film with tSPU = 3.0 s. A single SALAD cycle comprises: a Al(CH3)3 pulse indicated by a green vertical arrow, a purge for Al(CH3)3, a H2O pulse marked by a red vertical arrow, a purge for H2O, and a subsequent SPU step with a specific tSPU indicated by a red double-headed horizontal arrow. In Fig. 7, a unique feature in SALAD cycles is further revealed as follows: after the completion of a purge for H2O, an intermediate evacuation step, observed as a sudden drop in the pressure indicated by a black vertical arrow, occurs before initiating a SPU step by temporarily stopping the flow of argon carrier gas for ALD-AlOx that creates a laminar flow along the direction perpendicular to the surface normal of the substrate. The laminar flow established for ALD was found to create an impenetrable barrier that stopped Cu species generated in a SPU step from reaching the substrate. In this inaugural experiment, a SPU step was inserted after a full ALD cycle was completed due to its simplicity that avoided an interruption of an ALD cycle. Needless to say, a SPU step can be inserted at any point (e.g., after a Al(CH3)3 pulse, after a purge for Al(CH3)3, after a H2O pulse, or after a purge for H2O) within a single ALD cycle. A series of ACDM nanocomposite thin film samples was deposited on glass substrates (Corning 2497) by repeating a SALAD cycle comprising of an ALD step followed by a SPU step with a specific tSPU varied in the rage from 1.3 s to 7 s corresponding, given SPU = 0.04 nm/s, to the amount of SPU- Cu equivalent to 0.05-0.28 nm thickness. The total number of SALAD cycles was fixed to 300 for all the samples. All the glass substrates were used as they were directly pulled from their original container, and no pre-deposition cleaning procedure was done. The following assumption is made to quantitatively distinguish the samples prepared with varied tSPU: a single SALAD cycle yields a layer of ALD-AlOx with thickness of 0.13 nm followed by a layer of SPU-Cu with a nominal thickness of 0.04 nm/s x tSPU with an assumption that no intermixing of the two occurs. Although this assumption needs to be verified by carrying out detailed structural and chemical analysis of the samples, for the purpose of this paper, the ACDM nanocomposite thin film samples prepared with varied tSPU are contemplated as (AlOx)n(Cu)m where n/m = (ALD x 1 ALD cycle) / (SPU x tSPU) and n + m = 1. By solving for n and m for a given tSPU, the prepared seven variations of ACDM nanocomposite thin film samples are identified by using m as listed in Table 3. In addition to the ACDM nanocomposite thin film samples, a 120 nm thick SPU-Cu (i.e., no ALD- AlOx was deposited) sample and a 30 nm thick ALD-AlOx (i.e., no SPU- Cu was deposited) sample were prepared as two reference samples. 4. Results and Discussion 5 Table 3: n and m calculated for various tSPU used in the experiment providing seven variations of ACDM nanocomposite thin film samples identified by m. Energy dispersion spectroscopy (EDS) analysis of the ACDM nanocomposite thin film samples was carried out in a scanning electron microscope to obtain effective atomic composition of constituent chemical elements including Cu, oxygen (O), Al, and carbon (C) expected to be present in the samples. Specific conditions used in the EDS analysis are as follows: acceleration voltage of 15 kV and electron current of 1.6 nA. Shown in Fig. 8(a) is weight percent (wt%) of Cu (solid red line). Wt% of O (blue solid line), Al (orange solid line), and C (green solid line) are shown in (b) plotted as a function of m. The wt% of each of Cu, O, and Al -- referred to as [Cu], [O], and [Al] -- changes linearly as m changes while wt% of C -- referred to as [C] -- remains almost constant or decreases slightly as m increases. [Cu] monotonically increases while [Al] and [O] decreases as m increases. The ratio of [Al] to [O] remains in the rage of 1.3-1.5 for m = 0.29-0.68, reflecting the fact that x of ALD-AlOx is comparable or slightly smaller than 1.5 expected for stoichiometric Al2O3. Since ALD-AlOx cycles were fixed to 300 for all the samples, the increase in [C] in (a) and corresponding decreases in [Al] and [O] in (b) are a result of increase in [C] as m increases. The trend of [C] slightly decreasing from 8 to 5 wt% as m increases would be viewed as being consistent with the decrease in [Al] originated to ALD-AlOx made by using a precursor, trimethylaluminum, that contains C, as well as the fact that Al acting as a getter of such impurities as C present in the deposition environment. The EDS analysis clearly demonstrate SALAD's capability of explicitly controlling [C] by SPU embedded in the ALD-AlOx host. The linear change of [Cu], [O], and [Al] as m changes indicates that SPU and ALD processes done sequentially were independent. In other words, the two processes had minimum influence on one another. Fig. 8: Weight percent (wt%) of Cu in (a) and O (blue solid line), Al (orange solid line), and C (green solid line) in (b). The wt% of these chemical elements were obtained by using EDS spectra collected on the AlOx-Cu nanocomposite thin film structures prepared with various m. The wt% is plotted as a function of m. In this demonstration of SALAD, our focus was on optical properties -- a mirror image of dielectric properties -- of the ACDM nanocomposite thin film samples, which was expected to be modified by the inclusion of SPU- Cu in the samples. Measurements of reflectance spectra (R spectra) were carried out to elucidate unique optical properties that would emerge from the samples characterized by their complexities of the way two materials, ALD-AlOx and SPU-Cu, are sequentially deposited by SALAD. Fig. 9(a)(b) shows R spectra collected from the seven samples with various m in the range of 0.29-0.68 as previously defined on Table 3. The R spectra are presented in two different ranges of wavelength, 200-600 nm in (a) and 500-2500 nm in (b), to highlight the details of specific features appearing in these two spectral ranges. R spectra obtained from the two reference samples -- a 120 nm thick SPU-Cu (m = 1) and a 30 nm thick ALD-AlOx (m = 0) -- are also displayed in Fig. 9(a)(b), showing R dispersions generally expected for bulk forms of pure Cu and amorphous Al2O3. Fig. 9(a) clearly indicates that there exists a progressive development of R in the range of 200-600 nm as m increases. The wavelength range of 200-600 nm represents a range of wavelengths shorter than ~560 nm associated with the plasma reflection edge of bulk Cu. In other words, the features that appear in the R spectra in Fig. 9(a) are expected to be dominated by the inter-band transitions rather than contributions associated with free-electrons. The absorption coefficient of bulk Cu measured in the wavelength range of 200-600 nm was found to exhibit features originated to three inter-band transitions at various symmetry points L2'-L1, X5-X4', and L3-Fermi surface11; thus the undulating features that appear in the wavelength range of 200-600 nm in Fig. 9(a) are most likely associated with the progressive formation of well-defined electronic bands of Cu, which can be further viewed as the increase in the total volume of Cu as m increases by a fraction, highlighting precise and accurate control on delivering a specific amount of SPU-Cu physically separated by yet another precise and accurate control of an atomic layer of ALD- AlOx, being made possible by SALAD. The statement "the progressive formation of well-defined electronic bands of Cu" needs to be further clarified and validated by performing such complementary characterization 6 as Raman scattering spectroscopy and x-ray photoelectron spectroscopy to elaborate a physical and chemical picture of multiple Cu -- continuous or non-continuous -- layers, rather than bulk Cu, effectively separated by a single ALD-AlOx layer in the framework of the formation of electronic bands. On the R spectra in the range of wavelength from 600 to 2500 nm (i.e., the range of wavelength longer than the plasma reflection edge of bulk Cu) in Fig. 9(b), one would expect the dispersion of R collected from the ACDM nanocomposite thin film samples to reflect characteristics of R of the two constituent materials -- Cu and AlOx -- skillfully blended. Optical properties of bulk metals including Cu in this spectral range are often described by using the Drude-Lorentz theory in which free- electrons (i.e. electrons experiencing no restoring force) play the major role in determining complex dielectric constants (i.e., complex optical constants) relevant to the R spectra. In the description of the Drude-Lorenz theory, the equation of motion for the displacement of a free-electron includes the frictional damping force F that accounts for the retarding action of the surrounding medium. In such metals as Cu, F is often set to zero to explain the R spectrum of SPU-Cu (m = 1) being close to 100%. The series of R spectra for the different m in Fig. 9(b) shows no distinct tendency in the way by which the spectra change as m changes in contrast to the definitive changes in the R spectra in Fig. 9(a). Interestingly, the sharp increase in the R spectrum of SPU-Cu (m = 1) seen at ~560 nm seems to emerge in none of the R spectra collected from the samples with m = 0.29-0.68. In a qualitative perspective, F appears to be a factor that should not be set to zero for the samples with m = 0.29- 0.68, suggesting the ACDM nanocomposite thin film samples presumably because of the presence of ALD-AlOx layers even for m = 0.69 for which the total weight percent of Cu is ~80% as seen in Fig. 8(a). It is also conceivable that the inter-band transition discussed earlier for the range of wavelength shorter than the plasma reflection edge of 560 nm somehow extends well above the plasma reflection edge, suggesting that ALD-AlOx layers may have given rise to additional inter-band transitions associated with SPU-Cu in the ACDM nanocomposite thin film samples because of, for instance, the alternations on the density of states in 3d and 4s bands of Cu. is much that F larger in Fig. 9: Reflectance R spectra of the ACDM nanocomposite thin film structures prepared with various m along with those collected from SPU Cu (m = 1) and ALD AlOx (m = 0). The reflectance spectra are plotted over two different ranges: (a) 200-600 nm and (b) 500-2500 nm to highlight features distinctive in their respective ranges. Contributions to the unique R spectrum in Fig. 9(a) and (b) at the macroscopic-level rather than the microscopic-level should not be ruled out -- for instance, optical interference occurring in a structure in which two types of layers having refractive index significantly different one another. However, since the optical thickness of each of ALD-AlOx and SPU-Cu layers in the ACDM nanocomposite thin film samples is much smaller than a quarter of the wavelengths of the light used in measuring the R spectra, destructive and constructive interferences by which conventional distributed Bragg reflectors and band-pass filters operate are not relevant. The skin effect often used in discussing AC conductivity in metals would be another contributor to the peculiar R spectrum illustrated in Fig. 9(a) and (b). For instance, the skin depth of Cu is estimated to be 2.7-5.9 nm using the conventional formula12 for electromagnetic waves with wavelengths 500-2500 nm relevant to our experiment, respectively, which are up to two orders of magnitude larger than the thickness of a single SPU-Cu layer in the ACDM nanocomposite thin film samples, indicating that exploiting the concept of the skin effect is invalid. Another perspective, conventional nanocomposites AyB1- y made of a dielectric host made of element A that contains metal inclusions made of element B often show complex optical characteristics when the dielectric-metal compositional ratio y is in the range of 0.2 - 0.813, which is often described within the framework of effective medium theories including the Maxwell Garnett approximation intended to undertake the interaction between electromagnetic wave and such mediums as 7 a colloidal solution of metallic particles suspended in a dielectric host 14 . In the Maxwell Garnett approximation, a complex medium -- effective medium -- is assumed to exhibit effective permittivity approximated in terms of the permittivity and volume fractions of the independent constituents of the complex medium. This approximation, however, is clearly not valid in the ACDM nanocomposite thin film samples because of the presence of a structural periodicity (i.e., ALD-AlOx and SPU-Cu layers being stuck on top of each other) while the random distribution of metallic particles within a dielectric host is assumed in the Maxwell Garnett approximation. Further assessment is currently in progress to understand the unique R dispersion considering the structural periodicity introduced by SALAD. 5. Conclusion Our vision of creating a unique deposition system capable of two deposition techniques deemed incompatible each other has been realized by demonstrating SALAD that has yielded the ACDM nanocomposite thin film samples made of ALD-AlOx and SPU-Cu with optical properties that could not be simply assessed by interpolating individual optical properties of its constituent materials. The key to engineering SALAD was the modeling of the gas flow in the chamber; our CFD-FEA allowed us to optimize chamber design parameters such as FASD width, chamber width, and SPU throw distance. Our demonstration would open a new frontier of materials science exploration by providing unique thin film structures made of a wide range of materials, which would not be easily realized by employing a thin film deposition system based on a single deposition method. We are confident that the extreme level of control over deposition rate and the enhanced ability to deliver a variety of chemical elements is producing materials with structural complexities that lead to novel physical phenomena. 1 A. J. Elliot, G. A. Malek, R. Lu, S. Han, H. Yu, S. Zhao, J. Z. Wu, "Integrating atomic layer deposition and ultra-high vacuum physical vapor deposition for in situ fabrication of tunnel junctions," Review of Scientific Instruments 85, 073904 (2014). 2 R. Ravindran, K. Gangopadhyay, S. Gangopadhyay, "Permittivity enhancement of aluminum oxide thin films with the addition of silver nanoparticles," Applied Physics Letters 89, 263511 (2006). 3 M. Gorbunova, L. Lemkina, I. I. Lebedeva, D. Kisel'kov, L. Chekanova, "Synthesis and potential applications of silver -- porous aluminium oxide nanocomposites as prospective antiseptics and bactericides," Journal of Materials Science: Materials in Medicine 28, 40 (2017). 4 L. Maya, W. R. Allen, A. L. Glover, J. C. Mabon, "Gold nanocomposites," Journal of Vacuum Science & Technology B 13, 361 (1995). 5 N. M. Figueiredo, N. J. M. Carvalho, A. Cavaleiro, "An XPS study of Au alloyed Al-O sputtered coatings," Applied Surface Science 257, 5793 (2011). 6 V. Lozovski, M. Razumova, "Influence of inclusion shape on light absorption in thin Au/Teflon nanocomposite films," Journal of the Optical Society of America B 33, 8 (2016). 7 N. M. Figueiredo, T. Kubart, J. A> Sanchez-Garcia, R. Escobar Galindo, A. Climent-Font, A. Cavaleiro, "Optical properties and refractive index sensitivity of reactive sputtered oxide coatings with embedded Au clusters," Journal of Applied Physics 115, 063512 (2014). 8 T. Kazimierczuk, D. Frohlich, S. Scheel, H. Stolz, M. Bayer, "Giant Rydberg excitions in the copper oxide Cu2O," Nature 514, 343 (2014). 9 J. Ghijsen, L. H. Tjeng, J. van Elp, H. Eskes, J.Westerink, G. A. Sawatzky, M. T. Czyzyk, "Electronic structure of Cu2O and CuO," Physical Review B38, 11322 (1988). 10 B. K. Meyer, A. Polity, D. Reppin, M. Becker, P. Hering, P. J. Klar, T. Sander, C. Reindl, J. Benz, M. Eickhoff, C. Heiliger, M. Heinemann, J. Blasing, A. Krost, S. Shokovets, C. Muller, C. Ronning, "Binary copper oxide semiconductors: From materials towards devices," Physica Status Solidi B249, 1487 (2012). 11 B. Segall, Phys. Rev. 125, 109 (1962). 12 L.-T. Hwang, I. Turlik, "A review of the skin effect as applied to thin film interconnections," IEEE Transactions on Components, Hybrids, and Manufacturing Technology 15, 43 (1992). 13 R. W. Cohen, G. D. Cody, M. D. Coutts, B. Abeles, "Optical properties of granular silver and gold films," Physical Review B8, 3689 (1973). 14 M. Y. Koledintseva, R. E. DuBroff, R. W. Schwarts, "A Maxwell Garnett model for dielectric mixtures containing conducting particles at optical frequencies," Progress In Electromagnetics Research, PIER 63, 112 (2006). 8
1808.00874	2	1808	2018-12-23T21:57:11	Spectral purification of microwave signals with disciplined dissipative Kerr solitons	[ "physics.app-ph" ]	Continuous-wave-driven Kerr nonlinear microresonators give rise to self-organization in terms of dissipative Kerr solitons, which constitute optical frequency combs that can be used to generate low-noise microwave signals. Here, by applying either amplitude or phase modulation to the driving laser we create an intracavity potential trap, to discipline the repetition rate of the solitons. We demonstrate that this effect gives rise to a novel spectral purification mechanism of the external microwave signal frequency, leading to reduced phase noise of the output signal. We experimentally observe that the microwave signal generated from disciplined solitons follows the external drive at long time scales, but exhibits an unexpected suppression of the fast timing jitter. Counter-intuitively, this filtering takes place for frequencies that are substantially lower than the cavity decay rate. As a result, while the long-time-scale stability of the Kerr frequency comb repetition rate is improved by more than 4 orders of magnitude as a result of locking to the external microwave signal, the soliton stream shows a reduction of the phase noise by 30 dB at offset frequencies above 10 kHz.	physics.app-ph	physics	Spectral Purification of Microwave Signals with Disciplined Dissipative Kerr Solitons Wenle Weng,1, ∗ Erwan Lucas,1, ∗ Grigory Lihachev,2, 3 Valery E. Lobanov,2 Hairun Guo,1 Michael L. Gorodetsky,2, 3 and Tobias J. Kippenberg1, † 1 ´Ecole Polytechnique F´ed´erale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland 2Russian Quantum Center, Skolkovo 143025, Russia 3Faculty of Physics, M.V. Lomonosov Moscow State University, 119991 Moscow, Russia (Dated: December 27, 2018) Continuous-wave-driven Kerr nonlinear microresonators give rise to self-organization in terms of dissipative Kerr solitons, which constitute optical frequency combs that can be used to generate low-noise microwave signals. Here, by applying either amplitude or phase modulation to the driving laser we create an intracavity potential trap to discipline the repetition rate of the solitons. We demonstrate that this effect gives rise to a novel spectral purification mechanism of the external microwave signal frequency, leading to reduced phase noise of the output signal. We experimentally observe that the microwave signal generated from disciplined solitons is injection-locked by the external drive at long time scales, but exhibits an unexpected suppression of the fast timing jitter. Counter-intuitively, this filtering takes place for frequencies that are substantially lower than the cavity decay rate. As a result, while the long-time-scale stability of the Kerr frequency comb's repetition rate is improved by more than 4 orders of magnitude, the purified microwave signal shows a reduction of the phase noise by 30 dB at offset frequencies above 10 kHz. Introduction. -- Low-noise microwave signals play a vi- tal role in a wide range of industrial and scientific applications, including telecommunication networks [1], radar/LIDAR systems [2] as well as in fundamental re- search such as long baseline interferometry [3] and tests of fundamental constants [4, 5]. Traditionally, the mi- crowave signals with the best spectral purity were pro- vided by cryogenic microwave oscillators [6, 7]. Owing to the advancement of mode-locked-laser frequency combs and optoelectronics, new photonic-based ways of gener- ating ultralow-noise microwaves have been proposed and demonstrated, such as optical frequency division [8 -- 10], electro-optical frequency division [11], or Brillouin lasing in microresonators [12, 13]. Recently, dissipative Kerr solitons (DKS) in optical mi- croresonators [14, 15] have been attracting surging inter- ests thanks to their self-organizing mechanism that re- sults from the double-balance between nonlinearity and anomalous dispersion, as well as between parametric gain and cavity loss. DKS offer high coherence, broad band- width and microwave-repetition rate frequency combs (also referred to as soliton microcombs [16]), and have been applied successfully to ultrafast ranging [17, 18], dual-comb spectroscopy [19 -- 21], calibrating astrophysi- cal spectrometer [22, 23], as well as optical frequency syn- thesis [24]. Like mode-locked-laser frequency combs, soli- ton microcombs can function as a frequency link between the microwave/radio-frequency (RF) domain and the op- tical domain [25, 26]. In particular, microcomb-based microwave oscillators hold great promise of providing a robust, portable and power-efficient way to synthesize pure microwave tones [27]. In contrast to microresonator- based approaches of generating microwave signals using Brillouin lasers, the frequency of the generated signal is mainly determined by the cavity free spectral range (FSR), rather than the host material property of the resonator, thus offering control over the microwave cen- ter frequency. However, this flexibility comes at a price: reaching a good longterm stability requires the ability to control the comb repetition rate (frep) and the carrier- envelope offset (fceo) and discipline them to optical refer- ences or RF clocks. To obtain such ability most previous efforts focused on using active feedback to correct ther- mal drifts and noises [28, 29] and utilizing sophisticated structure design for appropriate actuation [26, 30, 31]. In this work, we use DKS in a crystalline microres- onator to purify a 14.09 GHz microwave signal. The phase noise of the purified signal approaches -130 dBc/Hz at 10 kHz offset frequency, which is at the level achieved by the state-of-the-art microresonator-based optoelec- tronic oscillators and the previously reported best re- sults obtained with undisciplined DKS and narrowband RF filter [2, 27]. We adapt the microwave injection- locking technique that was previously used to stabilize modulation-instability (MI) combs [25, 32] to discipline the soliton stream by creating intracavity potential gra- dient that traps the solitons. This mechanism not only relies on linear cavity filtering, but exploits further the dynamics of DKS, and allows to reduce substantially the phase noise of an external microwave drive. Owing to the dynamical attractor of the soliton state, the stability of the disciplined solitons exhibits strong robustness against incoherent perturbations contained in the injected signals [33], thus efficiently dissipating noises in a coherent sys- tem. This self-purifying mechanism leads to the reduc- tion of the injected microwave phase noise, allowing the nonlinear cavity in the soliton state to act as a passive spectral purifier that can improve the performance of an external off-the-shelf electronic oscillators. As depicted in Fig. 1, the disciplined-DKS-based microwave purifier 2 and (b). From a frequency domain perspective, the mod- ulation frequency defines frep through parametric four- wave-mixing. In the time domain, a modulated CW field traps solitons and disciplines frep correspondingly. In this proof-of-principle experiment we use a synthesizer to drive the AM/PM modulator but the input microwave signal could be derived from a frequency-multiplied clock oscillator or a voltage-controlled oscillator (VCO). The modulation frequency fmod is swept around the free- running frep (∼ 14.09 GHz) and we observe that frep is injection-locked by the input microwave signal. Fig. 3 (c) shows the evolution of the microwave spectrum of the DKS as we slowly swept the AM fmod. When the dif- ference between fmod and the free-running frep is larger than ∼ 400 Hz, multiple spectral components including fmod (the strongest), frep (the second strongest) and mul- tiple harmonics are observed in the spectra, indicating an absence of injection locking. As fmod is approach- ing the free-running frep, the spectrum displays typical frequency-pulling effect as frep is pulled towards fmod [37]. When the difference between fmod and free-running frep is less than ∼ 300 Hz all the spectral components merge into a major one, indicating that the frep is syn- chronized to fmod, i.e. the soliton stream is locked to the external drive. We measured the frequency insta- bilities of the injected-locked frep against fmod, which is also presented in Fig. 2 (f). The Allan deviation shows that at time scales of > 0.1 s the fluctuations of frep has been suppressed significantly -- up to more than 4 orders of magnitude at averaging time of 1000 s, indicating that the disciplined DKS tightly follow the injected microwave frequency. We acquire the locking range from the evolution of the RF spectrum, and repeat the measurement with varied modulation strength. As shown in Fig. 3 (e), with the normalized amplitude of the modulation sideband below 0.07, the locking range rises monotonically with almost perfect linearity as the modulation strength increases. With stronger modulation the slope of the locking range scaling increases, which is attributed to the appearance of higher-order modulation sidebands that increase the gradient of the potential and trap the solitons more ef- fectively [38 -- 41]. For the same reason, we observe that the locking range increases by nearly a factor of 2 when we measure the locking range with fmod around 2 × frep (∼ 28.18 GHz). Spectral purification effect. -- To characterize the spec- tral purity at frep, the out-coupled soliton stream is fil- tered by fiber Bragg grating filters (FBG) to suppress the pump light and then amplified by an EDFA and subse- quently attenuated to ∼ 5 mW before being registered by a fast photodetector. We use a phase noise analyzer to measure the phase noise of the 14.09 GHz signal output by the photodetector. Fig. 4 presents the single-sideband (SSB) phase noise level when PM injection locking was performed. One should note that very similar results FIG. 1. The concept of a microcomb-based microwave spec- tral purifier. A commercially-available electronic microwave oscillator (Rohde&Schwarz SMB100A) is used to modulate the pump laser, leading to the injection locking of solitons, thus providing a long-term frequency reference to the soliton repetition rate. The generated microwave exhibits a reduced phase noise level due to the nonlinear soliton dynamics, lead- ing to noise reduction of the microwave signal for Fourier fre- quencies far away from the carrier. constitutes in itself a frep-stabilized frequency comb and a spectrally pure microwave generator into a single de- vice. A 1555-nm laser Experiment. -- The experimental setup is shown in Fig. 2 (a). is amplified by an Erbium-doped fiber amplifier (EDFA) and 200 mW op- tical power is coupled into a z-cut magnesium fluoride (MgF2) whispering-gallery-mode resonator with a FSR of 14.09 GHz via a tapered fiber. A single-soliton-state DKS comb is generated by scanning the laser over a res- onance with a loaded quality-factor (Q) of 1.3 × 109 to reach the step-like range where solitons are formed [14]. To stabilize the effective laser detuning with respect to the cavity resonance, we apply phase modulation to the laser with an electro-optic modulator (EOM) to generate Pound-Drever-Hall (PDH) error signals. The laser fre- quency is locked to the high-frequency PDH sideband by setting the lock point of the servo to the center of the sideband resonance which is indicated in Fig. 2 (e). The frequency of the laser is then compared with a tooth of a stabilized fiber-laser-based comb, and the frequency dif- ference is stabilized at 20 MHz through a slow thermal ac- tuation on the cavity by active control of the pump power with an acousto-optic modulator (AOM). As illustrated in Fig. 2 (c), with the two servos this "pre-stabilization" scheme stabilizes both the pump laser frequency and the pump-cavity detuning. As a result, the stability of frep is improved by up to 2 orders of magnitude at time scales of > 10 s (see Fig. 2 (f)), allowing the time-consuming mea- surement of phase noise via cross correlation to be carried out properly. One should note that the fiber-laser-based comb can be replaced with a laser stabilized by a refer- ence cavity [34] or an atomic vapor cell [35], and that with improved thermal isolation [27] or self-referenced stabilization [36] the entire setup can be more compact. The injection locking of the soliton repetition rate is implemented by applying amplitude modulation (AM) or phase modulation (PM) on the pump laser, at a fre- quency close to the FSR. Intuitive illustrations of how the injection locking works are presented in Fig. 3 (a) CW LaserA/PMoffset freq.phase noiseoffset freq.phase noise3 FIG. 2. (a) The experimental setup. (b) The MgF2 resonator used in the experiment. (c) Illustration of the PDH offset locking and the pre-stabilization scheme. (d) Optical spectrum of the soliton microcomb. (e) Generated comb power as the laser is scanned across the pumped resonance (upper) and the corresponding PDH error signal (lower). The red dot indicates the locking point. (f) Allan deviations of frep when the Kerr comb is pre-stabilized and DKS-disciplined respectively. We counted frep with a Π-type frequency counter that is referenced to the same frequency source (relative frequency instability < 1× 10−12 at 1 s averaging time) to which fmod is referenced. FIG. 3. (a) In frequency domain, the difference between the pump laser and the modulated sidebands sets the microcomb frep. (b) In time domain, the potential of the modulated CW field traps solitons, thus locking frep of the soliton train. (c) Evolution of the spectrum around 14.09 GHz. (d) Snapshots with different fmod detuned from free-running frep (indicated in the upper-right corners) showing the typical states of unlocked (-500 Hz and 500 Hz), quasi-locked (-300 Hz and 300 Hz) and injection-locked (0 Hz). (e) Locking ranges with varied AM strengths for modulation frequencies around frep (blue circles) and 2 × frep (red circles) respectively. were also observed with AM injection locking. At low offset frequencies between 1 -- 100 Hz the injection-locked DKS show improved noise level that is in agreement with the phase noise level of the input RF tone due to the bet- ter long-term frequency stability provided by the injected microwave signal, as confirmed by the Allan deviations. This result shows that the soliton stream is strictly dis- ciplined by the potential trap at low frequency ranges. Remarkably, at offset above 100 Hz the spectrum of the injection-locked frep mostly maintains the intrinsic high quality, which is several orders of magnitude lower than the input microwave in terms of phase noise level. We note that this purifying effect cannot be explained by the cavity filtering since the frequency range where the purification is observed is ∼ 3 orders of magnitude lower than the loaded cavity resonance bandwidth (∼ 150 kHz). At offset frequencies above 30 kHz a reduction of the in- put microwave phase noise level by 30 dB is achieved, showing the exceptional spectral purifying ability of the disciplined DKS. Simulation of soliton spectral purification. -- In order to study the mechanism of the observed spectral purifi- cation, we performed simulations of PM-to-PM transfer function based on the Lugiato-Lefever equation (LLE) frequency 2 mm CW LaserEDFAOSAESAEOMAOMPower ServoPDH ServoFiber Laserdetuning(PDH locking)~20 MHz(power servo)cavityresonance(cid:31)ber lasercomb linepumpPDH sidebandsabc-30-20-10Detuning [MHz]-0.100.1Error [V]020103000.51Power [a.u.]15201540156015801600Wavelength [nm]-60-40-200Power [dBm]edfCombNormalized Allan dev. at 14.09GHz10-510-410-310-210-110010110210310410510-1610-1510-1410-1310-1210-1110-1010-910-8Pre-stabilized onlySoliton disciplinedMicrowave sourceCounter limitsAveraging time [s]-2000 -1000 0 1000 2000RF frequency (centered on 14.093 GHz) [Hz]-500-400-300-200-1000100200300400500Modulation frequency offset [Hz]-120-100-80-60-40-200RF power [dBm]RF power [Arb.Units]RF frequency (centered on 14.093 GHz) [Hz]-2000 -100001000 2000-500 Hz-300 Hz 0 Hz 300 Hz 500 HztimeCW backgroundsolitontpumpsidebandsidebandffffffffabcdesolitonfrequency00.050.10.15Normalized amplitude of modulation sideband020040060080010001200Locking range [Hz]14.09 GHz mod.28.18 GHz mod.4 FIG. 4. Phase noise spectra of the soliton repetition rate with and without PM injection locking. The phase noise of the input microwave signal is also presented, showing that the injection locking reduces the noise level by nearly 40 dB for offsets at 100 kHz. The crosses and the dashed line show the noise floor of the phase noise analyzer. FIG. 5. Experimentally measured and numerically simulated phase noise transfer indices for the PM-to-PM noise transfer from injected microwave signal to frep of the DKS stream. The flat floors of the experimental data at frequencies above 100 kHz are due to the noise floor of the phase noise analyzer during the IQ measurement, which are indicated by the dash- dot lines. = 2 µ±1i e±iΩt) 2 (1) [42]. The model is similar to the one described in [43], which is expressed as: ∂t ∂ Aµ(t) (cid:16)− κ − igF(cid:2)A2A(cid:3) (cid:17) Aµ of DKS respectively (related via A(t) =(cid:80) µ + + i(2πδ) + iDint(µ) µ0 + δ(cid:48) κexsin(δ(cid:48) √ µ where Aµ and A are the spectral and temporal envelopes Aµe−iµD1t), κ is the cavity loss rate, g is the single photon induced Kerr frequency shift, κex is the coupling rate and sin2 denotes the power of the laser pumping the central mode, µ0/±1 is the Kronecker delta, and F[ ]µ represents the δ(cid:48) µ'th frequency component of the Fourier series. We in- clude third oder dispersion in Dint(µ). A pair of PM sidebands are included in the last term of the equation, where indicates the amplitude of the modulation side- bands, and Ω is the frequency difference between the FSR and the input microwave signal. Adapting the technique used in [44], we introduce phase modulation on the microwave signal with phase deviation of 0.1 radian and varied modulation frequen- cies from 200 Hz to 1 MHz. The phases of the purified microwave signal can be derived from the comb spectra with Ψ(t) = Arg † Aµ A µ−1 (2) µ where ωin is the frequency of the input microwave sig- nal. We use pump power of 200 mW and = 0.32 for the numerical simulation [45]. The results are presented in Fig. 5. The simulated transfer function follows a typical first-order lowpass filtering effect, showing a magnitude (cid:34) eiωint(cid:88) (cid:35) that is close to unity at low frequency (200 Hz). For higher offset frequencies the magnitude decreases with a slope of -20 dB/decade, reaching a minimum of ∼ −63 dB around 500 kHz, thus revealing a significant phase noise suppression in the soliton state. To verify the simulated results, we apply PM with varied phase deviation on the injected microwave signal and record the resulting phase deviation on the soliton repetition rate with an in-phase- and-quadrature (IQ) demodulator [45]. The experimen- tally measured transfer functions are plotted in the same figure. From the comparison we see that at low frequen- cies the experimental results and the simulation are in satisfactory agreement. However, at frequencies above ∼ 100 kHz the experimental curves show flat floors, which are attributed to the detection noise floor introduced by the analyzer we use to perform the measurement. This instrumental noise floor is confirmed by increasing the modulation strength, which improves the dynamic range of our measurement. Conclusion. -- We have experimentally and numerically demonstrated a novel phase noise purifying mechanism by disciplining dissipative solitons with potential traps. The comb repetition rate drift, which is a major lim- itation in microcavities, was thereby suppressed, while this parameter was stabilized to a reference oscillator. The high frequency noise of the trapping signal was self- purified, at frequency offsets well below the cavity reso- nance bandwidth. Our technique reveals the unique dy- namical stability of the self-organized temporal solitons. The exceptional phase noise level achieved with the puri- fied microwaves shows that disciplined DKS are compet- itive with other state-of-the-art optical-microresonator- based microwave oscillators in terms of generating low- 100102104106Offset frequency [Hz]-170-160-150-140-130-120-110-100-90-80-70-60-50-40-30-20-10Phase noise [dBc/Hz]Pre-stabilized onlySoliton disciplinedMicrowave sourceAnalyzer floor102103104105106Frequency [Hz]-60-50-40-30-20-100PM-to-PM transfer index (in power) [dB]Phase Deviation 0.1 radPhase Deviation 1 radPhase Deviation 2 radSimulationnoise microwave signals with miniaturized device. It could also facilitate the application of microcombs in co- herently averaged dual-comb spectroscopy [46] and co- herent optical telecommunication [47]. This supported (Driven publication was by Contract and Nonequilibrium Quan- D18AC00032 tum Systems, DRINQS) from the Defense Advanced Research Projects Agency (DARPA), Defense Sciences Office (DSO) and funding from the Swiss National Science Foundation under grant agreement no. 163864 and no. 165933, and by the Russian Foundation for Basic Research under project 17-02-00522. W.W. acknowl- edges support by funding from the European Union's Horizon 2020 research and innovation programme under Marie Sklodowska-Curie IF grant agreement no. 753749 (SOLISYNTH). ∗ W. Weng and E. Lucas contributed equally to this work. † [email protected] [1] D. M. Pozar, Microwave and RF design of wireless sys- tems (Wiley New York, 2001). [2] L. Maleki, Nature Photonics 5, 728 (2011). [3] S. Grop, P. Bourgeois, N. Bazin, Y. Kersal´e, E. Rubi- ola, C. Langham, M. Oxborrow, D. Clapton, S. Walker, J. De Vicente, et al., Review of Scientific Instruments 81, 025102 (2010). [4] P. L. Stanwix, M. E. Tobar, P. Wolf, M. Susli, C. R. Locke, E. N. Ivanov, J. Winterflood, and F. van Kann, Physical review letters 95, 040404 (2005). [5] M. Nagel, S. R. Parker, E. V. Kovalchuk, P. L. Stanwix, J. G. Hartnett, E. N. Ivanov, A. Peters, and M. E. Tobar, Nature communications 6, 8174 (2015). [6] S. Grop, P.-Y. Bourgeois, R. Boudot, Y. Kersal´e, E. Ru- and V. Giordano, Electronics Letters 46, 420 biola, (2010). [7] J. G. Hartnett, N. R. Nand, and C. Lu, Applied Physics Letters 100, 183501 (2012). [8] T. M. Fortier, M. S. Kirchner, F. Quinlan, J. Tay- lor, J. Bergquist, T. Rosenband, N. Lemke, A. Lud- low, Y. Jiang, C. Oates, et al., Nature Photonics 5, 425 (2011). [9] F. Quinlan, T. M. Fortier, H. Jiang, A. Hati, C. Nel- son, Y. Fu, J. C. Campbell, and S. A. Diddams, Nature Photonics 7, 290 EP (2013). [10] X. Xie, R. Bouchand, D. Nicolodi, M. Giunta, W. Hansel, M. Lezius, A. Joshi, S. Datta, C. Alexandre, M. Lours, et al., Nature Photonics 11, 44 (2017). [11] J. Li, X. Yi, H. Lee, S. A. Diddams, and K. J. Vahala, Science 345, 309 (2014). [12] J. Li, H. Lee, and K. J. Vahala, Nature communications 4, 2097 (2013). [13] W. Loh, J. Becker, D. C. Cole, A. Coillet, F. N. Baynes, S. B. Papp, and S. A. Diddams, New Journal of Physics 18, 045001 (2016). [14] T. Herr, V. Brasch, J. Jost, C. Wang, N. Kondratiev, M. Gorodetsky, and T. Kippenberg, Nature Photonics 8, 145 (2014). [15] N. Akhmediev and A. Ankiewicz, in Dissipative solitons 5 (Springer, 2005) pp. 1 -- 17. [16] T. J. Kippenberg, A. L. Gaeta, M. Lipson, and M. L. Gorodetsky, Science 361, eaan8083 (2018). [17] P. Trocha, M. Karpov, D. Ganin, M. H. Pfeiffer, A. Kordts, S. Wolf, J. Krockenberger, P. Marin-Palomo, C. Weimann, S. Randel, et al., Science 359, 887 (2018). [18] M.-G. Suh and K. J. Vahala, Science 359, 884 (2018). [19] M.-G. Suh, Q.-F. Yang, K. Y. Yang, X. Yi, and K. J. Vahala, Science 354, 600 (2016). [20] N. Pavlov, G. Lihachev, S. Koptyaev, E. Lucas, M. Kar- and pov, N. Kondratiev, I. Bilenko, T. Kippenberg, M. Gorodetsky, Optics Letters 42, 514 (2017). [21] A. Dutt, C. Joshi, X. Ji, J. Cardenas, Y. Okawachi, K. Luke, A. L. Gaeta, and M. Lipson, Science advances 4, e1701858 (2018). [22] E. Obrzud, M. Rainer, A. Harutyunyan, M. H. Ander- son, J. Liu, M. Geiselmann, B. Chazelas, S. Kunder- mann, S. Lecomte, M. Cecconi, A. Ghedina, E. Molinari, F. Pepe, F. Wildi, F. Bouchy, T. J. Kippenberg, and T. Herr, Nature Photonics 13, 31 (2019). [23] M.-G. Suh, X. Yi, Y.-H. Lai, S. Leifer, I. S. Grudinin, G. Vasisht, E. C. Martin, M. P. Fitzgerald, G. Dopp- mann, J. Wang, D. Mawet, S. B. Papp, S. A. Diddams, C. Beichman, and K. Vahala, Nature Photonics 13, 25 (2019). [24] D. T. Spencer, T. Drake, T. C. Briles, J. Stone, L. C. Sinclair, C. Fredrick, Q. Li, D. Westly, B. R. Ilic, A. Blue- stone, et al., Nature 557, 81 (2018). [25] S. B. Papp, K. Beha, P. Del'Haye, F. Quinlan, H. Lee, K. J. Vahala, and S. A. Diddams, Optica 1, 10 (2014). [26] P. Del'Haye, A. Coillet, T. Fortier, K. Beha, D. C. Cole, K. Y. Yang, H. Lee, K. J. Vahala, S. B. Papp, and S. A. Diddams, Nature Photonics 10, 516 (2016). [27] W. Liang, D. Eliyahu, V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, D. Seidel, and L. Maleki, Nature commu- nications 6, 7957 (2015). [28] J. Jost, E. Lucas, T. Herr, C. Lecaplain, V. Brasch, M. Pfeiffer, and T. Kippenberg, Optics letters 40, 4723 (2015). [29] S.-W. Huang, J. Yang, M. Yu, B. H. McGuyer, D.-L. Kwong, T. Zelevinsky, and C. W. Wong, Science ad- vances 2, e1501489 (2016). [30] S. B. Papp, P. Del'Haye, and S. A. Diddams, Physical Review X 3, 031003 (2013). [31] J. Lim, A. A. Savchenkov, E. Dale, W. Liang, D. Eliyahu, V. Ilchenko, A. B. Matsko, L. Maleki, and C. W. Wong, Nature Communications 8, 8 (2017). [32] S. B. Papp, P. Del'Haye, and S. A. Diddams, Optics Express 21, 17615 (2013). [33] F. Leo, S. Coen, P. Kockaert, S.-P. Gorza, P. Emplit, and M. Haelterman, Nature Photonics 4, 471 (2010). [34] D. G. Matei, T. Legero, S. Hafner, C. Grebing, R. Weyrich, W. Zhang, L. Sonderhouse, J. M. Robin- son, J. Ye, F. Riehle, and U. Sterr, Phys. Rev. Lett. 118, 263202 (2017). [35] W. Liang, A. A. Savchenkov, V. S. Ilchenko, D. Eliyahu, A. B. Matsko, and L. Maleki, IEEE Photonics Journal 9, 1 (2017). [36] W. Weng, J. D. Anstie, P. Abbott, B. Fan, T. M. Stace, and A. N. Luiten, Physical Review A 91, 063801 (2015). [37] B. Razavi, IEEE Journal of Solid-State Circuits 39, 1415 (2004). [38] H. Taheri, A. B. Matsko, and L. Maleki, The European Physical Journal D 71, 153 (2017). 6 [39] H. Taheri, A. A. Eftekhar, K. Wiesenfeld, A. Adibi, et al., IEEE Photon. J. 7 (2015). [40] J. K. Jang, M. Erkintalo, S. Coen, and S. G. Murdoch, Nature communications 6, 7370 (2015). [41] V. E. Lobanov, G. Lihachev, and M. L. Gorodetsky, EPL (Europhysics Letters) 112, 54008 (2015). [42] L. A. Lugiato and R. Lefever, Physical review letters 58, 2209 (1987). [43] H. Guo, E. Lucas, M. H. P. Pfeiffer, M. Karpov, M. An- derson, J. Liu, M. Geiselmann, J. D. Jost, and T. J. Kippenberg, Physical Review X 7, 041055 (2017). [44] A. B. Matsko and L. Maleki, JOSA B 32, 232 (2015). [45] See Supplementary Material for details, which includes Refs. [48-50]. [46] I. Coddington, N. Newbury, and W. Swann, Optica 3, 414 (2016). [47] P. Marin-Palomo, J. N. Kemal, M. Karpov, A. Kordts, J. Pfeifle, M. H. Pfeiffer, P. Trocha, S. Wolf, V. Brasch, M. H. Anderson, et al., Nature 546, 274 (2017). [48] X. Yi, Q.-F. Yang, X. Zhang, K. Y. Yang, X. Li, and K. Vahala, Nature Communications 8, 14869 EP (2017). [49] Y. Wang, F. Leo, J. Fatome, M. Erkintalo, S. G. Mur- doch, and S. Coen, Optica 4, 855 (2017). [50] E. Lucas, H. Guo, J. D. Jost, M. Karpov, and T. J. Kippenberg, Phys. Rev. A 95, 043822 (2017). Supplementary material for "Spectral purification of microwave signals with disciplined dissipative Kerr solitons" 7 LOCKING RANGE MEASUREMENT WITH PM INJECTION LOCKING In the main text we show the measured locking ranges with AM applied to the pump laser. We also measure the locking ranges with PM scheme and with varied PM strength, which are presented in Fig. 6. We note that the locking range can be influenced by mode coupling between different mode families (see the next section), and that the resulting effect on the locking range can vary with temperature and fiber-resonator coupling condition. However, the results show the similar linear dependence of the locking range on the modulation strength with relatively small PM strength. FIG. 6. Measured locking ranges with laser-resonance detuning of 11.8 MHz. The normalized amplitude of modulation sideband corresponds to 2, where is defined in Eq. 1 in the main text. MODE CROSSINGS' EFFECT ON THE LOCKING RANGE We apply PM injection locking by modulating the pump laser with frequencies close to the cavity FSR (∼14.09 GHz) with an EOM. We measure the locking ranges at different laser-resonance detunings but with the same PM strength. For the measurement the laser-resonance detuning is tuned from high frequency (30 MHz) to low frequency (11.8 MHz), which we referred to as backward tuning. We observe that the locking ranges change dramatically and erratically as the detuning changes, as presented in Fig. 7. Then we tune the detuning back to high frequency (forward tuning) and measure the locking range at detuning of 18.8 MHz, which is also presented in the same figure. From this comparison we observe strong bistability of the locking range at the almost same detuned positions, as indicated by the green double-arrow. We suspect that such behavior is induced by the mode coupling between different mode families [43], which can produce strong single-mode dispersive waves [48] that can modulate the CW background and trap the solitons [49] and consequently nullify the injection locking. To verify our speculation, we measure the DKS frep and the comb spectrum continuously as we tune the detuning backward and then forward. Fig. 8 shows the experimental results. Indeed, bistability of the repetition rate shows with different tuning directions, due to the appearance and disappearance of the single-mode dispersive waves that can shift frep through soliton recoil [50]. 00.050.10.150.2Normalized amplitude of modulation sideband020040060080010001200Locking range [Hz]8 FIG. 7. Shift of the DKS frep as the laser-resonance detuning is tuned forward (to high frequency) and backward (to low frequency) respectively. Four comb spectra corresponding to the four points labeled on the curves are presented, showing that the strong bistability of frep shift is caused by the mode coupling between different mode families, which also causes the appearance and disappearance of the single-mode dispersive wave indicated by the red arrows. FIG. 8. Shift of the DKS frep as the laser-resonance detuning is tuned forward (to high frequency) and backward (to low frequency) respectively. Four comb spectra corresponding to the four points labeled on the curves are presented, showing that the strong bistability of frep shift is caused by the mode coupling between different mode families, which also causes the appearance and disappearance of the single-mode dispersive wave indicated by the red arrows. NUMERICAL SIMULATION OF PM-TO-PM TRANSFER FUNCTION LLE-like model described in [43] is used to perform the simulation with adaptive Runge-Kutta method. We include both the second-order dispersion (D2/2π = 2 kHz) and third-order dispersion (D3/2π = −4 Hz) in Dint(µ). Due to the high-order dispersion effect, the soliton repetition rate displays a frequency shift of ∼ −7.3 kHz with respect to the cavity FSR (14 GHz). We set the input microwave signal (Ω) to match the repetition rate so the solitons can be trapped by the modulated intracavity background. The central pump power is 200 mW and the PM sideband power 1015202530Detuning [MHz]0200400600800Locking range [Hz]BackwardForward510152025303540Resonance-laser detuning [MHz]-20-15-10-5051015Repetition rate shift [kHz]ForwardBackward12341510152015301540155015601570158015901600Wavelength [nm]-60-40-200Power [dBm]1510152015301540155015601570158015901600Wavelength [nm]-60-40-200Power [dBm]1510152015301540155015601570158015901600Wavelength [nm]-60-40-200Power [dBm]1510152015301540155015601570158015901600Wavelength [nm]-60-40-200Power [dBm]1234is set to be 10 mW, which corresponds to 2 4 = 0.05. After tens of photon decay time the position of the soliton with respect to the modulated background does not change anymore, indicating that frep is injection-locked. In Fig. 9 we plot the trapped soliton and the modulated intracavity background. 9 FIG. 9. Temporal profile of a soliton that is trapped by the modulated intracavity background. The pump power is 200 mW. Phase modulation with = 0.32 (5 mW for the modulation sideband power) is used for the simulation. After the soliton is trapped, we apply PM on the injected microwave signal. This modulation is directly converted into PM on the optical modulation sidebands. We choose the modulation amplitude (PM deviation) to be 0.1 radian and let the simulation run for 8 -- 20 modulation periods. The phase response of the disciplined soliton repetition rate is derived with Eq. 2 in the main text. Then we apply fast Fourier transform (FFT) on the phase response in the time domain to obtain the spectrum of the phase variation [44]. Fig. 10 shows the spectrum of the phase variation of the injection-locked frep as PM at 1 MHz is applied to the optical modulation sidebands. With the phase response spectrum we can calculate the PM-to-PM transfer indices at particular PM frequencies. We choose discrete PM frequencies from 200 Hz to 1 MHz to simulate the transfer function shown in the main text. For comparison, we also carry out simulations of PM injection locking with larger and AM injection locking. The simulated results are summarized in Fig. 11. As the modulation strength increases, the transfer function amplitude rises, showing that the soliton trapping bandwidth is enhanced. Moreover, AM injection locking exhibits higher transfer indices at offset frequencies higher than ∼ 100 kHz. EXPERIMENTAL MEASUREMENT OF PM-TO-PM TRANSFER FUNCTION To verify the simulated PM-to-PM noise transfer function, we implement PM on the input microwave signals via internal function of the synthesizer that is used to drive the optical phase modulator. The PM amplitude is set to be 0.1 radian, 1 radian and 2 radian respectively. After the soliton repetition rate is injection locked, the PM on the microwave signals is activated, and the in-phase-and-quadrature (IQ) demodulation function of the spectrum analyzer is utilized to measure the phase and amplitude responses of the microwave signals generated by the fast photodetector that detects the DKS repetition rate. The demodulation frequency on the analyzer is set equal to the synthesizer frequency and both devices are referenced to a common clock oscillator. In Fig. 12 we plot the measured results of the IQ phasors at 4 different PM frequencies shown in the complex plane when the PM phase deviation is 0.1 radian. We note that the approach can also be used to measure the PM-to-AM noise transfer function. As shown in the main text, as the PM frequency on the input microwave signal increases, the phase deviation of the purified signal decreases. To quantify this effect, the phase deviation is extracted from the IQ measurement (cid:18) ∆φ(measured)(ω) (cid:19) ∆φ(in)(ω) and the transfer function is obtained as 20 log10 . Eventually, the response amplitude is below 10 FIG. 10. Spectrum of the phase variation of the injection-locked frep. The PM amplitude (phase deviation) on the optical sidebands is 0.1 radian. The PM frequency is 1 MHz. FIG. 11. Simulated PM-to-PM transfer functions with different injection-locking schemes and varied laser modulation strengths. The optical sideband powers are indicated in the legend. the instrumental noise floor of the phase noise analyzer, which is evidently displayed as the flat floor on the transfer function at high frequencies. To increase the dynamical range of the measurement, the measurement is repeated with larger phase deviations (1 radian and 2 radians) on the modulation and re-take the IQ measurement. Fig. 13 shows 4 sets of measured results at 4 different PM frequencies. We see that at very high PM frequency (1 MHz) the response is limited to the same instrumental noise floor. However, because of the larger input phase deviation, the noise floor level of the transfer function decreases, as shown by Fig. 5 in the main text. 102103104105106Offset frequency [Hz]-70-60-50-40-30-20-100PM-to-PM transfer index (in power) [dB]PM 10mW (ε2=0.2)PM 20mW (ε2=0.4)PM 40mW (ε2=0.8)AM 10mW (ε2=0.2)11 FIG. 12. Measured results of the microwave signal from the IQ test. The amplitude and the phase responses to the injected PM on the input microwave signal can be derived from the data in complex plane. The 4 sets of data are corresponding to 4 different PM frequencies that are indicated in the figure. The phase deviation of the PM on the input microwave signal is set to be 0.1 radian. FIG. 13. Measured results of the microwave signal from the IQ test with phase deviation of 1 radian.
1910.09105	1	1910	2019-10-21T01:33:25	Electrochemically-stable ligands bridge photoluminescence-electroluminescence gap of quantum dots	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Colloidal quantum dots (QDs) are promising emitters for electroluminescence devices (QD light-emitting-diodes, QLEDs). Though QDs have been synthesized with efficient and stable photoluminescence, inheriting their superior luminescence in QLEDs remains challenging. This is commonly attributed to unbalanced charge injection and/or interfacial exciton quenching in the devices, instead of lack of suited QD materials. Here, a general but previously overlooked degradation channel in QLEDs, i.e., operando electrochemical reactions of surface ligands with injected charge carriers, is identified after systematic studies of various combination of core/shell QDs and ligands. Applying electrochemically-inert ligands to highly photoluminescent QDs is developed to bridge their photoluminescence-electroluminescence gap. This material-design principle is general for boosting electroluminescence efficiency and lifetime of the QLEDs, resulting in record-long operational lifetimes for both red-emitting QLEDs (T95 > 3800 hours at 1000 cd m-2) and blue-emitting QLEDs (T50 >10,000 hours at 100 cd m-2). Our study provides a critical guideline for the QDs to be used in optoelectronic and electronic devices.	physics.app-ph	physics	Electrochemically-stable ligands bridge the photoluminescence- electroluminescence gap of quantum dots Chaodan Pu,1† Xingliang Dai,1,2† Yufei Shu,1† Meiyi Zhu,1 Yunzhou Deng,1,2 Yizheng Jin1,2 & Xiaogang Peng1 1 Centre for Chemistry of High-Performance & Novel Materials, Department of Chemistry, Zhejiang University, Hangzhou 310027, China. 2 State Key Laboratory of Silicon Materials, Zhejiang University, Hangzhou 310027, China. †These authors contributed equally to this work. Correspondence authors: Prof. Yizheng Jin ([email protected]); Prof. Xiaogang Peng ([email protected]) 1 Abstract Colloidal quantum dots (QDs) are promising emitters for electroluminescence devices, including QD light-emitting-diodes (QLEDs). Though QDs have been synthesized with efficient, stable, and high colour-purity photoluminescence, inheriting their superior luminescent properties in QLEDs remains challenging. This is commonly attributed to unbalanced charge injection and/or interfacial exciton quenching in the devices. Here, a general but previously overlooked degradation channel in QLEDs, i.e., operando electrochemical reactions of surface ligands with injected charge carriers, is identified. We develop a strategy of applying electrochemically-inert ligands to QDs with excellent luminescent properties to bridge their photoluminescence- electroluminescence gap. This material-design principle is general for boosting electroluminescence efficiency and lifetime of the QLEDs, resulting in record-long operational lifetimes for both red-emitting QLEDs (T95 > 3800 hours at 1000 cd m-2) and blue-emitting QLEDs (T50 >10,000 hours at 100 cd m-2). Our study provides a critical guideline for the QDs to be used in optoelectronic and electronic devices. 2 Colloidal QDs are solution-processable semiconductor nanocrystals coated with a monolayer of surface ligands1. Progresses on synthetic chemistry of core/shell QDs have led to a unique class of materials possessing efficient, stable, and high colour-purity photoluminescence properties2-6. Electroluminescence of the core/shell QDs is expected to harness the unique combination of their superior photoluminescence properties and excellent solution processability, enabling high- performance and cost-effective QLEDs, electrically driven single-photon sources, and potentially electrically-pumped lasers7-17. However, fabricating high-performance QLEDs remains to be challenging and empirical, which is considered as a result of unbalanced charge injection and interfacial exciton quenching18. Despite extensive efforts on material screening and device engineering, only a few exceptional examples are documented for red and green devices with decent external quantum efficiency and operational lifetime9,10,15,17. The operational lifetime of blue QLEDs remains to be very short (typically 100-1000 hours for an initial brightness of 100 cd m-2 to drop 50%)10,19, shadowing the future of QLED technology. These facts indicate that at the fundamental level new design principles for QDs and/or QLEDs are necessary for bridging their photoluminescence- electroluminescence gap. Photoluminescence and electroluminescence of QDs differ from each other in the excitation processes. In photoluminescence, an exciton is generated and confined mostly in the centre of a core/shell QD by absorbing a photon. In electroluminescence, an exciton is generated by injecting an electron and a hole separately into a core/shell QD. Surface ligands are inevitably accessible during the carrier-injection processes. This means that, though surface ligands are usually optimized for achieving high-performance photoluminescence, their electronic and electrochemical properties 3 should be taken into consideration in terms of designing QDs for high-performance electroluminescence devices. Here, we report that electrochemical stability of QDs under device operation conditions is an intrinsic but previously overlooked issue. Given the inner inorganic lattice of a QD is electrochemically stable, the electrochemical degradation of QDs in the devices is largely associated with their inorganic- ligand interface. We systematically investigate the core/shell QDs with various combinations of outer shells and surface ligands, generalizing the concept of electrochemically-inert ligands to be essential for bridging the photoluminescence-electroluminescence gap of QDs. Ligand-dependent electroluminescence of the CdSe/CdS core/shell QDs with ideal photoluminescence properties. The recently developed CdSe/CdS core/shell QDs with nearly ideal photoluminescence properties4 are applied as the first model system. The as-synthesized QDs, denoted as CdSe/CdS-Cd(RCOO)2 QDs, comprise of 3-nm CdSe cores and 7-nm (10 monolayers) CdS shells coated with two types of ligands. These two types of ligands are carboxylate ligands bonded onto the surface cadmium sites on the polar facets and cadmium-carboxylate ligands weakly adsorbed onto the non-polar facets20,21 (Fig. 1a, left scheme). An established surface-treatment procedure22 converts the CdSe/CdS-Cd(RCOO)2 QDs to the QDs solely with fatty amine ligands (CdSe/CdS-RNH2) (Fig. 1a, right scheme and Supplementary Fig. 1). According to literature, only the standard Cd2+/Cd0 reduction potential of cadmium carboxylates is within the bandgap of CdSe and that of the other components of all ligands are far outside the bandgap (Supplementary Fig. 2)23,24. Optical measurements (Fig. 1b, 1c and Supplementary Fig. 3) show that both types of the CdSe/CdS core/shell QDs, either in solution or 4 in film, exhibit nearly identical, stable, and highly efficient photoluminescence. However, the CdSe/CdS-RNH2 QDs and the CdSe/CdS-Cd(RCOO)2 QDs deliver strikingly different electroluminescence performance in an identical QLED structure (Fig. 1d-g). With the surface-treated CdSe/CdS-RNH2 QDs, both current density and luminance of the QLEDs show a steep increase simultaneously at ~1.65 V (Fig. 1e). A device with the best efficiency shows a peak external quantum efficiency (EQE) of 20.2% (Fig. 1f), and the average peak EQE of our devices is 18.6%. Since the out-coupling efficiency of our devices is simulated to be ~23.5% by following a literature method25, the average peak internal quantum efficiency (IQE) is estimated to be ~80%. The device lifetime (T50, time for the luminance decreasing by 50%), is estimated to be ~90,000 hours at 100 cd m-2 by applying an acceleration factor of 1.816 (Fig. 1g). The QLEDs using the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs deliver an extremely low peak EQE of 0.2% and a negligible T50 of ~0.3 h at 100 cd m-2 (Fig. 1f, g), indicating dominating and detrimental non-radiative recombination of the injected charges. It is interesting that the devices with the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs possess a voltage gap of 0.8 V between the threshold voltage for current raise and the turn-on voltage for light emission, while this voltage gap is almost zero for the QLEDs using the surface-treated CdSe/CdS-RNH2 QDs (Fig. 1e). Operando electrochemical reduction of Cd2+ ions in the CdSe/CdS-Cd(RCOO)2 QDs in QLEDs. To explain the drastically different electroluminescence performance between the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs and the surface-treated CdSe/CdS-RNH2 QDs (Fig. 1), we propose a scenario of electrochemical reduction of the cadmium-carboxylate ligands by the injected electrons. 5 This scenario is akin to the chemical reduction of the ligands cadmium chalcogenide QDs with cadmium-carboxylate ligands26,27. As shown in Supplementary Fig. 2, the standard reduction potential of cadmium carboxylates (Cd2+/Cd0) is slightly below the bottom of the conduction band of bulk CdSe and CdS23,26. The proposed operando electrochemical reaction shall generate Cd0 species, which effectively quench the luminescence of QDs28, resulting in poor electroluminescence. A series of experiments are carried out to verify this hypothesis. We monitor the relative changes of photoluminescence efficiency of the CdSe/CdS core/shell QDs in working QLEDs at different voltages29 (experimental setup shown in Supplementary Fig. 4). The relative photoluminescence efficiency of the surface-treated CdSe/CdS-RNH2 QDs is almost constant when the device bias is swept from 0 to 2 V. Above 2 V, the photoluminescence efficiency of the surface-treated CdSe/CdS-RNH2 QDs show little yet reversible reduction, which is likely due to increase of the electric field. In contrast, for the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs, the photoluminescence efficiency rapidly decreases when the bias is above 1.2 V (Fig. 2a), which equals to the threshold voltage associated with current raise and is much below the turn-on voltage for this type of QLEDs (Fig. 1e). The reduction of photoluminescence efficiency is not recovered when a zero bias (Fig. 2b) or a negative bias (data not shown) is applied to the QLED. Being irreversible at zero bias means that the possible redox products are stable without an external electric field, suggesting the excess holes are either remotely trapped in the hole-injection/hole-transport layers or resulting in stable oxidation products. The current of a diode is negligible under negative bias, which implies an attempt to reverse the suggested Cd2+/Cd0 reduction by applying negative bias would be difficult in the QLEDs. We demonstrate that electrons but not holes injected into the devices are responsible for the 6 decrease of photoluminescence efficiency of the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs under positive bias (> 1.2 V), supporting the picture of operando reduction of cadmium-carboxylate ligands. Single-carrier devices are fabricated to distinguish the effects of the two types of charge carriers. The results (Fig. 2c) indeed show that photoluminescence efficiency of the CdSe/CdS-Cd(RCOO)2 QDs drops in an electron-only device operating at a constant current density of 10 mA cm-2 and remains to be constant in a hole-only device operating at a constant current density of 30 mA cm-2. As expected, photoluminescence efficiency of the surface-treated CdSe/CdS-RNH2 QDs is stable in either an electron-only device or a hole-only device (Fig. 2c). We conduct voltammetric measurements on the surface-treated CdSe/CdS-RNH2 QDs, free cadmium carboxylates, and the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs in anhydrous tetrahydrofuran (Fig. 2d), verifying the potential alignment documented in literature30,31. The surface- treated CdSe/CdS-RNH2 QDs exhibit a defined cathodic peak at ~0.89 V (vs normal hydrogen electrode (NHE)), corresponding to their lowest unoccupied molecular orbital (LUMO)30,31. In comparison, the reduction peak potential for cadmium carboxylates is less negative, agreeing with the literature23. The characteristics of the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs exhibit combined features of cadmium carboxylates and the CdSe/CdS-RNH2 QDs. These data indicate that the cadmium-carboxylate ligands should be electrochemically active prior to electron injection into the LUMO of the CdSe/CdS core/shell QDs. Finally, we confirm that the product of the operando reduction reaction is Cd0. A QLED with as- synthesized CdSe/CdS-Cd(RCOO)2 QDs is biased for ~50 hours at 2 V. The top electrode and the electron-transporting layer of this device is removed to expose the QD layer (see captions of Fig. 2e 7 and 2f for details). While the QDs exposed to nitrogen remain to be non-emissive, their photoluminescence efficiency can be largely recovered after placing them in oxidative environments, either a solution of dibenzoyl peroxide or oxygen gas (Fig. 2e). Importantly, the photoluminescence spectrum of the recovered QDs is identical to that of the QDs before applying bias stress (Fig. 2e). The QDs along the reduction (in QLED)-oxidation (in oxygen) cycle are further studied by Auger electron spectroscopy, which is sensitive to surface atoms (attenuation length < 1.0 nm)32 and can readily distinguish the cadmium valence states (Cd0 and Cd2+)33. Results (Fig. 2f and Supplementary Fig. 5) illustrate that the cadmium species on the QDs after the bias stress at 2V show signals of Cd0. Conversely, the cadmium species on the other samples of the QDs -- either before the bias stress or after the oxidation by oxygen -- are Cd2+. Bridging the photoluminescence-electroluminescence gap of the CdSe/CdS core/shell QDs by gradual ligand replacement. Infrared spectra (Fig. 3a) show that the surface treatment outlined in Fig. 1a would completely replace all carboxylate and cadmium-carboxylate ligands by fatty amine ones22. As reported in the literature, polar surfaces of zinc-blende II-VI QDs (e.g., {100} and {111} facets) are usually terminated with positively-charged cationic sites (i.e., surface cadmium ions for CdSe and CdS lattice), which are coordinated with negatively-charged carboxylate ligands to fulfil their four coordination bonds and maintain charge neutrality20,21,34-36. The non-polar surfaces ({110} facets as the representative ones) are charge-neutral, on which only neutral cadmium-carboxylate ligands can weakly adsorb20,21,34-36. The surface Cd cations in the former case are a part of the (surface) lattice, and the ones in the latter case should be more like the Cd2+ ions in the cadmium-carboxylate 8 molecules21. We apply an amine-washing procedure (Fig. 3b), which is milder than the amine-based surface treatment, to gradually replace the carboxylate and cadmium-carboxylate ligands on CdSe/CdS-Cd(RCOO)2 QDs. Bonding between the lattice cadmium sites of the nanocrystals and the carboxylate ligands is about one order of magnitude stronger than that between the neutral nonpolar facets and the adsorbed cadmium-carboxylate ligands20,21. Thus, the washing procedure selectively removes the weakly-bonded cadmium-carboxylate ligands. The residual carboxylate groups at the plateau in Fig. 3b (~15% of the original ones) should be the tightly-bonded carboxylate ones on the polar facets. Fig. 3b reveals that the photoluminescence quantum yield of the QDs with different percentages of the residual carboxylates remains to be near unity in solution, which results in a constant photoluminescence quantum yield of ~80% for all QD thin films (Fig. 3c). However, internal quantum yield and operational lifetime (shown in logarithm scale) of electroluminescence for the series of QLEDs demonstrate a strong dependence of ligand composition of the QDs (Fig. 3c and Supplementary Fig. 6). Simultaneously, the voltage gap between the current-raise voltage and the turn- on voltage is also gradually reduced (Fig. 3c). This voltage gap always co-exists with the operando reduction of Cd2+ (or Zn2+ below) ions, implying the early current-raise being associated with the operando reduction reactions. In summary, the results in Fig. 3c confirm that the photoluminescence-electroluminescence gap -- both efficiency and operation lifetime -- of the CdSe/CdS core/shell QDs is gradually bridged by removing the weakly-adsorbed cadmium-carboxylate ligands. Because the plateau portion of the residual carboxylates (~15%) does not contribute to the photoluminescence-electroluminescence gap, 9 the cause for the poor performance of the QLEDs based on the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs should be a result of the operando reduction of the free Cd2+ ions in the form of cadmium carboxylates. Below, if similar amine-based surface treatments are applied, we would not specify the small amount of negatively-charged carboxylate ligands due to their electrochemically-benign nature. Bridging the photoluminescence-electroluminescence gap of other types of red-emitting QLEDs by electrochemically-inert ligands. The outer shells of core/shell QDs are a part of their inorganic- ligand interface. Thus, the outer shells and their specific ligands might both play a role in determining electrochemical properties of the QDs in QLEDs. Besides CdS, ZnS is the other type of most common outer shells of the colloidal core/shell QDs for electroluminescence applications14,37,38. Similar to the cadmium-carboxylate ligands for the CdS outer shells, zinc fatty acid salts (Zn(RCOO)2) are common metal-carboxylate ligands for the ZnS outer shells28. Because the standard reduction potential of Zn2+ is only ~0.3 V more negative than that of Cd2+ (Supplementary Fig. 2), the zinc-carboxylate ligands may also be electrochemically active in QLEDs. In this regard, we adopt another model system for studying the ligand-induced photoluminescence-electroluminescence gap of red-emitting QLEDs, namely, CdSe/CdS/ZnS core/shell/shell QDs. In comparison with the CdSe/CdS core/shell QDs studied above, these core/shell/shell QDs possess the same CdSe core but replace the ten monolayers of CdS shells by a combination of five monolayers of CdS inner shells and three monolayers of ZnS outer shells. During the epitaxial growth of the ZnS outer shells, Zn(RCOO)2 and fatty acids are the only source of possible ligands for the resulting QDs (see Methods). 10 Results in Fig. 4a-d confirm that, though the effects are less dramatic, a significant photoluminescence-electroluminescence gap exists for the as-synthesized CdSe/CdS/ZnS core/shell/shell QDs with zinc-carboxylate ligands (CdSe/CdS/ZnS-Zn(RCOO)2). This gap is removed for the surface-treated CdSe/CdS/ZnS core/shell/shell QDs with amine ligands (CdSe/CdS/ZnS-RNH2 QDs). Both the CdSe/CdS/ZnS-Zn(RCOO)2 and the CdSe/CdS/ZnS-RNH2 QDs possess the same photoluminescence quantum yield of ~80% in thin films. The relative photoluminescence efficiency of the as-synthesized CdSe/CdS/ZnS-Zn(RCOO)2 QDs in both working electron-only devices and QLEDs decreases gradually (Fig. 4a, b), while the surface-treated CdSe/CdS/ZnS-RNH2 QDs offer stable and efficient photoluminescence in these working devices. The QLEDs using the as-synthesized CdSe/CdS/ZnS-Zn(RCOO)2 QDs show low average EQEs of 2.7 ± 0.5% and negligible device lifetime, but the surface-treated CdSe/CdS/ZnS-RNH2 QDs demonstrate high-performance QLEDs with an average EQE of 17.8 ± 0.4% (Fig. 4c) and excellent operational lifetime, i.e., a typical T95 lifetime (time for the luminance decreasing to 95% of the original value) of ~600 hours at 1000 cd m-2 (Fig. 4d). In addition to the ligand systems discussed above, there are some other common types of ligands for colloidal QDs, including organophosphines5,10,15, metal phosphonates39,40, and thiols2. Fig. 4e illustrates the photoluminescence-electroluminescence efficiency comparison for both CdSe/CdS core/shell and CdSe/CdS/ZnS core/shell/shell QDs with different ligands. Table 1 includes the absolute photoluminescence quantum yield of the QDs films, electroluminescence efficiency, and lifetime of the QLEDs. The neutral cadmium-phosphonate ligands cause a very large photoluminescence- electroluminescence gap, similar to that induced by the cadmium-carboxylate ligands (raw data shown 11 in Supplementary Fig. 7a, b). Thiol ligands significantly quench both photoluminescence and electroluminescence of the CdSe/CdS-Cd(RCOO)2 QDs (Supplementary Fig. 7a, b) but barely affect the photoluminescence and electroluminescence of CdSe/CdS-RNH2 QDs (Supplementary Fig. 7 c-f), implying detrimental effects of the neutral cadmium-thiolate ligands but electrochemically-benign nature of the thiolate ligands41. The effects of neutral tri-octylphosphine (TOP) ligands are quite similar to those of the amine ones in terms of eliminating the photoluminescence-electroluminescence gap (raw data shown in Supplementary Fig. 7 c-f). Replacing cadmium carboxylates by zinc carboxylates as ligands results in a mild reduction of the photoluminescence-electroluminescence gap for the CdSe/CdS QDs. Magnesium-carboxylate (Mg(RCOO)2) ligands largely eliminate the photoluminescence-electroluminescence gap of the resulting QLEDs (Supplementary Fig.7 g, h), which are consistent with the high reduction potential of Mg2+ ions (Supplementary Fig. 2). In addition to CdS and ZnS, zinc and cadmium chalcogenide alloys are sometimes applied as outer shells. Experimental results based on the CdSe/CdZnSe/CdZnS core/shell/shell QDs illustrate that a significant photoluminescence-electroluminescence gap exists when zinc-carboxylate ligands are in place (Table 1 and Supplementary Fig. 8). Replacement of the neutral zinc-carboxylate ligands by TOP ligands leads to devices exhibiting an extremely long operation T95 lifetime of 3800 hours at 1000 cd m-2 (Table 1 and Supplementary Fig. 8), surpassing the stability record of QLEDs15. In summary, above results (Fig. 4e, Table 1, Supplementary Fig. 7, and Supplementary Fig. 8) suggest that similar to fatty amines, neutral organophosphines are electrochemically-inert ligands for the core/shell QDs used in QLEDs. For the ligands of Cd/Zn-based salts, no matter what anionic groups (phosphonates, carboxylates, or thiolates) are, they are all detrimental to performance of the QLEDs. 12 From cadmium carboxylates, zinc carboxylates, to magnesium carboxylates, the detrimental effects to electroluminescence of the CdSe/CdS QDs alleviate gradually. All common types of semiconductor outer shells are electrochemically-inert in the QLEDs as long as electrochemically-inert ligand systems are in place. All these results, including both electroluminescence efficiency and lifetime, are consistent with the above conclusion, that the degradation channel induced by the operando electrochemical reactions of surface ligands is critical for the performance of QLEDs. Improving the operational lifetime of blue-emitting QLEDs using electrochemically-inert ligands. The short operational lifetime of blue-emitting QLEDs is currently the bottleneck for commercialization of the QLED technology. In this regard, blue-emitting CdSeS/ZnSeS/ZnS core/shell/shell QDs with either zinc-carboxylate (CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs) or amine ligands (CdSeS/ZnSeS/ZnS-RNH2 QDs) are comparatively studied. Photoluminescence of the CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs under electric bias is unstable in working electron-only devices and QLEDs (Fig. 5a, b), indicating similar electrochemical degradation discussed above. Conversely, photoluminescence of the CdSeS/ZnSeS/ZnS-RNH2 QDs under electric bias is stable during the operation of both electron-only devices and QLEDs (Fig. 5a, b). Both electroluminescence efficiency and device operation lifetime of the QLEDs with the CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs are poor (Fig. 5c, d). In comparison, blue-emitting QLEDs with the CdSeS/ZnSeS/ZnS-RNH2 QDs show decent efficiency, as characterized by a peak EQE of ~10% (Fig. 5c), though their photoluminescence quantum yield in thin film (~60%) is lower than that (~80%) of the CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs (Supplementary Fig. 9). The operational lifetime of the 13 QELDs with the CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs is also substantially enhanced (~85 hours at an initial brightness of 1450 cd m-2) (Fig. 5d). By applying an experimentally-determined acceleration factor of 1.88 (Supplementary Fig. 9), our devices possess a lifetime of >10,000 hours at 100 cd m-2, representing the most stable blue QLED reported so far. Discussion and outlook. In literature, QLEDs with high efficiencies -- sometimes also with decent device lifetime -- have been reported by several groups using the core/shell QDs9,10,15-17,42-47. In all these reports (Table S1), a common feature is that excessive electrochemically-inert ligands identified above, i.e., fatty amines and/or organophosphines, are used in the synthetic systems. It is known that, in such a system, the neutral cadmium/zinc-carboxylate ligands can be largely removed during isolation of the QDs20,34,48, which would result in unintentional fulfilment of the criterion for electrochemically-stable QDs for QLEDs. We emphasize that, without full control of surface ligands of QDs by design, fabricating QLEDs with high efficiency, long lifetime, and good reproducibility would remain to be empirical. It is worthy to note that electrochemically-inert ligands can only bridge the photoluminescence- electroluminescence gap -- both efficiency and lifetime -- of the QDs, and high efficiency of photoluminescence and electroluminescence requires a suited combination of QD structures and electrochemically-inert ligands. For instance, with the ZnSe inner shells and thin ZnS outer shells -- generally true for the blue-emitting QLEDs in Fig. 5 and those non-cadmium InP QLEDs49, removal of zinc carboxylate ligands by either amines or phosphines is less controllable and sometimes reduces photoluminescence efficiency (Supplementary Fig. 9), which would counteract the improvement of 14 device performance by the electrochemically-inert ligands. Thus, the design of new core/shell structures coupled with further development of electrochemically-inert ligands, including the surface- treatment procedures, is the direction to pursuit for achieving ideal performance of various QLEDs. In this regard, though they are under investigation, magnesium fatty acid salts (also other electrochemically-stable metal ions) are interesting candidates for broadening the spectrum of electrochemically-inert ligands (Fig. 4e and Table 1). In conclusion, it is indeed interesting to see that a simple, general, and materials-based strategy can effectively bridge the photoluminescence-electroluminescence gap of QDs, leading to red-emitting and blue-emitting QLEDs with record-long operational lifetime. These advancements lead us to believe that there are no fundamental obstacles to fully exploit the readily achievable superior photoluminescence properties of QDs to realize high-performance and solution-processed electroluminescence devices, such as QLEDs, quantum light sources, and lasers. Our study further suggests that the electrochemical stability of ligands is a critical design parameter for QDs used in optoelectronic and electronic devices because the operation of these devices all involves charge injection into the QDs. 15 References 1. Steigerwald, M. L. & Brus, L. E. Semiconductor Crystallites: a Class of Large Molecules. Acc. Chem. Res. 23, 183- 188, (1990). 2. Chen, O. et al. Compact high-quality CdSe-CdS core-shell nanocrystals with narrow emission linewidths and suppressed blinking. Nat. Mater. 12, 445-451, (2013). 3. Cao, H. J. et al. Design and Synthesis of Antiblinking and Antibleaching Quantum Dots in Multiple Colors via Wave Function Confinement. J. Am. Chem. Soc. 138, 15727-15735, (2016). 4. Zhou, J. H., Zhu, M. Y., Meng, R. Y., Qin, H. Y. & Peng, X. G. Ideal CdSe/CdS Core/Shell Nanocrystals Enabled by Entropic Ligands and Their Core Size-, Shell Thickness-, and Ligand-Dependent Photoluminescence Properties. J. Am. Chem. Soc. 139, 16556-16567, (2017). 5. Park, Y. S., Lim, J. & Klimov, V. I. Asymmetrically strained quantum dots with non-fluctuating single-dot emission spectra and subthermal room-temperature linewidths. Nat. Mater. 18, 249-255, (2019). 6. Hanifi, D. A. et al. Redefining near-unity luminescence in quantum dots with photothermal threshold quantum yield. Science 363, 1199-1202, (2019). 7. Colvin, V. L., Schlamp, M. C. & Alivisatos, A. P. Light-Emitting-Diodes Made from Cadmium Selenide Nanocrystals and a Semiconducting Polymer. Nature 370, 354-357, (1994). 8. Coe, S., Woo, W. K., Bawendi, M. & Bulovic, V. Electroluminescence from single monolayers of nanocrystals in molecular organic devices. Nature 420, 800-803, (2002). 9. Dai, X. L. et al. Solution-processed, high-performance light-emitting diodes based on quantum dots. Nature 515, 96- 99, (2014). 10. Yang, Y. X. et al. High-efficiency light-emitting devices based on quantum dots with tailored nanostructures. Nat. Photonics 9, 259-266, (2015). 11. Oh, N. R. et al. Double-heterojunction nanorod light-responsive LEDs for display applications. Science 355, 616-619, (2017). 12. Lin, X. et al. Electrically-driven single-photon sources based on colloidal quantum dots with near-optimal antibunching at room temperature. Nat. Commun. 8, 1132, (2017). 13. Lim, J., Park, Y. S. & Klimov, V. I. Optical gain in colloidal quantum dots achieved with direct-current electrical pumping. Nat. Mater. 17, 42-49, (2018). 14. Li, X. Y. et al. Bright colloidal quantum dot light-emitting diodes enabled by efficient chlorination. Nat. Photonics 12, 159-165, (2018). 15. Cao, W. R. et al. Highly stable QLEDs with improved hole injection via quantum dot structure tailoring. Nat. Commun. 9, 2608, (2018). 16. Zhang, Z. X. et al. High-Performance, Solution-Processed, and Insulating-Layer-Free Light-Emitting Diodes Based on Colloidal Quantum Dots. Adv. Mater. 30, 1801387, (2018). 17. Lim, J., Park, Y. S., Wu, K. F., Yun, H. J. & Klimov, V. I. Droop-Free Colloidal Quantum Dot Light-Emitting Diodes. Nano Lett. 18, 6645-6653, (2018). 18. Shirasaki, Y., Supran, G. J., Bawendi, M. G. & Bulovic, V. Emergence of colloidal quantum-dot light-emitting technologies. Nat. Photonics 7, 13-23, (2013). 19. Acharya, K. P., Titov, A., Ying, C. Y., Hyvonen, J. & Holloway, P. H. High-Performance Quantum Dot Light Emitting Diodes and their Challenges SID Symp. Dig. Tech. Papers. 49, 969-972, (2018). 20. Singh, S. et al. Colloidal CdSe Nanoplatelets, A Model for Surface Chemistry/Optoelectronic Property Relations in Semiconductor Nanocrystals. J. Am. Chem. Soc. 140, 13292-13300, (2018). 16 21. Zhang, J. et al. Identification of Facet-Dependent Coordination Structures of Carboxylate Ligands on CdSe Nanocrystals. J. Am. Chem. Soc. 141, 15675-15683, (2019). 22. Pu, C. D. & Peng, X. G. To Battle Surface Traps on CdSe/CdS Core/Shell Nanocrystals: Shell Isolation versus Surface Treatment. J. Am. Chem. Soc. 138, 8134-8142, (2016). 23. Shim, M. & Guyot-Sionnest, P. N-type colloidal semiconductor nanocrystals. Nature 407, 981-983, (2000). 24. Simon, T. et al. Redox shuttle mechanism enhances photocatalytic H-2 generation on Ni-decorated CdS nanorods. Nat. Mater. 13, 1013-1018, (2014). 25. Neyts, K. A. Simulation of light emission from thin-film microcavities. J. Opt. Soc. Am. A 15, 962-971, (1998). 26. Zhao, J., Holmes, M. A. & Osterloh, F. E. Quantum Confinement Controls Photocatalysis: A Free Energy Analysis for Photocatalytic Proton Reduction at CdSe Nanocrystals. ACS Nano 7, 4316-4325, (2013). 27. Hartley, C. L. & Dempsey, J. L. Electron-Promoted X-Type Ligand Displacement at CdSe Quantum Dot Surfaces. Nano Lett. 19, 1151-1157, (2019). 28. Boles, M. A., Ling, D., Hyeon, T. & Talapin, D. V. The surface science of nanocrystals. Nat. Mater. 15, 141-153, (2016). 29. Shirasaki, Y., Supran, G. J., Tisdale, W. A. & Bulovic, V. Origin of Efficiency Roll-Off in Colloidal Quantum-Dot Light-Emitting Diodes. Phys. Rev. Lett. 110, 217403, (2013). 30. Impellizzeri, S. et al. Structural Implications on the Electrochemical and Spectroscopic Signature of CdSe-ZnS Core- Shell Quantum Dots. J. Phys. Chem. C 114, 7007-7013, (2010). 31. Amelia, M., Lincheneau, C., Silvi, S. & Credi, A. Electrochemical properties of CdSe and CdTe quantum dots. Chem. Soc. Rev. 41, 5728-5743, (2012). 32. Hoener, C. F. et al. Demonstration of a Shell Core Structure in Layered CdSe-ZnSe Small Particles by X-Ray Photoelectron and Auger Spectroscopies. J. Phys. Chem. 96, 3812-3817, (1992). 33. Sen, P., Hegde, M. S. & Rao, C. N. R. Surface oxidation of cadmium, indium, tin and antimony by photoelectron and Auger spectroscopy. Appl. Surf. Sci. 10, 63-74, (1982). 34. Anderson, N. C., Hendricks, M. P., Choi, J. J. & Owen, J. S. Ligand Exchange and the Stoichiometry of Metal Chalcogenide Nanocrystals: Spectroscopic Observation of Facile Metal-Carboxylate Displacement and Binding. J. Am. Chem. Soc. 135, 18536-18548, (2013). 35. Owen, J. The coordination chemistry of nanocrystal surfaces. Science 347, 615-616, (2015). 36. Saniepay, M., Mi, C. J., Liu, Z. H., Abel, E. P. & Beaulac, R. Insights into the Structural Complexity of Colloidal CdSe Nanocrystal Surfaces: Correlating the Efficiency of Nonradiative Excited-State Processes to Specific Defects. J. Am. Chem. Soc. 140, 1725-1736, (2018). 37. Lee, K. H. et al. Over 40 cd/A Efficient Green Quantum Dot Electroluminescent Device Comprising Uniquely Large- Sized Quantum Dots. ACS Nano 8, 4893-4901, (2014). 38. Lee, K. H. et al. Highly Efficient, Color-Pure, Color-Stable Blue Quantum Dot Light-Emitting Devices. ACS Nano 7, 7295-7302, (2013). 39. Mashford, B. S. et al. High-efficiency quantum-dot light-emitting devices with enhanced charge injection. Nat. Photonics 7, 407-412, (2013). 40. Tan, R. et al. Monodisperse Hexagonal Pyramidal and Bipyramidal Wurtzite CdSe-CdS Core-Shell Nanocrystals. Chem. Mater. 29, 4097-4108, (2017). 41. Jeong, S. et al. Effect of the thiol-thiolate equilibrium on the photophysical properties of aqueous CdSe/ZnS nanocrystal quantum dots. J. Am. Chem. Soc. 127, 10126-10127, (2005). 42. Cho, I. et al. Multifunctional Dendrimer Ligands for High-Efficiency, Solution-Processed Quantum Dot Light- Emitting Diodes. Acs Nano 11, 684-692, (2017). 17 43. Kim, D. et al. Polyethylenimine Ethoxylated-Mediated All-Solution-Processed High-Performance Flexible Inverted Quantum Dot-Light-Emitting Device. ACS Nano 11, 1982-1990, (2017). 44. Acharya, K. P. et al. High efficiency quantum dot light emitting diodes from positive aging. Nanoscale 9, 14451- 14457, (2017). 45. Lin, Q. et al. Nonblinking Quantum-Dot-Based Blue Light-Emitting Diodes with High Efficiency and a Balanced Charge-Injection Process. ACS Photonics 5, 939-946, (2018). 46. Sun, Y. et al. Investigation on Thermally Induced Efficiency Roll-Off: Toward Efficient and Ultrabright Quantum- Dot Light-Emitting Diodes. ACS Nano, (2019). 47. Chen, S. et al. On the degradation mechanisms of quantum-dot light-emitting diodes. Nat Commun 10, 765, (2019). 48. Yang, Y., Li, J. Z., Lin, L. & Peng, X. G. An efficient and surface-benign purification scheme for colloidal nanocrystals based on quantitative assessment. Nano. Res. 8, 3353-3364, (2015). 49. Li, Y. et al. Stoichiometry-Controlled InP-Based Quantum Dots: Synthesis, Photoluminescence, and Electroluminescence. J. Am. Chem. Soc. 141, 6448-6452, (2019). 18 Acknowledgements This work was financially supported by the National Key R&D Program of China (2016YFB0401600), the National Natural Science Foundation of China (91833303), and the Fundamental Research Funds for the Central Universities (2017XZZX001-03A). We thank Prof. Peng Wang and Mr. Yanlei Hao (Zhejiang University, China) for the assistances on providing the access to voltammetric equipment and the fabrication of QLEDs, respectively. Author contributions X.P. and Y.J. conceived the idea, supervised the work, and wrote the manuscript. C.P. participated in conceiving the idea, developed the surface treatment processes for all QDs, and conducted voltammetric measurements. X.D. carried out the fabrication and characterizations of QLEDs and single-carrier devices. Y.S. developed the CdSe/CdS/ZnS core/shell/shell QDs and worked on surface modification of all QDs. M.Z. carried out the single-dot photoluminescence experiments. Y.D. calculated the out-coupling efficiency of the QLEDs. All authors discussed the results and commented on the manuscript. Data Availability The data that support the finding of this study are available from the corresponding authors upon reasonable request. 19 Competing interests The authors declare no competing interests. Additional Information Reprints and permissions information is available at www.nature.com/reprints. Readers are welcome to comment to the online version of the paper. Correspondence and requests for materials should be addressed Y.J. ([email protected]) or X.P. ([email protected]). 20 Figure 1 Optical properties and QLED performance of the CdSe/CdS-Cd(RCOO)2 QDs and the CdSe/CdS-RNH2 QDs. a, Ligand exchange from cadmium-carboxylates (with a small amount of negatively-charged carboxylates) to primary amines. b, Absorption and steady-state photoluminescence (PL) spectra. c, Time-resolved PL spectra with mono-exponential PL decay lifetime (τ) and quantum yield (QY). d, QLED structure: indium tin oxide (ITO)/ poly(ethylenedioxythiophene):polystyrene sulphonate (PEDOT:PSS, ~35 nm)/poly (N,N9-bis(4- butylphenyl)-N,N9-bis(phenyl)-benzidine) (poly-TPD, ~30 nm)/poly(9-vinlycarbazole) (PVK, ~5 nm)/QDs (~40 nm)/Zn0.9Mg0.1O nanocrystals (~60 nm)/Ag. e, Current density and luminance versus 21 driving voltage characteristics of the QLEDs. f, EQE versus driving voltage characteristics of the QLEDs. g, Stability data of the QLEDs driven at a constant current density of 100 mA cm-2. 22 Figure 2 Operando electrochemical reduction of the CdSe/CdS-Cd(RCOO)2 QDs. a, Relative PL efficiency of the QDs in the QLEDs as a function of the driving voltage. b, Relative PL efficiency of the CdSe/CdS-Cd(RCOO)2 QDs in a QLED (driven at a constant voltage of 2 V for 50 hours, followed by 0 V for 10 hours). Insets are the corresponding photographs of the devices under ultraviolet irradiation (scale bar: 0.5 mm). c, Relative PL efficiency of the QDs in electron-only devices (ITO/2,2',2''-(1,3,5-Benzinetriyl)-tris(1-phenyl-1-H-benzimidazole) (TPBi, ~25 nm)/QDs (~40 nm)/Zn0.9Mg0.1O (~60 nm)/Ag, driven at a constant current density of 10 mA cm-2). Inset, the relative PL efficiency of the QDs in hole-only devices (ITO/PEDOT:PSS (~35 nm)/poly-TPD (~30 nm)/PVK 23 (~5nm)/QDs (~40 nm)/4,4'-Bis(N-carbazolyl)-1,1'-biphenyl (CBP, ~25 nm)/MoOx/Au), driven at a constant current density of 30 mA cm-2, sharing the same coordinate axes with the main plot. d, The voltammetric curves of the two types of CdSe/CdS core/shell QDs and cadmium oleate. e and f, PL and differential auger electron spectra of the CdSe/CdS-Cd(RCOO)2 QDs in QLEDs under different conditions. The top electrodes are delaminated by adhesive tapes and the oxide electron-transporting layers are etched by a dilute acetonitrile solution of acetic acid, a known inert solution for the QD layer. 24 Figure 3 Cadmium-carboxylate ligand concentration-dependent PL and EL properties of the CdSe/CdS-Cd(RCOO)2 QDs. a, FTIR spectra of the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs (red) and the completely surface-treated CdSe/CdS-RNH2 QDs (black). Vibration bands of RNH2 and Cd(RCOO)2 are labelled as νN-H and νCOO-, respectively. b, Residual carboxylate percentages and PL QY of the CdSe/CdS core/shell QDs in solution versus different concentration of primary amine used in the ligand-exchange procedures. c, PL QY of the QD films, IQE of the QLEDs, operational lifetime of the QLEDs, and voltage gap of the QLEDs versus the residual percentage of carboxylate ligands identified by FTIR. 25 Figure 4 Electrochemical stability and QLED performance of the red-emitting QDs with various types of ligands. a, Relative PL efficiency of the CdSe/CdS/ZnS core/shell/shell QDs in electron-only devices (ITO/TPBi (~25 nm)/QDs (~40 nm)/Zn0.9Mg0.1O (~60 nm)/Ag) driven at a constant current density of 100 mA cm-2. Inset, the relative PL efficiency of the QDs in hole-only devices (ITO/PEDOT:PSS (~35 nm)/poly-TPD (~30 nm)/PVK (~5nm)/QDs (~40 nm)/CBP (~25 nm)/MoOx/Au) driven at a constant current density of 30 mA cm-2, sharing the same coordinate axes with the main plot. b, Relative PL efficiency of the CdSe/CdS/ZnS core/shell/shell QDs in QLEDs driven at a constant current density of 100 mA cm-2. c, EQE versus driving voltage characteristics and d, stability data (driven at a constant current of 100 mA cm-2) of the QLEDs based on the 26 CdSe/CdS/ZnS core/shell/shell QDs. e, The IQE/PL QY ratio (IQE of the QLED divided by PL QY of the QD film) for the CdSe/CdS core/shell and the CdSe/CdS/ZnS core/shell/shell QDs with different ligands. 27 Figure 5 Electrochemical stability and QLED performance of the blue-emitting CdSeS/ZnSeS/ZnS core/shell/shell QDs. A, Relative PL efficiency of the QDs in the electron-only devices (ITO/TPBi (~25 nm)/QDs (~20 nm)/Zn0.9Mg0.1O (~50 nm)/Al) driven at a constant current density of 100 mA cm-2. Inset, the relative PL efficiency of the QDs in the hole-only devices (ITO/PEDOT:PSS (~35 nm)/poly-TPD (~30 nm)/PVK (~5nm)/QDs (~20 nm)/CBP (~25 nm)/MoOx/Au) driven at a constant current density of 30 mA cm-2, which shares the same coordinate axes with the main plot. b, Relative PL efficiency of the QDs in the QLEDs (ITO/PEDOT:PSS (~35 nm)/poly-TPD (~30 nm)/PVK (~5 nm)/QDs (~20 nm)/Zn0.9Mg0.1O nanocrystals (~50 nm)/Al, driven at a constant current density of 100 mA cm-2). c, EQE versus driving voltage of the blue-emitting QLEDs. Insets, PL and EL spectra (top-left) and a photograph of a working QLED (bottom-right), respectively. d, Stability data of the blue-emitting QLEDs driven at the constant-current mode. Inset: expanded plot for the QLEDs with the CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs. 28 Table 1 PL and EL performance of the CdSe/CdS QDs, CdSe/CdS/ZnS QDs and CdSe/CdZnSe/CdZnS QDs with different ligands. The data with gray color is from QDs with electrochemically reactive ligands and the data with black color is from QDs with electrochemically inert ligands. a The device lifetimes are estimated at an initial brightness of 100 nits if not noted. b These QDs are determined to retain their negatively-charged carboxylate ligands (~15% of residual carboxylate ligands). c The percentage of carboxylate ligands of these QDs after ligand exchange by TOP was ~25%. This value is the lower limit for maintaining the dispersity of these QDs after TOP treatment. 29 Inorganic structure Ligands Film PL QY (%) EQE (%) Device Lifetimea T50 (h) CdSe/CdS Cd(RCOO)2 77 ± 2 0.25 ± 0.08 ~ 0.3 Cd(RPOOO) 70 ± 3 0.1-0.3 ~ 1 Zn(RCOO)2 75 ± 2 2-4 ~ 5-7 Mg(RCOO)2 42 ± 3 8-9 ~ 8000 RNH2 + RSH 77 ± 2 18-20 ~ 90,000 RNH2b 76 ± 2 18.5 ± 0.6 ~ 90,000 RNH2 76 ± 2 18.6 ± 0.6 ~ 90,000 CdSe/CdS/ ZnS Zn(RCOO)2 80 ± 2 2.7 ± 0.5 ~ 3-5 TOPc 79 ± 2 15.4 ± 0.6 T95 @1000 nits ~ 400 RNH2 + RSHb 78 ± 2 17-19 T95 @1000 nits ~ 550 RNH2b 79 ± 2 17.8 ± 0.4 T95 @1000 nits ~ 600 CdSe/ CdZnSe/ CdZnS Zn(RCOO)2 74 ± 2 5-7 T95 @1000 nits ~ 80 TOPc 75 ± 2 15.8 ± 0.9 T95 @1000 nits ~ 3800 Methods Materials. Poly-TPD (average molecular weight, ~60000 g mol-1) was purchased from Xi'an polymer light technology Corp. PVK (average molecular weight, 25,000 -- 50,000 g mol-1), sulfur (99.98%), zinc acetate dihydrate (98%), 2-ethylhexanethiol (97%) and oleylamine (70%) were purchased from Sigma Aldrich. Octylamine (98%), tri-octylphosphine (TOP, 97%), cadmium acetate (98%), magnesium acetate hydrate (98%), benzoyl peroxide (BPO, 97%, dry wt.), 1-octadecene (ODE, 90%) and oleic acid (90%) were purchased from Alfa-Aesar. Chlorobenzene (extra dry, 99.8%), m-xylene (extra dry, 99%), octane (extra dry, >99%), ethanol (extra dry, 99.5%), tetrahydrofuran (THF, 99.9%, anhydrous) and decanoic acid (99%) were purchased from Acros. Dimethyl sulphoxide (DMSO, HPLC grade) and ethyl acetate (HPLC grade) were purchased from J&K Chemical Ltd. 1-Ethyl-3-methylimidazolium Bis(trifluoromethanesulfonyl)imide (EMI-TFSI, 98%) and Ferrocene (98%) were purchased from Energy Chemical. Red CdSe/CdZnSe/CdZnS core/shell/shell QDs and blue CdSeS/ZnSeS/ZnS core/shell/shell QDs were purchased from Najing technology Co., Ltd. Syntheses of QDs and oxide nanocrystals. The CdSe/CdS core/shell QDs (ten monolayers of CdS shell) were synthesized according to ref. 14 with some modifications. Briefly, core CdSe dots (first exciton peak: 550 nm) were used for shell growth. ODE (4 mL), oleic acid (0.85 mL), decanoic acid (0.15 mL), and cadmium acetate (1 mmol) were mixed and degassed at 150 oC for 10 min, followed by injecting of a hexane solution (1 mL) containing the CdSe core QDs. The mixture was further degassed for 10 min. Then the temperature of the solution was raised to 260 oC. A solution of sulfur (0.5 mmol) in ODE (5 mL) was dropwisely introduced into the reaction flask at a rate of 4 mL/h. After 60 min, the reaction was stopped. The QDs were precipitated by adding acetone and then purified 30 twice by using toluene and methanol for dissolution and precipitation, respectively. The resulting CdSe/CdS core/shell QDs, i.e., CdSe/CdS-Cd(RCOO)2 QDs, were dissolved in toluene. For the synthesis of CdSe/CdS/ZnS core/shell/shell QDs (five monolayers of CdS shell and three monolayers of ZnS shell), the CdSe/CdS core/shell QDs (five monolayers of CdS shell) were synthesized and precipitated following the above procedures. For the growth of ZnS shell, ODE (4 mL), oleic acid (0.4 mL), decanoic acid (0.1 mL), and zinc acetate (0.5 mmol) were mixed and degassed at 150 oC for 10 min, followed by injecting of a hexane solution (1 mL) containing CdSe/CdS QDs. The mixture was further degassed for 10 min. Then the temperature of the solution was raised to 300 oC. A solution of sulfur (0.5 mmol) in ODE (5 mL) was dropwisely introduced into the reaction flask at a rate of 4 mL/h. After 30 min, the reaction was stopped. The QDs were precipitated by adding acetone and then purified twice by using toluene and methanol for dissolution and precipitation, respectively. The resulting CdSe/CdS/ZnS core/shell/shell QDs, i.e., CdSe/CdS/ZnS-Zn(RCOO)2 QDs were dissolved in toluene. The Zn0.9Mg0.1O nanocrystals were synthesized according to ref. 9. For a typical synthesis, a DMSO solution (30 mL) of magnesium acetate hydrate (0.3 mmol) and zinc acetate hydrate (2.7 mmol) mixed dropwise with an ethanol solution (10 mL) of TMAH (5 mmol) and stirred for 1 h in ambient conditions. The resulting Zn0.9Mg0.1O nanocrystals were precipitated by adding ethyl acetate and dispersed in ethanol. The ethanol solution of Zn0.9Mg0.1O nanocrystals (~ 30 mg mL-1) was filtered with 0.22 μm PTFE filter before use. Surface modification of the QDs. The as-synthesized CdSe/CdS-Cd(RCOO)2 QDs were dissolved in a mixture of oleylamine (2 mL), sulfur (0.1 mmol) and ODE (1 mL). The temperature was raised to 31 110 oC and maintained for 20 min, followed by precipitation of QDs by adding methanol (5 mL). The procedures were repeated to ensure complete conversion of the surface ligands. The precipitated QDs were dissolved in ODE (2 mL) and oleylamine (2 mL) and then annealed at 240 oC for 20 min. The resulting CdSe/CdS-RNH2 QDs were precipitated by adding methanol. Regarding the ligand exchange of CdSe/CdS-Cd(RCOO)2 QDs by amine, the as-synthesized CdSe/CdS-Cd(RCOO)2 QDs were dispersed in a mixture of oleylamine, octylamine and ODE, with a total volume of 6 mL. The volume ratio of oleylamine and octylamine was maintained at 3: 2 and the total concentrations of the amines were varied from 0-4 mol/L. The mixture was heated up to 150 oC and annealed for 20 min. Then the solution was cooled. The QDs were precipitated by adding methanol (12 mL). Next, the QDs were reprecipitated by using 1 mL hexane and 8 mL acetonitrile. Regarding the ligand exchange of CdSe/CdS/ZnS-Zn(RCOO)2 QDs by amine, the as-synthesized QDs were dispersed in a mixture of oleylamine (3.6 mL) and octylamine (2.4 mL). The mixture was heated to 50 oC and annealed for 20 min. Then the solution was cooled. The QDs were first precipitated by adding methanol (12 mL) and then reprecipitated by using 1 mL hexane and 8 mL acetonitrile. The number of the ligand-exchange cycle was repeated four times, leading to CdSe/CdS/ZnS-RNH2 QDs. Regarding the ligand exchange of CdSeS/ZnSeS/ZnS-Zn(RCOO)2 QDs by amine, the as- synthesized QDs were dispersed in a mixture of oleylamine (1 mL) and ODE (2 mL). The mixture was heated to 220 oC and annealed for 10 min. Then the solution was cooled, leading to CdSeS/ZnSeS/ZnS- RNH2 QDs. The QDs were precipitated by adding methanol (6 mL). Regarding the CdSe/CdS/ZnS-TOP QDs, the ligand-exchange procedure was similar to that of amine for CdSe/CdS/ZnS-Zn(RCOO)2 QDs, except that pure TOP (6 mL) was used instead of 32 oleylamine and octylamine. Regarding the CdSe/CdZnSe/CdZnS-TOP QDs, the ligand-exchange procedure was similar to that for CdSe/CdS/ZnS-TOP QDs, except that the CdSe/CdZnSe/CdZnS-Zn(RCOO)2 QDs were used. The ligand-exchange procedure with 2-ethylhexanethiol was performed by adding 1 mL thiol into 1 mL solution of QDs (either the CdSe/CdS-RNH2 QDs or the CdSe/CdS/ZnS-RNH2 QDs) and then stirred for 1h in N2 atmosphere at 70 oC. After cooling of the mixture, the QDs were precipitated by adding ethanol and re-dispersed in octane. Regarding the CdSe/CdS-Cd(RPOOO) QDs, the CdSe/CdS-Cd(RCOO)2 QDs were dispersed in toluene (2 mL). Octylphosphonic acid (0.25 mmol) was added into the mixture. The solution was heated to 50 oC and maintained for 1 h. Then ethanol was added to precipitate the QDs. Regarding the CdSe/CdS-Zn(RCOO)2 QDs, ODE (4 mL), oleic acid (0.4 mL), decanoic acid (0.1 mL), and zinc acetate (0.5 mmol) were mixed and degassed at 150 oC for 10 min, followed by injecting of a hexane solution (1 mL) containing CdSe/CdS-RNH2 QDs. The mixture was further degassed for 10 min. Then the temperature of the solution was raised to 200 oC. After annealed for 10 min, the reaction was stopped. The QDs were precipitated by adding acetone and then purified twice by using toluene and methanol for dissolution and precipitation, respectively. The resulting CdSe/CdS- Zn(RCOO)2 QDs were dissolved in toluene. Regarding the CdSe/CdS-Mg(RCOO)2 QDs, the procedure was similar to that for CdSe/CdS- Zn(RCOO)2 QDs except that zinc acetate was replaced by magnesium acetate. The relative amount of the residue carboxylate ligands of QDs were determined by Fourier 33 transform infrared spectra. The samples were prepared by dispersing the QDs in dodecane, followed by adding hydrochloric acid. The mixture was vortexed for 10 min to convert the carboxylate ligands (including both metal carboxylate sales and carboxylate ions) to carboxylic acid. Next, the mixture was centrifuged for 2 min at 4000 rpm. The concentration of carboxylic acid in the supernatant was determined by a standard curve (Supplementary Fig. 6a). Device fabrication. The QLEDs were fabricated by depositing materials onto ITO coated glass substrates (~1.1 mm in thickness, sheet resistance: ~20 Ω sq-1). PEDOT:PSS solutions (Baytron P VP Al 4083, filtered through a 0.22 μm N66 filter) were spin-coated at 3000 rpm for 60 s and then baked at 150 °C for 15 min. The PEDOT:PSS-coated substrates were subjected to an oxygen plasma treatment for 5 min and then transferred into a nitrogen-filed glove box (O2 < 1 ppm, H2O < 1 ppm). Poly-TPD (in chlorobenzene, 8 mg mL-1), PVK (in m-xylene, 1.5 mg mL-1), QDs (in octane, ~20 mg mL-1 for the CdSe/CdS core/shell red dots, ~15 mg mL-1 for the CdSe/CdS/ZnS core/shell/shell red dots and the CdSe/CdZnSe/CdZnS core/shell/shell red dots, ~12 mg mL-1 for the CdSeS/ZnSeS/ZnS core/shell/shell blue dots), and Zn0.9Mg0.1O nanocrystals (in ethanol, 30 mg mL-1) were layer-by-layer deposited by spin coating at 2000 rpm for 45 s. The poly-TPD and PVK layers were baked at 130 °C for 20 min and at 150 °C for 30 min, respectively, before deposition of the next layer. Finally, metal electrodes (100 nm) were deposited using a thermal evaporator (Trovato 300C, base pressure: ~2×10- 7 torr) through a shadow mask. The device area was 4 mm2 as defined by the overlapping area of the ITO and metal electrodes. All devices were encapsulated in a glove-box using commercially available ultraviolet-curable resin. Characterizations. The absorption spectra were obtained by using an Agilent Cary 5000 34 spectrophotometer. The photoluminescence spectra were obtained by using an Edinburgh Instruments FLS920 fluorescence spectrometer. The time-resolved fluorescence spectra were measured by the time-correlated single-photon counting method using an Edinburgh Instruments FLS920 spectrometer. The samples were excited by a 405 nm pulsed diode laser (EPL-405). The absolute photoluminescence quantum efficiencies of the QDs were measured by applying a two-step (for solutions) method or a three-step (for thin films) method50. A system consisting of a Xenon lamp, optical fibre, a QE65000 spectrometer (Ocean Optics) and a home-designed integrating sphere was used. Transmission electron microscope observations were carried out using Hitachi 7700 operated at 80 keV. Fourier transform infrared spectroscopy analyses were conducted using a Nicolet 380 spectrometer. Energy-dispersive X-ray spectroscopy analyses of the QDs were carried out on an Ultral 55 scan transmission electron microscope. Auger electron spectroscopy measurements were performed using a PHI 710 Scanning Auger Microscopy. The thicknesses of the films were measured using a KLA Tencor P-7 Stylus Profiler. Single-dot photoluminescence experiments were conducted on a home-built epi-illumination fluorescence microscope system equipped with a Zeiss 63× oil immersion objective (N.A.:1.46). Samples were excited by a 405 nm ps pulse laser with 1 MHz repetition for photoluminescence measurement. The photoluminescence intensity trajectories of single dots were recorded by an electron multiplier charge-coupled device (EMCCD, Andor iXon3). The exposure time per frame was 30 ms. Voltammetric measurements were carried out by using a CHI600D electrochemical workstation located in a glove box. Typically, a solution of THF (1 mL) containing EMI-TFSI (0.1 mmol) was used as an electrolyte and a solution of Ferrocene in THF (0.1 M) was added as a reference. Silver 35 wire and platinum coil were used as quasi-reference and counter electrodes, respectively. Solutions (0.1 mL) of QDs in toluene was added for measurements. The scan speed was 20 mV s−1. All QLEDs were characterized under ambient conditions (room temperature: 22~24 °C and relative humidity: 40~60%). A system consisting of a Keithley 2400 source meter and an integration sphere (FOIS-1) coupled with a QE-Pro spectrometer (Ocean Optics) was used to measure the current density-luminance-voltage curves. The electroluminescence characteristics of the QLEDs were cross- checked at Cavendish Laboratory (Richard Friend group) and Nanjing Tech University (Jianpu Wang group). The operational lifetimes of the QLEDs were measured by using an aging system designed by Guangzhou New Vision Opto-Electronic Technology Co., Ltd. Photoluminescence intensity changes of the QDs in working devices were measured via a home- build system controlled by Labview. The system (Supplementary Fig. 4) consists of lock-in amplifiers (SR830), source-meter (Keithley 2400), electric-meter (Keithley 2000), and photodetectors (Thorlabs PDA100A). The devices were excited by a cw 445 nm (or 405 nm) laser with an optical chopper for frequency modulation (933 Hz). A Keithley 2400 source-meter was used to electrically drive the devices. Lock-in amplifiers combined with photodetectors were used to independently measure the intensity of photoluminescence from QDs and excitation light. The intensity of excitation light was kept to be low to ensure that the photo-excitation do not interfere with the operation of the devices. 50. de Mello, J. C., Wittmann, H. F. & Friend, R. H. An improved experimental determination of external PL quantum efficiency. Adv. Mater. 9, 230-232 (1997). 36 Figure S1 Chemical and structural characterizations of the CdSe/CdS-Cd(RCOO)2 QDs and the CdSe/CdS-RNH2 QDs. a, Atomic ratios of Cd and (Se + S) of the QDs, confirming that the ratio of cadmium to anions (Se+S) for the CdSe/CdS-RNH2 QDs is stoichiometric within experimental error and the CdSe/CdS-Cd(RCOO)2 QDs possess excess cadmium ions. b and c, Transmission electron microscope images of the QDs. Scale bar: 50 nm. 37 Figure S2 Energy level diagram of several bulk semiconductors and standard reduction potentials of Cd2+/Cd0, Zn2+/Zn0 and Mg2+/Mg0. 38 Figure S3 Photoluminescence properties of the CdSe/CdS-Cd(RCOO)2 QDs and the CdSe/CdS- RNH2 QDs. a and b, Photoluminescence intensity-time traces of single CdSe/CdS-Cd(RCOO)2 and CdSe/CdS-RNH2 QD, respectively, indicating non-blinking characteristics. c and d, Stable and efficient photoluminescence of two types of QDs in thin films. e and f, The corresponding time- resolved photoluminescence decay curves for both types of QDs in thin films. 39 Figure S4 Experimental Setup for the simultaneous photoluminescence-electroluminescence measurements. The equipment operando monitors the relative changes of photoluminescence efficiency of QDs in working devices by using low-frequency chopped, low-intensity photo-excitation. 40 Figure S5 Differential auger electron spectra for the CdSe/CdS-RNH2 QDs (biased in QLEDs) and two control samples of CdS nanocrystals and metal Cd foils. 41 Figure S6 Ligand composition, photoluminescence, and electroluminescence of the CdSe/CdS core/shell QDs with different amounts of residual carboxylate ligands. a, FTIR calibration curve for measuring the concentration of fatty acids by using the absorbance of the C=O vibration of fatty acids. b, FTIR spectra of the fatty acids obtained by digesting the QDs with hydrochloric acid (see experimental for details). c, d, and e, J-L-V, EQE-V and stability data of the QLEDs based on the QDs with different percentages of residual carboxylates. 42 Figure S7 Characteristics of the red-emitting QLEDs based on the QDs with different ligands. Photoluminescence QY of thin films is provided for each sample in the plot. a and b, CdSe/CdS core/shell QDs with cadmium phosphonate ligands (Cd(RPOOO)) or cadmium thiolate ligands (Cd(RS)2). c and d, CdSe/CdS core/shell QDs with mixed ligands of amine and TOP (RNH2 + TOP) or amine and thiol (RNH2 + RSH). e and f, CdSe/CdS/ZnS core/shell/shell QDs with TOP ligands or mixed ligands of amine and thiol (RNH2 + RSH). g and h, CdSe/CdS core/shell QDs with zinc- carboxylate (Zn(RCOO)2) ligands or magnesium-carboxylate (Mg(RCOO)2) ligands. 43 Figure S8 Photoluminescence and electroluminescence properties of the CdSe/CdZnSe/CdZnS- Zn(RCOO)2 QDs and the CdSe/CdZnSe/CdZnS-TOP QDs. a, Transient photoluminescence spectra. b-d, J-L-V, EQE-V and stability data of the QLEDs based on the two types of QDs. 44 Figure S9 Optical properties and electroluminescence behaviours of the blue-emitting CdSeS/ZnSeS/ZnS core/shell/shell QDs. a, Absorption and steady-state photoluminescence spectra. b, Transient photoluminescence spectra of the QDs in solution. The photoluminescence QY of QDs in film are also shown. c, Operational lifetime of the blue QLEDs using the CdSeS/ZnSeS/ZnS-RNH2 QDs measured at different initial brightness. These devices were tested under a constant-current mode. The data are fitted by an empirical equation, L0 n × T50 = constant, leading to an acceleration factor of 1.88. 45 Table S1 Comparison of the state-of-the-art QLEDs in literature. QD Ligands in the structure synthetic system CdSe/CdS Decylamine & Dodecanethiol EL peak (nm) 640 CdSe/CdZnS Oleic acid & TOP 624 Device structure ITO/PEDOT:PSS/P-TPD/ PVK/QDs/PMMA/ZnO/Ag ITO/PEDOT:PSS/P-TPD/ PVK/QDs/ZnMgO/Ag Turn-on Peak voltage (V) EQE @1 cd m-2 (%) Lifetime (h) Ref. T50@ 100 cd m-2 1.7 20.5 100,000 9 1.7 18.2 190,000 16 Oleic acid & TOP & ITO/ZnO/QDs/CPB/ 658 1.9 11.36 ______ 42 1-dodecanethiol MoOx/Al CdSe/CdS CdSe/Cd1-x ZnxSe1-ySy/ Oleic acid & TOP 631 ITO/PEDOT/TFB/QD/ZnO/Al 1.7 15.1 2,200,000 ZnS CdSe/Cd1-x ITO/ZnO-PVP/QDs/TCTA/ ZnxSe/ZnSey Oleic acid & TOP 611 1.9 13.5 1,330,000 Oleic acid & TOPO 537 MoOx/Al ITO/PEDOT:PSS /TFB/ QDs/ZnO/Al ITO/ZnO/QDs/PEIE/P-TPD/ 2.3 14.5 90,000 Oleic acid & TOP ~520 ~3.2 15.6 ______ 43 S1-y CdSe@ZnSe/ ZnS CdSe@ZnS/ ZnS ZnxCd1−xSy Se1−y CdZnSeS@ Oleic acid & TOP 526 TOP & ~540 MoOx/Al ITO/PEDOT:PSS/TFB/ QDs/ZnO/Al ITO/PEDOT/TFB/QDs/ ZnS 1-dodecanethiol ZnMgO/Al ZnCdSe/ZnS// Trioctylamine & ITO/PEDOT/TFB/QDs/ ZnS TOP Cd1− xZnxS/ Oleic acid 479 467 PMMA/ZnO/Al ITO/PEDOT/TFB/QDs/ ZnS & TBP ZnO/Al 15 17 10 2.2 21 T90 @ 2500 nits 2.0 16.6 490 _____ 2.4 16.2 355 ~2.4 ~7 T50 @ 1000 nits 23 44 46 45 47 46
1806.03381	1	1806	2018-06-08T23:39:05	Sub-picosecond photon-efficient 3D imaging using single-photon sensors	[ "physics.app-ph" ]	Active 3D imaging systems have broad applications across disciplines, including biological imaging, remote sensing and robotics. Applications in these domains require fast acquisition times, high timing resolution, and high detection sensitivity. Single-photon avalanche diodes (SPADs) have emerged as one of the most promising detector technologies to achieve all of these requirements. However, these detectors are plagued by measurement distortions known as pileup, which fundamentally limit their precision. In this work, we develop a probabilistic image formation model that accurately models pileup. We devise inverse methods to efficiently and robustly estimate scene depth and reflectance from recorded photon counts using the proposed model along with statistical priors. With this algorithm, we not only demonstrate improvements to timing accuracy by more than an order of magnitude compared to the state-of-the-art, but this approach is also the first to facilitate sub-picosecond-accurate, photon-efficient 3D imaging in practical scenarios where widely-varying photon counts are observed.	physics.app-ph	physics	Sub-picosecond photon-efficient 3D imaging using single-photon sensors Felix Heide, Steven Diamond, David B. Lindell, Gordon Wetzstein Stanford University, Department of Electrical Engineering Active 3D imaging systems have broad applications across disciplines, including biological imaging, remote sensing and robotics. Applications in these domains require fast acquisition times, high timing resolution, and high detection sensitivity. Single-photon avalanche diodes (SPADs) have emerged as one of the most promising detector technologies to achieve all of these requirements. However, these detectors are plagued by measurement distortions known as pileup, which fundamentally limit their precision. In this work, we develop a probabilistic image formation model that accurately models pileup. We devise inverse methods to effi- ciently and robustly estimate scene depth and reflectance from recorded photon counts using the proposed model along with statistical priors. With this algorithm, we not only demon- strate improvements to timing accuracy by more than an order of magnitude compared to the state-of-the-art, but this approach is also the first to facilitate sub-picosecond-accurate, photon-efficient 3D imaging in practical scenarios where widely-varying photon counts are observed. Active 3D imaging has broad applications across disciplines, including remote sensing, non-line-of-sight imaging, autonomous driving, robotics, and microscopic imaging of biological samples 1–11. Key requirements for most of these applications include high timing accuracy, fast acquisition rates, wide dynamic operating ranges, and high detection sensitivity. In particular, remote sensing and automotive applications 1–4, 6 demand acquisition ranges from < 1 meter to kilometers, while non-line-of-sight imaging 7–9, for example, relies on the information encoded by the few returning photons of multiply scattered indirect light, in addition to the directly reflected light. To facilitate these applications, ultra-sensitive detectors have been developed that allow for individual photons returning from a pulsed illumination source to be recorded. The most sensit- ive time-resolved detectors to date are single-photon avalanche diodes (SPAD), which have been rapidly established as a core detector technology for 3D imaging 2, 12–18. SPADs are reverse-biased photodiodes operated above their breakdown voltage, i.e. in Gei- ger mode 12. Photons incident on the active surface of the SPAD can trigger an electron avalanche, which is subsequently time-stamped. By repeatedly timestamping photons returning from a syn- chronously pulsed illumination source, typically operating at MHz rates, a histogram of photon counts over time can be accumulated. The histogram approximates the intensity of the returning light pulse and characterizes the propagation delay, enabling distance, reflectance and 3D geometry to be recovered. While a SPAD can be operated in a free-running mode 19, which allows photon events to be detected at all arrival times simultaneously, some modes of operation enable gated detection wherein only photons within a specified time window between pulses are detected 2 0. In 1 this gated mode 21, 22, they achieve high accuracy and robustness to ambient light by electronically gating out all but a narrow temporal slice 20. However, gated acquisition requires a sequential sweep of the temporal slice over the full target range, which restricts it to applications that do not require fast acquisition rates. SPAD detectors operated in free-running mode do not suffer from this limitation, but whether these sensors are used for 3D imaging, microscopy, non-line-of-sight imaging, or any other application, they are subject to a fundamental phenomenon which severely limits accuracy: pileup distortions 23, 24. Pileup is a limitation inherent to the SPAD detector's working principle. After every triggered electron avalanche, the detector needs to be quenched before it is able to detect further photon arrival events. During this dead time, which is typically in the order of tens to hundreds of ns, the detector is inactive. Thus, earlier photons of a single laser pulse are more likely to trigger an avalanche while later ones are more likely to fall into the dead time, thus being ignored. This causes severe measurement skew, which is known as pileup. Pileup cannot be corrected in hardware and results in round-trip time-stamping errors that are orders of magnitudes larger than the timing precision of the detector. To avoid pileup distortion, active imaging systems can be operated in a low-flux regime, where the average number of incident photons per measurement is extremely low (i.e., (cid:28) 1). State- of-the-art depth and reflectivity estimation techniques have demonstrated successful operation in these challenging conditions using probabilistic image formation models and statistical priors in the inverse methods 18, 25, 26. However, many 3D imaging applications, for example in robotics, biological imaging, or automotive sensing, operate in environments where both objects returning a low number of signal photons and objects reflecting higher numbers of photons are essential for decision making. The large variance in acquired photon counts of objects at different depths or resulting from varying reflectivity of different objects makes it crucial for a practical 3D imaging technique to operate robustly in both low-flux and high-flux conditions simultaneously (see Fig. 1). In this work, we introduce a new estimation algorithm that overcomes existing limitations of active 3D imaging systems with free-running SPADs. Specifically, we demonstrate sub-picosecond timing accuracy for pulsed light sources with a full width at half maximum (FWHM) wider than 50 ps. The proposed method improves the accuracy of existing depth and albedo estimation al- gorithms by more than an order of magnitude across a wide dynamic range, from low-flux to high- flux measurements. These benefits are enabled by an image formation model and corresponding inverse method that lift assumptions of low-flux models 18, 25 to broad operating conditions that are distorted by pileup. To this end, we introduce a probabilistic image formation model that includes pileup and we derive efficient inverse methods for the depth and albedo estimation problem with this model. The inverse methods exploit statistical priors of both depth and albedo. Unlike pre- vious work, our reconstruction framework jointly estimates all unknown parameters, overcoming algorithmic limitations that restrict the timing precision of existing methods. The proposed meth- ods not only facilitate highly accurate and fast 3D imaging but they also open up a new operating regime of photon-efficient 3D imaging in conditions with drastically-varying photon counts. 2 Results Imaging setup. The experimental acquisition setup is illustrated in Fig. 1. The light source is a pulsed, collimated laser that is scanned horizontally and vertically using a 2-axis mirror gal- vanometer. Light scattered back from the scene is focused on the SPAD (Micro Photon Devices PD-100-CTD) using a microscope objective. A PicoQuant PicoHarp 300 module is used for time- stamping, histogram accumulation, and synchronization of laser and detector. The dead time of the SPAD detector is 75 ns, corresponding to a path length of 22.5 m. Due to the fact that this dead time is multiple orders of magnitude larger than the desired timing accuracy, the pileup distortions of measured histograms can only be ignored in the ultra-low-flux regime with < 10−2 average photons detected per pulse. Observation model. We discretize the scene into a grid of m × n points at distances z ∈ Rm×n and denote the reflectance of these points as α ∈ [0, 1]m×n. Each point is probed with N pulses. We first describe the observation model for a single point (i, j) by modeling the temporal shape of the laser pulse g as a mixture of Gaussians g(t) = K X k=1 ak exp(cid:18)− (t − bk)2 c2 k (cid:19) , (1) where the parameters ai, bi, ci ∈ R are calibrated in a pre-processing step (cf. Methods). Given this parametric impulse model, the photon detections on the SPAD detector can be modeled as an inhomogeneous Poisson process with rate function rij(t, αij, zij, s) = µαij g(t − 2zij/c) + s, (2) where c is the speed of light, µ is the SPAD photon detection probability, and s is background detections from ambient light and dark count 18. The free-running SPAD records photon detections over a time interval [0, T ] discretized into uniform bins of size ∆, where T is the inverse of the laser repetition rate. Let hijk denote the ac- cumulated counts in bin k over the N pulses and λijk(αij, zij, s) = R (k+1)∆ rij(t, αij, zij, s)dt the Poisson rate for the number of detections in each bin k. The proposed probabilistic re- construction method infers the latent variables z, α using maximum-a-posteriori (MAP) estim- ation, which relies on the probability P (hijλij) of observing a histogram hij ∈ ZT given means λij(αij, zij, s) ∈ RT and priors on depth and albedo. k∆ Prior work on 3D imaging using SPAD detectors focuses on the low-flux regime18, 25, 27, 28 in which the expected number of photon detections per pulse is significantly smaller than one. In the low-flux regime we may neglect dead time and approximate the observations hijk as being conditionally independent across bins k, with corresponding Poisson probability mass function 3 P (hijkλij). For medium and high-flux measurements the conditional independence approxim- ation breaks down because dead time ensures that at most one photon detection is recorded per pulse 23. The probability of a single histogram bin, taking dead time into account, is given by the multinomial distribution , ( λijk!!hij' ' X k=1 T '=1 exp − Y '−1Xk=1 λijk! − exp − P (hijλij) = N ! exp(−1T λij)N−1T hij hij1!··· hijT !(N − 1T hij)! 3) where 1 is the vector of all ones. We refer to the Supplemental Methods for a detailed derivation of this probabilistic model. The observation model holds across the full measurement range, from low-flux to high-flux regimes. Reconstruction algorithm. After scanning a scene, temporal histograms hij are available for each point (i, j). To reconstruct scene reflectance and depth from these histograms, we find the MAP estimate of α and z, along with the ambient term s, using the observation model and the prior assumption that the gradients of reflectivity and depth maps are sparse. Inspired by prior work on depth and reflectivity estimation 18, 25, 26, we place transverse anisotropic total variation (TV) priors directly on α and z. The MAP estimate is given by solving the optimization problem α,z,s Pm i=1Pn minimize j=1 − log P (hijλij(αij, zij, s)) + γ1k∇αk1 + γ2k∇zk1, (4) where α and z are the unknown variables, and ∇ is the gradient operator. To solve this prob- lem, we develop a proximal algorithm for this non-convex optimization problem that decouples the sparsity-promoting prior terms from the likelihood term. Specifically, we minimize the joint objective by introducing slack variables vα, vz, vs for albedo, depth and ambient terms. We then optimize each unknown term in an alternating fashion (see Supplemental Material for details). As a result of this proximal optimization scheme, the log-likelihood minimization becomes separable in the measurement ij as (5) (s − vs)2, ξ 2 (α − vα)2 + ξ 2 (z − vz)2 + ξ 2 minimize α,z,s − log P (hijλij(α, z, s)) + which is an optimization problem over a non-linear tri-variate loss function and quadratic proximal closeness terms with scalar weight ξ. We solve the minimization problem with per-pixel parallel Newton Methods, as derived in the Supplemental Methods. This reconstruction method improves prior SPAD depth imaging techniques both by account- ing for dead time in the observation model, which is crucial for high and medium-flux scenarios, and by jointly estimating α and z directly from the raw histogram data. Previous approaches, on the other hand, apply a sequence of transformations to the data, estimating α and z in separate stages 18, 25, 27, 28, which limits the reconstruction performance. Experimental validation. We evaluate the proposed method on measurements experimentally acquired with the setup illustrated in Fig. 1. In Fig. 2, we assess the performance of the proposed 4 method on two scenes with highly varying reflectance and depth profiles. Both scenes contain objects with complex geometries and varying reflectance properties, including specular behavior for the "Statue of David" scene and Lambertian reflectance with spatially varying albedo in the "Bas-relief" scene. For both scenes, we capture a ground truth reference measurement with a 5% Neutral Density filter in the laser path which eliminates pileup distortions by damping the source intensity. To minimize shot noise fluctuations at the low count rates, we acquire very long sequences of 6 s length per spot at 4 MHz laser repetition rate. We scan every scene at a spatial resolution of 150 × 150 points. Hence, a full reference measurement is acquired in 150 · 150 · 6 s = 37.5 h per scene. The ground truth depth is extracted from long-exposure measurements using log-matched filtering 28 with the impulse response calibrated using a planar target captured under the same acquisition settings, i.e. unaffected by pileup. The experimental measurements, which serves as input for the proposed method, are acquired without any filters in the optical path. For each scene in Fig. 2, depth and albedo reconstructions along with the corresponding er- ror maps are shown. We show perspective renderings of the point cloud reconstructions for two different viewpoints. These results verify that the proposed method achieves high-quality depth and albedo reconstructions unaffected by scene-dependent pileup or shot-noise distortions. Spe- cifically, we compare our approach against the conventional log-matched filter estimate 28 as well as Coates' pileup correction method 23 followed by a Gaussian fit. The effect of pileup becomes apparent in the error map for the log-matched filter estimate resulting in more than 5 mm depth error for the "Statue of David" scene. Existing methods do not effectively suppress pileup and therefore suffer from strongly scene-dependent depth precision. In contrast, our method achieves sub-picosecond accuracy independent of scene depth and reflectance, despite the FWHM of the laser pulses being longer than 50 ps. We include additional comparisons in the Supplemental Res- ults where we also evaluate the proposed approach in the low-flux regime and demonstrate that our probabilistic method achieves close to an order of magnitude increased temporal resolution. In Table 3, we validate the sub-picosecond accuracy of the proposed approach without using spatial priors, i.e. relying solely on the probabilistic pileup model for individual histogram meas- urements. Specifically, we acquire a sequence of single-point measurements of a planar Lamber- tian target which is moved along the optical axis using a linear motion stage for 100 measurement points uniformly spaced with 0.5 mm distance. The setups for this measurement scenario and for an additional horizontal scanline measurement are shown in the Supplemental Information. In addition to the conventional log-matched filter estimate 28 and Coates' method 23, we compare the proposed method against and Shin et al.'s method 18 applied on the Coates-corrected histogram data, which adds censoring and background signal suppression 18 to Coates' method. Table 3 shows the average absolute error in depth and round-trip time for both of these measurements. The results demonstrate that, even without spatial priors, the proposed probabilistic method outperforms exist- ing state-of-the-art approaches by more than an order of magnitude. Adding the proposed spatial prior reduces temporal error by a factor of 2× on average. We refer to the Supplemental Results for additional experimental results and extensive evaluation in simulation. 5 Optimal regime of photon counts. Next, we analyze the performance of the proposed method with a fixed dwell or exposure time for a varying incident photon flux. The proposed probabil- istic method achieves optimal accuracy in an unconventional, pileup-affected regime with around 1 photon detection per pulse (see Fig. 4). This plot is generated in simulation without the use of spatial priors for the 450 nm Alphalas LD-450-50 laser with N = 104 shots. As expected, the log-matched filtering approach performs best when the measurements are neither too noisy nor too much affected by pileup. Existing sequential pileup correction methods, such as Coates' method 23, alleviate pileup and effectively extend the range where optimal performance can be reached bey- ond 1 expected photon detection per pulse. The accuracy of the proposed method is identical to previous methods for the low-flux regime, because we only consider a single pixel and no spatial priors are used. As the photon count increases, our approach substantially improves upon all exist- ing methods by up to two orders of magnitude. Optimal precision is achieved for expected photon counts around 1 per pulse, which is significantly higher than the low-flux regime in which existing methods operate. These results motivate an optimal photon flux regime for 3D imaging that is far outside the conventional low-flux regime. In the Supplemental Results, we also analyze the effect of the histogram bin width on timing accuracy in this optimal flux regime. The accuracy is consist- ent across a broad range of histogram bin-widths from sub-2 ps up to 80 ps, demonstrating that the proposed approach not only improves on the state-of-the-art by more than an order of magnitude in accuracy, but also reduces the timing resolution requirements on the detector side. Discussion Despite laser pulse widths being larger than 50 ps FWHM, the proposed probabilistic image form- ation model and corresponding inverse methods achieve sub-picosecond accuracy for active 3D imaging. Moreover, the proposed methods achieve this precision across a wide dynamic range, from low-flux to high-flux measurements. These capabilities are achieved by accurately modeling pileup distortions in the image formation model and by solving the inverse problem jointly for all latent variables using statistical priors. This paves the way for fast and precise photon-efficient 3D imaging systems in practical scenarios where widely-varying photon counts are observed in a scene. Applications across disciplines may significantly benefit from these capabilities, including 3D mapping and navigation, autonomous driving, robotic and machine vision, geographic inform- ation science, industrial imaging, and microscopic imaging. Methods Equipment details. The hardware setup consists of a time-resolved sensor, picosecond laser, illu- mination and collection optics, and a set of scanning mirrors to achieve a raster scan illumination pattern. The sensor is a PDM SPAD from Micro Photon Devices with a 100 µm × 100 µm sensor area, 27 ps timing jitter (measured with a 100 kHz laser at 675 nm), and 40.9 dark counts per second. Timing of photon arrivals is captured with a PicoHarp 300 Time-Correlated Single Photon Counting (TCSPC) module. The illumination source is a 450 nm or 670 nm picosecond laser 6 (ALPHALAS PICOPOWER-LD-450-50, PICOPOWER-LD-670-50). The 450 nm and 670 nm versions have pulse widths of 90 ps and 50 ps and average power of 0.406 mW and 0.11 mW re- spectively at a 10 MHz pulse repetition rate. The collection optics are designed to extend the field of view of the SPAD across the area scanned by the illumination source and consist of a 75 mm objective lens (Thorlabs AC508-075-A-ML), a 30 mm relay lens (Thorlabs AC254-030-A-ML) and a microscope objective (Olympus UPLFLN 20x objective). The laser spot is minified using a 50 mm (Thorlabs AC254-050-A-ML) and 250 mm (Thorlabs AC254-250-A-ML) lens relay and scanned with mirrors driven by a two-axis galvanometer (Thorlabs GVS012). Calibration details. All parameters of the image formation model are described in the Observa- tion model section. The detector's photon detection probability µ = 0.34 at 450 nm (µ = 0.33 at 670 nm) and dark count rate d = 40.9 c/s were calibrated using the method of Polyakov 29. The remaining unknowns are {ak, bk, ck k ∈ 1, . . . , K} for the Gaussian mixture model g. We use K = 8 and calibrate for the 450 nm Alphalas LD-450-50 laser the values {(0.57, 220.4, 32.0), (0.0059, 203.8, 10.2), (0.003, 214.6, 18.7), (-0.57, 220.3, 32.0), (0.003, 255.6, 47.5), (0.002, 297.9, 67.4), (0.009, 199, 7.1), (0.0003, 357.4, 143.2)}, and for 670 nm Alphalas LD-670-50 laser the values {(0.04, 199.4, 6.4), (0.004, 206.5, 7.0), (-0.001, 187.9, 12.5), (0.005, 204.8, 17.8), (0.003, 225.1, 27.2), (0.002, 254.6, 42.3), (0.0008, 301.0, 69.3), (0.0003, 388.7, 136.0)}. Measurements of a scene consisting of a single diffuse reflector placed at 1 m distance are acquired to estimate these parameters. Specifically, a n N D fi lter wi th 1% tr ansmission is pl aced in th e illumination path, and a low-noise histogram of a single point reflector is accumulated with N = 1 09 pulses, requiring approximately 17 minutes at a 1 MHz laser repetition rate. The high-absorption ND filter ensures a low-flux regime where pileup can be ignored and where the histogram counts are Poisson distributed. For the high number of 109 shots, these histogram measurements can be well approximated by a Gaussian distribution. Finally, the mixture parameters are estimated using effi- cient, conventional expectation-maximization algorithms 30. Please see the Supplemental Methods for calibration results for the for the 450 nm Alphalas LD-450-50 laser and the 670 nm Alphalas LD-670-50 laser. Algorithm parameters. The algorithm hyper-parameters are γ1 = γ2 = 10−1. These values were determined using simulations. Initial values of α = 1, z = 0 are used for the reconstruction al- gorithm in the Reconstruction algorithm section. In practice, warm-starting with the log-matched filter or Coates' estimate reduces the iterations needed to c onverge. The termination criteria for the proposed non-convex optimization algorithm are detailed in the Supplemental Methods. The algorithm run time with unoptimized Matlab code averages around 100 secs on a Intel i7 2.6GHz notebook computer. Note that the likelihood proximal operator, which dominates the algorithm runtime, is embarrassingly parallel across all measurement points, and hence is suited for imple- mentation on modern GPU hardware. Code and data availability. The code and data used to generate the findings of this study will be made public on GitHub. 7 References 1. Schwarz, B. Lidar: Mapping the world in 3d. Nature Photonics 4, 429 (2010). 2. Bronzi, D. et al. Automotive three-dimensional vision through a single-photon counting spad camera. IEEE Transactions on Intelligent Transportation Systems 17, 782–795 (2016). 3. McCarthy, A. et al. Kilometer-range, high resolution depth imaging via 1560 nm wavelength single-photon detection. Opt. Express 21, 8904–8915 (2013). 4. Pawlikowska, A. M., Halimi, A., Lamb, R. A. & Buller, G. S. Single-photon three-dimensional imaging at up to 10 kilometers range. Opt. Express 25, 11919–11931 (2017). 5. Raibert, M., Blankespoor, K., Nelson, G. & Playter, R. Bigdog, the rough-terrain quadruped robot. IFAC Proceedings Volumes 41, 10822–10825 (2008). 6. Cot´e, J.-F., Widlowski, J.-L., Fournier, R. A. & Verstraete, M. M. The structural and radiative consistency of three-dimensional tree reconstructions from terrestrial lidar. Remote Sensing of Environment 113, 1067–1081 (2009). 7. Velten, A. et al. Recovering three-dimensional shape around a corner using ultrafast time-of- flight imaging. Nature communications 3, 745 (2012). 8. M. O'Toole, G. W., D.B. Lindell. Confocal Non-Line-of-Sight Imaging Based on the Light Cone Transform. Nature 338–341 (2018). 9. Gariepy, G., Tonolini, F., Henderson, R., Leach, J. & Faccio, D. Detection and tracking of moving objects hidden from view. Nature Photonics 10, 23–26 (2016). 10. Morris, P. A., Aspden, R. S., Bell, J. E., Boyd, R. W. & Padgett, M. J. Imaging with a small number of photons. Nature communications 6, 5913 (2015). 11. Chen, Y., Muller, J. D., So, P. T. & Gratton, E. The photon counting histogram in fluorescence fluctuation spectroscopy. Biophysical journal 77, 553–567 (1999). 12. Aull, B. F. et al. Geiger-mode avalanche photodiodes for three-dimensional imaging. Lincoln Laboratory Journal 13, 335–349 (2002). 13. Rochas, A. et al. First fully integrated 2-d array of single-photon detectors in standard cmos technology. IEEE Photonics Technology Letters 15, 963–965 (2003). 14. Niclass, C., Rochas, A., Besse, P.-A. & Charbon, E. Design and characterization of a cmos 3-d image sensor based on single photon avalanche diodes. IEEE Journal of Solid-State Circuits 40, 1847–1854 (2005). 15. Richardson, J. A., Grant, L. A. & Henderson, R. K. Low dark count single-photon avalanche diode structure compatible with standard nanometer scale cmos technology. IEEE Photonics Technology Letters 21, 1020–1022 (2009). 8 16. Villa, F. et al. Cmos imager with 1024 spads and tdcs for single-photon timing and 3-d time- of-flight. IEEE journal of selected topics in quantum electronics 20, 364–373 (2014). 17. Bronzi, D. et al. 100 000 frames/s 64× 32 single-photon detector array for 2-d imaging and 3-d ranging. IEEE journal of selected topics in quantum electronics 20, 354–363 (2014). 18. Shin, D. et al. Photon-efficient imaging with a single-photon camera. Nature Communications 7 (2016). 19. Migdall, A., Polyakov, S. V., Fan, J. & Bienfang, J. C. Single-photon generation and detection: physics and applications, vol. 45 (Academic Press, 2013). 20. Tosi, A. et al. Fast-gated single-photon avalanche diode for extremely wide dynamic-range applications. In Design and Quality for Biomedical Technologies II, vol. 7170, 71700K (In- ternational Society for Optics and Photonics, 2009). 21. Cova, S., Longoni, A. & Ripamonti, G. Active-quenching and gating circuits for single-photon avalanche diodes (spads). IEEE Transactions on nuclear science 29, 599–601 (1982). 22. Busck, J. & Heiselberg, H. Gated viewing and high-accuracy three-dimensional laser radar. Applied optics 43, 4705–4710 (2004). 23. Coates, P. The correction for photon 'pile-up' in the measurement of radiative lifetimes. Journal of Physics E: Scientific Instruments 1, 878 (1968). 24. O'Connor, D. Time-correlated single photon counting (Academic Press, 2012). 25. Kirmani, A. et al. First-photon imaging. Science 343, 58–61 (2014). 26. Shin, D., Xu, F., Wong, F. N. C., Shapiro, J. H. & Goyal, V. K. Computational multi-depth single-photon imaging. Opt. Express 24, 1873–1888 (2016). 27. Gariepy, G., Tonolini, F., Henderson, R., Leach, J. & Faccio, D. Detection and tracking of moving objects hidden from view. Nature Photonics (2015). 28. Shin, D., Kirmani, A., Goyal, V. K. & Shapiro, J. H. Photon-efficient computational 3-d and reflectivity imaging with single-photon detectors. IEEE Transactions on Computational Imaging 1, 112–125 (2015). 29. Polyakov, S. V. Single-Photon Detector Calibration (Elsevier, 2015). 30. Murphy, K. P. Machine learning: a probabilistic perspective (The MIT Press, Cambridge, MA, 2012). Acknowledgements This work was in part supported by a National Science Foundation CAREER award (IIS 1553333), by a Sloan Fellowship, by the DARPA REVEAL program, and by the KAUST Office of 9 Sponsored Research through the Visual Computing Center CCF grant. The authors would like to thank Rafael Setra, Kai Zang, Matthew O'Toole, Amy Fritz, and Mark Horowitz for fruitful discussions in early stages of this project. Supplementary information Supplementary Information is attached to this submission. Competing interests The authors declare that they have no competing financial interests. Author contributions F.H. and G.W. conceived the idea. F.H. derived the algorithm. F.H. and S.D. im- plemented the algorithm. F.H processed all data. F.H., D.B.L., and G.W. designed the experiments. D.B.L. captured experimental data. F.H., S.D., and G.W. wrote the manuscript. G.W. supervised all aspects of the project. Correspondence Correspondence should be addressed to Felix Heide and Gordon Wetzstein. 10 Figure 1: Sub-picosecond 3D Imaging Framework. (a) A collimated, pulsed laser il- luminates the scene at a single point. The laser is laterally scanned using a 2-axis mirror galvanometer. Timing and control electronics time-stamp each detected photon arrival rel- ative to the last emitted pulse and accumulate these events in a histogram of spatio-temporal photon counts (b). This histogram is processed to estimate both reflectivity and depth in- formation (c). Two points are highlighted, one corresponding to high-flux (d) and the other to low-flux (e) measurements. Whereas the latter are noisy, high-flux measurements suffer from pileup distortion which introduce a significant bias for the depth estimation of con- ventional algorithms. The proposed estimation method accurately models both of these scenarios, allowing for reflectance information and travel time to be estimated with sub- picosecond accuracy from severely distorted measurements. 11 a b e d c 2.5 2 1.5 1 0.5 0 0 2.5 2 1.5 1 0.5 0 0 Ground Truth Scene Response Measurement 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 Pulsed Laser 2-A xis G alv o B e a m -Ste erin g n d g a n pin n str u ctio m e - S t a e c o R Ti m r c t o e t e D D A P S Scene Low-flux Measurement Point Laser Pulse High-flux Scene Point Normalized Counts Time [ps] Normalized Counts Time [ps] Measurement Probabilistic Reconstruction Low-flux High-flux Figure 2: Experimental reconstructions. A recorded spatio-temporal distribution of photon counts (a,e) is processed to estimate a 3D point cloud (b,c,f,g) that contains both depth and albedo information, here shown for two different scenes (photographs shown in (d,h)). The color-coded errors maps (d,h) directly compare the results of several depth estimation techniques, including log-matched filtering 28, Coates' method 23 followed by Gaussian fit (on high-flux measurement), and the proposed method. 12 Scene Target Planar Linear Stage Planar Linear Stage Pulse Log-matched Filter Wavelength 450 nm 450 nm 670 nm 670 nm Depth Error 4.54 mm 2.79 mm 3.19 mm 8.13 mm Coates' Method 23 4.40 mm 2.65 mm 3.00 mm 8.09 mm First Photon 18 on Coates 23 3.39 mm 1.98 mm 2.37 mm 8.00 mm Proposed Method 0.14 mm 0.24 mm 0.16 mm 0.23 mm Log-matched Filter Round-trip Error 15.13 ps 9.31 ps 10.64 ps 27.11 ps Coates' Method 23 14.68 ps 8.85 ps 9.99 ps 26.97 ps First Photon 18 on Coates 23 11.30 ps 6.58 ps 7.91 ps 26.68 ps Proposed Method 0.46 ps 0.91 ps 0.52 ps 0.76 ps Figure 3: Experimental validation of sub-picosecond accuracy on recorded single-pixel data without spatial priors. The average depth and round-trip time error for two scenes are shown, for the 450 nm Alphalas LD-450-50 laser (FWHM of 90 ps) and the 670 nm Alphalas LD-670-50 laser (FWHM of 50 ps), respectively. The background level is 5% for all scenes. We compare reconstructions of the conventional log-matched filter estim- ate 28, Coates' method 23 followed by a Gaussian fit, Shin et al. 18 on Coates-corrected measurements, and the proposed method. 13 Figure 4: Optimal photon count regime. Depth reconstruction accuracy for varying photon counts for the 450 nm Alphalas LD-450-50 laser (FWHM of 90 ps). The con- ventional log-matched filter, Coates' method 23, and the proposed method are compared. The optimal number of photon counts lies around the unconventional region of 1 photon detected per pulse on average, independent of the impulse response and for a broad range of histogram bin widths, see Supplemental Results. 14 Low-flux Medium-flux High-flux Log-matched Filter Coates' Method Proposed 102 101 100 10-1 Log Error [mm] - 450nm Source 10-2 10-4 10-2 100 102 Intensity [Expected Photon Detections] Supplementary Materials and Methods Image Formation Model This section introduces an observation model for imaging with single photon avalanche diodes (SPADs). For this purpose, we assume that a single SPAD, operating in free-running mode, and a pulsed laser are co-axially mounted and focused on a surface patch at some distance z. Additional collection optics may be mounted on the detector. The imaged surface patch is assumed to exhibit Lambertian reectance properties. Illumination and Response We illuminate the scene with N laser pulses, which denes the measurement acquisition time for a given laser repetition rate. Each laser pulse is char- acterized by the continuous temporal impulse response function g(t) ∈ R+, which describes the source intensity as a function of time. The temporally resolved irradiance at the detector can be modeled as r(t) = α g(t − 2z/c) + s, (1) with α ∈ [0, 1] as the reectance of the scene patch at distance z and c the speed of light. The additive component s models ambient illumination present in the scene. The intensity πz2 ρ Ac Ap cos ](n, o) combines all intensity loss in the light transport from the source α = α to camera via the single scene patch, consisting of the following components: distance fallo, absorption in the detector optics ρ, perspective foreshortening due to the orientation of the patch with normal n to the optical axis o, patch size Ap and aperture size Ac. As estimating these parameters is not central to the proposed method, we combine these terms in the factor α ∈ [0, 1]. Detector Model Photons impinging on the active area of the SPAD detector generate an electron avalanche. The time of occurrence of the resulting rise in voltage is digitized using a time-to-digital converter (TDC), which is synchronized with the laser. After each electron avalanche, the detector must be reset or quenched. As a consequence, no new photon arrival event can be detected for a certain amount of time known as the deadtime. The full width at half maximum (FWHM) of the laser pulse g(t) is commonly orders of magnitude smaller than this deadtime, while the time between pulses 1/f is usually larger than the deadtime. We therefore assume that each emitted laser pulse results in at most one detected photon event. Moreover, we assume that the SPAD is fully quenched between successive laser pulses. Oftentimes, the timing logic quantizes detected photon arrival events into a histogram with T discrete bins. The laser and timing logic are synchronized so that each bin i cor- responds to a consistent uniform time interval Ii = [ti, ti+1) relative to the beginning of the emitted pulse. Throughout an acquisition, N pulses are emitted and all detected photon arrival events are combined into a single histogram h ∈ ZT . We model the probability of detecting a histogram h given a response r(t) and N laser pulses as follows. Photon detections in the SPAD are either due to the incident response r or dark count d (false photon detections), with the latter modeled as a constant rate over 1 time, following Kirmani et al. [1] and Shin et al. [2]. Fluctuation in the delay between the actual time of arrival of a photon and the resulting timestamp are modeled as jitter using the probability density function (PDF) w(t). With these denitions, the number of photon detections in the SPAD during each interval Ii can be modeled as an inhomogeneous Poisson process, with mean µ(cid:0)r(z, α) w(t)(cid:1) + d(t) dt λi =Z ti+1 = µ(cid:0)α g(ti − 2z/c) + s(cid:1) + d, ti (2) where µ > 0 is the quantum eciency and d is the dark count rate per interval. The system (g ? w)(t)dt absorbs both the detector jitter as well as the laser impulse response g(τ ) =R τ +I impulse response. τ Parametric Impulse Response Model We represent the impulse response of the meas- urement system using a parametric continuous model that is compact and allows for fast statistical estimation. Specically, we model g as a Gaussian mixture model g(τ ) = K X k=1 ak exp(cid:18)− (τ − bk)2 c2 k (cid:19) (3) with K mixture components. The model coecients {ak, bk, ckk ∈ 1, . . . , K} are estimated in a precalibration step using measurements acquired over long acquisition periods to reduce the eect of noise. Specically, we nd the optimal least-squares t of the parameterized model in Eq. 3 to the measured histogram. Probabilistic Measurement Model Let λ denote the arrival rate for all histogram bins. We can express the probability P (n, iλ) of n photon detections in bin i as a Poisson random variable with mean λi that is P (n, iλ) = exp(−λi) λn i n! . The probability P (no detections) of detecting no photons during a pulse is then P (no detections) = P (n = 0, iλ) T i=1 T Y Y exp(−λi) = exp(−1T λ), i=1 = λ0 i 0! (4) (5) where 1 denotes a vector with all entries equal to one. Upon a photon detection, the detector-timing module increments the histogram bin in which the earliest photon detection occurred or does nothing if no photons were detected. 2 This means that photon arrivals are more likely to be recorded in earlier bins than later bins, a histogram distortion known as pile-up [3]. Specically, the probability P (iλ) of recording a photon arrival event in bin i is P (iλ) = P (n = 0, kλ)P (n > 0, iλ) exp(−λk) ∞ X n=1 exp(−λi) λn i n! (6) exp(−λk) exp(−λi)(exp(λi) − 1) = i−1Yk=1 i−1Yk=1 i−1Yk=1 = exp − = i−1Xk=1 In the third row, we use the Taylor seriesP∞n=1 xn n! = exp(x) − 1. Due to the fact that the dead time of the detector is less than the time between successive pulses, we assume that each of the N laser pulses is an independent trial. The probability P (hλ) of detecting the full histogram h is the multinomial distribution λk! − exp − λk! . i X k=1 P (hλ) = N ! h1!··· hT !(N − 1T h)! P (no detections)N−1T h = N ! h1!··· hT !(N − 1T h)! exp(−1T λ)N−1T h which follows from Eqs. (5) and (6). P (iλ)hi T i=1 Y i=1 exp − Y T λk! − exp − i−1Xk=1 i X k=1 (7) λk!!hi , Low-Flux Regime In a low-ux regime, where the probability of observing multiple photons reected back to the detector from a single emitted pulse is small, the eect of pile-up is negligible and can be ignored [1]. In this scenario, where the average number of photons per pulse is smaller than or equal to one, the detection probability becomes independent of previously detected events, and Eq. (6) becomes Plow-ux(iλ) ≈ P (n > 0, iλi) = 1 − P (n = 0, iλi) = 1 − exp(−λi), (8) 3 The probability P (hλ) of detecting the full histogram h then simplies to Plow-ux(hλ) = = ≈ T i=1(cid:18)N Y i=1(cid:18)N Y Y i=1 T T hi(cid:19)Plow-ux(iλ)hiP (no detections)N−1T h hi(cid:19)(1 − exp(−λi))hi exp(−λi)(N−hi) exp(−λiN ) (λiN )hi hi! = T Y i=1 Poisson(λiN ), (9) where the binomial is approximated as a Poisson distribution in the low-ux regime following the Poisson limit theorem (cf. [1, 2, 4]). As the count in each histogram bin reduces to a Poisson-distributed random variable, independent of other histogram bins, low-ux photon ecient imaging solves an estimation problem to recover the latent counts λ, which are subsequently used to estimate reectivity α and depth z. Model Assumptions The image formation model from Eq. 7 relies on the common as- sumption that SPAD measurements between laser pulses are independent. This assumption neglects afterpulsing, where an electron avalanche may be triggered by charge carriers gen- erated in the device by previous laser pulses. In our experimental setup, the afterpulsing probability is around 1% and is therefore ignored. Probabilistic Reconstruction Method In this section, we describe statistical estimation methods for recovering depth and albedo from SPAD measurements. The proposed model is independent of the illumination level and generalizes beyond the low-ux regime. First, we introduce a Bayesian maximum-a- posteriori (MAP) estimation approach with spatial priors on depth and reectivity images. The resulting optimization problem is solved using a proximal optimization algorithm, which decomposes the challenging joint estimation problem into a sequence of separable problems, one with an objective involving the per-pixel likelihood, the others involving the prior terms. Next, we describe per-pixel likelihood of the latent intensity vector λ using the model in- troduced in Eq. (7). Following this, a direct estimation of the depth and albedo (equivalent to reectance α in our notation) is described under the same model. Finally, we review the relationship of the proposed method to traditional approaches such as Coates pileup correction [3]. Bayesian MAP Estimation Thus far, we have only considered a single SPAD detector. In the following, we generalize our model to SPAD arrays or scenes scanned by sequentially scanning dierent scene points with a single SPAD. Assume that a scene is recorded at discrete points (m, j) ∀m ∈ {1, . . . , X}, j ∈ 4 {1, . . . , Y }, acquiring a SPAD image consisting of X×Y histograms h(m,j). Next, we describe how to recover a vectorized depth image z ∈ RXY and vectorized reectivity image α ∈ RXY from such histogram measurements. To stream-line notation in the following, we do not include the ambient term as an unknown. This unregularized oset term is incorporated into the solver method analogously to the the reectivity image, as described in the main manuscript. Given histogram measurements, we adopt a Bayesian approach and solve a maximum-a-posteriori (MAP) estimation problem to recover depth z and reectivity α as zMAP, αMAP = argmax z,α = argmax P (z, αh(0,0), . . . , h(Z,Y )) P (h(0,0), . . . , h(Z,Y )z, α)P (z, α) ≈ argmax = argmin z,α z,α Ym,j z,α Xm,j z,α Xm,j P (h(m,j)z(m,j), α(m,j))P (z)P (α) − log P(cid:0)h(m,j)λ(z(m,j), α(m,j))(cid:1) − log P (z) − log P (α) Λ(cid:0)h(m,j), z(m,j), α(m,j)(cid:1) + Γz(z) + Γα(α), (10) = argmin which maximizes the posterior over all measured histograms h(m,j). The third row in the above equation makes the simplifying assumptions that the scene depth and reectance priors are independent and that the likelihood of the histogram for each pixel is independent. The former assumption is a useful simplication for our reconstruction method, while the latter assumption is reasonable for the static scenes we capture. Although the specic prior distributions for depth Γz(z) and reectivity Γα(α) have not been dened yet, the objective in the last row of Eq. (10) exposes structure that can be leveraged in the optimization algorithm. The objective is a sum of three components. The rst is a data delity term originating from the likelihood, which is separable in the pixels, and there are two separate regularizer terms for depth and reectance, which result from the respective prior distributions. In the following, we will describe all of these three components in detail, starting with the prior terms. Priors on Depth and Reectance We assume Laplacian priors on the gradients of the depth and reectivity images, which leads to Total Variation regularization terms Γz, Γα in Eq. (10) P (z) ∝ exp ( −γ1k∇zk1 ) , P (α) ∝ exp ( −γ2k∇αk1 ) , Γz(z) = γ1k∇zk1, Γα(α) = γ2k∇αk1, (11) (12) where γ1, γ2 are prior weights that are estimated empirically from a representative dataset of depth and albedo images, respectively. 5 Likelihood The log-likelihood term in Eq. (10) is separable in the measurements (m, j). With the arrival rate vector λ from Eq. (2) a function of α, z, we can expand Λ(h, z, α) as follows Λ(h, z, α) = − log P (hλ(z, a)) = 1T λ(z, a)(N − 1T h) − = 1T λ(z, a)(N − 1T h) + with T i=1 T X X i=1 hi log(cid:16)e−Pi−1 i−1Xk=1 λi(z, a) = µ (α g(ti−2z/c)+s)+d k=1 λk(z,a) − e−Pi k=1 λk(z,a)(cid:17) hi log(cid:0)1 − e−λi(z,a)(cid:1) , X λk(z, a) − hi i=1 T (13) Having dened the objective function from Eq. (10), we now describe the algorithm to solve this minimization problem. We propose an ecient proximal optimization method that leverages the separable structure of the objective from Eq. (10) in the measurements (m, j). Following [5, 6] we reformulate the objective by splitting the unknowns in the three components while enforcing consensus in the constraints minimize Xm,j Λ(cid:0)h(m,j), z(m,j), α(m,j)(cid:1) + γ1k∇zk1 + γ2k∇αk1 = minimize Xm,j Λ(cid:0)h(m,j), x(m,j)(cid:1) } ∇ (cid:21) subject to (cid:20) ∇ {z } {z } =(cid:20) y1 y2 (cid:21) {z }y {z } (cid:20) z α (cid:21) {z }x = minimize g(x) + f1(y1) + f2(y2) + γ2ky2k1 + γ1ky1k1 {z f1(y1) f2(y2) g(x) K (14) kKx − y + uk2 2, ρ 2 fi(yi) + 6 2 X i=1 Lρ(x, y, u) = g(x) + subject to Kx = y, where we stack the unknowns z, α in the vector x = [zT , αT ]T . We have introduced two slack variables y1,2 ∈ R2XY for the gradients of the depth and albedo images ∇z,∇α ∈ R2XY , and enforce consensus between the gradients and the slack variables in the constraint. Furthermore, we have introduced functions g and f1,2 for the individual objective components in Eq. (14). We solve this objective using a linearized variant of the Alternating Method of Multipliers (ADMM) [5, 6]. To this end, we rst formulate the augmented Lagrangian of Eq. (14) as where we use the logarithmic identity log(a − b) = log(a) + log(1 − b/a) in the last row. Proximal Optimization which introduces here a scaled Lagrange variable u. The ADMM algorithm solves the saddle- point problem of minimizing Lρ w.r.t. each primal variable x, y, while maximizing the Lρ w.r.t. u. In block-coordinate descent fashion, a single ADMM iteration minimizes x and y in two separate sub-steps, and performs a dual ascent with the step size ρ as the last step. Linearized ADMM is a variant of the standard ADMM algorithm, where we replace the term in the augmented Lagrangian Eq. (14) with its linearization plus quadratic regularization: (ρ/2)kKx − yk + ukk2 2 (15) Performing dual ascent on the linearized augmented Lagrangian results in Algorithm 1. ρKT (Kx − yk + uk) + (µ/2)kx − xkk2 2. (16) 2, (x0, y0, u0). Algorithm 1 Linearized ADMM to solve Problem (14) 1: Initialization: µ > ρkKk2 2: for k = 1 to V do xk+1 = prox g 3: yk+1 j = prox fj uk+1 j = uk 5: 6: end for (xk − (ρ/µ)KT (Kxk − yk + uk)) (Kjxk+1 j + uk j + (Kjxk+1 − yk+1 j ) ∀j ∈ {1, 2} ) ∀j ∈ {1, 2} 4: µ ρ j Likelihood Update (17) Prior Updates (18) This algorithm contains two minimization steps, one w.r.t x in step (17), and another for y in step (18). Both minimization steps are expressed using proximal operators as a common notational shorthand [5], which is dened as proxφf (v) = argmin x (cid:18)f (x) + 2(cid:19) , 1 2φkx − vk2 where φ > 0 and v ∈ RXY . Proximal operators are alternatives to classical gradient and Hessian oracles used by rst and second order optimization methods. Algorithm 1 applies separate proximal operators for each f1,2 as the objective Eq. (14) is separable in the two components of y. Note that for convex objectives the algorithm is identical to the Pock- Chambolle algorithm [7], which is in fact linearized ADMM applied to the dual problem. For nonconvex objectives, such as Eq. 13, the two algorithms may give dierent results. Next, we specify and describe all relevant algorithm parameter choices and termination criteria. Algorithm Parameters and Termination Criteria For Algorithm 1 we use the para- meter choices ρ = 1/kKk2, µ = kKk2, x0 = 0, y0 = 0, and u0 = 0. We use the termination the criteria from [6], which stops the algorithm when the norms of the primal residual rk+1 = Kxk+1 − yk+1 and dual residual sk+1 = ρKT (yk+1 − yk) fall below given thresholds. The threshold for the primal residual is pri = √mabs + rel max{kKxkk2,kykk2}, 7 where we use abs = rel = 10−4. The dual threshold is pri = √nabs + relkKT ukk2. Proximal Prior Update in Eq. (18) Evaluating the two proximal operator updates from step (18) reduces to the well-known shrinkage operator [5], that is prox fj ν (v) = prox γj ν k·k1 (v) = max(cid:18)1 − , 0(cid:19) (cid:12) v, γj νv Shrinkage (19) which is a point-wise function that can be eciently evaluated in parallel for each vector component (i.e., each pixel (m, j)). Likelihood Proximal Update in Eq. (17) The proximal operator update in Eq. 17 is given by (20) ∀(m, j), (cid:0)α − vα (m,j)(cid:1)2 µ 2 2 kx − vk2 (m,j)(cid:1)2 + 2 µ 2 µ 2 g(x) + kx − vk2 Λ(cid:0)h(m,j), x(m,j)(cid:1) + (cid:0)z − vz Λ(cid:0)h(m,j), z, α(cid:1) + x x Xm,j µ 2 prox g µ (v) = argmin = argmin ⇔ (cid:16)prox g µ (v)(cid:17)(m,j) = argmin z,α (m,j), vα where vz (m,j) ∈ R extract the depth and reectivity of pixel (m, j) and z, α ∈ R are the optimization variables. The form of the update reveals why we linearized the augmented Lagrangian in the proposed optimization method: the minimization for depth and reectance is separable in the pixels (m, j). Unlike Eq. 18, the proximal update cannot be expressed in closed form. Instead we compute the joint optimization over all histograms as a sequence of bi-variate non-linear optimization problems that can be solved in parallel for all pixels (m, j). In the following, we describe how each of these non-linear problems is solved. (21) hi log(cid:0)1 − e−λi(z,α)(cid:1) } , (22) T X i=1 λk(z, α) − µ 2 µ 2 (α − vα)2 } i−1Xk=1 {z Λ(λ(z,α)) T i=1 hi µ 2 Υ(z,α) minimize Λ(h, z, α) + (z − vz)2 + {z X = minimize 1T λ(z, α)(N − 1T h) + (α − vα)2 (z − vz)2 + µ 2 + Bivariate Per-Pixel Likelihood Optimization Given pixel (m, j), we denote the cor- responding histogram h and proximal slack terms vz, vα for notational simplicity below. Recalling Eq. (13), we solve the following minimization problem jointly for z, α: 8 which introduces a bivariate function Υ that is the sum of the composite function Λ (λ(z, α)) and quadratic proximal closeness terms. We solve the non-linear bivariate minimization problem with the Newton Method given in Algorithm 2. We terminate this iteration once Algorithm 2 Bivariate Newton Method to solve Problem (21) 1: Initialization: z0 = vz, α0 = vα. 2: for k = 1 to V do (cid:20) zk+1 αk+1 (cid:21) =(cid:20) zk αk (cid:21) −(cid:2)HΥ(zk, αk)(cid:3)−1" ∂Υ(zk,αk) ∂Υ(zk,αk) ∂α ∂z # 4: end for 3: Newton Update (23) optimality is achieved and the gradient norm (cid:13)(cid:13)[∂Υ/∂z, ∂Υ/∂α]T(cid:13)(cid:13)2 < newton falls below the threshold newton = 10−4. The Newton update step (23) requires evaluation of the gradient ∇Υ and Hessian HΥ. Both are derived in the following. Gradient of Proximal Likelihood in Eq. (23) We use the chain rule to derive the gradient of the composite part of Υ as " ∂Υ(z,α) ∂Υ(z,α) ∂α ∂z # = ∇(cid:16)Λ ◦ λ(cid:17) (z, α) + µ(cid:20) z − vz α − vα (cid:21) α − vα (cid:21) = (Jλ (z, α))T ∇Λ(λ (z, α)) + µ(cid:20) z − vz =(cid:20)∂λ(z, α) ∂λ(z, α) (cid:21)T ∂α ∂z ∇Λ(λ (z, α)) + µ(cid:20) z − vz α − vα (cid:21) , (24) where J denotes here the Jacobian matrix. The gradient of Λ can be derived directly from Eq. (22) and dened in terms of its partials as ∂ Λ ∂λi = (N − i X k=1 hk) − hie−λi 1 − e−λi ∀i ∈ {1, . . . , T}. (25) The Jacobian of the arrival rates can be derived using Eqs. (2) and (3), dened in terms of its partials as ∀i ∈ {1, . . . , T}, (26) ∀i ∈ {1, . . . , T}. (27) ∂λi ∂z ∂λi ∂α 2µα = c K K c2 k k=1 ake− (ti−2z/c−bk )2 X ake− (ti−2z/c−bk )2 X k=1 c2 k = µ ·(cid:18)2 (ti − 2z/c − bk) c2 k (cid:19) Hessian of Proximal Likelihood in Eq. (23) In similar fashion to the gradient deriv- ations above, we apply the chain rule to estimate the 2 × 2 Hessian matrices for each pixel. 9 Specically, it is HΥ(z, α) = (Jλ (z, α))T HΛ (Jλ (z, α)) + (∇Λ)T ⊗ Hλ(z, α), (28) where the Jacobian matrices have already been dened in Eq. (26). The Hessian matrix HΛ is a diagonal matrix with ∂2 Λ ∂2λi ∂2 Λ λiλj = hie−λi (1 − e−λi)2 = 0 ∀i ∈ {1, . . . , T}, (29) ∀i 6= j. (30) Finally, the Hessian Hλ(z, α) can be derived immediately from Eqs.(2) and (3) as ∀i ∈ {1, . . . , T}, ∀i ∈ {1, . . . , T}, (cid:19) 2k 2 c − (31) By reusing the partials and function evaluations for the gradients, the Hessian can be com- puted eciently along with the gradient evaluations in the Newton update step. We run Newton iterations in parallel for each pixel, which can be implemented very eciently using modern graphics processing units (GPUs). Empirical Verication We validate the proposed probabilistic image formation model and reconstruction method by acquiring raw histogram measurements at dierent incident intensities. We use a scene con- sisting of a single diuse reector at a distance of 1 m from the detector and laser. To control the incident intensity, and thus the magnitude of the pileup eect, we place neutral-density (ND) lters of various optical densities in front of the detector. Using a low-transmission (1%) ND lter allows us to enforce a low-ux regime where pileup can be eliminated, though at the cost of a 100× increased acquisition time. Even when the measurements are distorted by pileup, the proposed reconstruction method is able to t a model which closely matches a low-ux reference measurement (see Fig. 1). The individual plots in the gure validate that the proposed image formation model and the probabilistic reconstruction generalize across dierent impulse response functions. In all cases, the proposed method estimates a model that matches the unseen reference measurement with high accuracy. 10 K Hλ(z, α) = X ake− (ti−2z/c−bk )2 X k=1 k=1 c2 k K ake− (ti−2z/c−bk )2 c2 k ∂2λ(z,α) ∂2z ∂2λ(z,α) ∂2α ∂z∂α ∂z∂α ∂2λ(z,α) ∂2λ(z,α)   with ·(cid:18)4(ti − 2z/c − bk)2 ·(cid:18)2(ti − 2z/c − bk) (cid:19) c4 k c2 k ∂2λi(z, α) ∂2z = 4µα2 c2 ∂λi(z, α) ∂α = 2µ c ∂2λi(z, α) ∂2α = 0. Supplementary Figure 1: Verication of the proposed forward model and re- construction method. The three plots show measurements for the 670 nm Alphalas LD-670-50 laser (FWHM of 50 ps), 450 nm Alphalas LD-450-50 laser (FWHM of 90 ps), and pulse-broadened 450 nm Alphalas LD-450-50 laser (broadened to 200 ps). Each plot displays raw histogram counts in blue, acquired without an ND lter, and a corresponding low-ux measurement captured with a 1% ND lter with negli- gible pileup. We show the model g with coecients recovered from the distorted raw counts using the proposed method. For all impulse response functions, the estimated model ts the unseen reference measurement with low residual error. Relation to Coates Method Coates method is a classical approach to pile-up correction [3]. It is a special case of our framework where we solve for λ rather than z and α with the prior P (λ) =(1, 0, if λi ≥ 0, for all i otherwise. The log-likelihood in Eq.10 is convex in λ with the above prior, so we nd the optimum by setting the gradient in Eq. 25 to zero, yielding λi = − log 1 − (32) hi k=1 hk! . N −Pi−1 After applying Coates method to recover λ, we can estimate depth z and reectivity α using a variety of methods, including those used in the low-ux regime [2, 4]. Additional Calibration Details This section provides additional detail on the calibration procedure used to estimate para- meters in the image formation model, described in the corresponding section. We assume that the detector's photon detection probability µ and dark count rate d have been calibrated independently of the setup discussed in this work, using established calibration methods for 11 450nm Laser Impulse Response 670nm Laser Impulse Response Broadened 670nm Laser Impulse Response 0 50 200 100 Temporal Bin [4ps] 150 250 300 18 16 14 12 10 02468 0 0 50 200 100 Temporal Bin [4ps] 150 250 300 0 0 50 200 100 Temporal Bin [4ps] 150 250 300 Measurement No-Pileup Measurement Estimated Model Fit 25 20 15 10 5 25 20 15 10 5 HistogramCount[Kilocount] Supplementary Figure 2: Calibration of the laser impulse response function. The left plot shows the impulse response of the red Alphalas LD-670-50 laser (FWHM of 50 ps in magenta) and the right plot shows the impulse response of the blue Alphalas LD-450-50 laser (FWHM of 90 ps). The blue plot shows the respective raw histogram counts captured with a 1% transmission ND lter to eliminate pileup. The respective FWHM is indicated by a magenta line. The red line shows the t of the model g(τ ) from the Image Formation Model section with K = 8 Gaussian components which exhibits only low residual error. single photon detectors (see, for example, Polyakov [8]). The remaining unknowns of the proposed model are the coecients {ak, bk, ckk ∈ 1, . . . , K} of the Gaussian mixture model for the overall system impulse response g from Eq. (3). To estimate these parameters, we acquire measurements of a scene consisting of a single diuse reector placed at 1 m distance, similar to the validation of the proposed forward model described at the end of the Image Formation Model section. We place an ND lter with 1% transmission in front of the laser illumination and capture a low-noise histogram of a single reector point by accumulating N = 109 pulses, which requires approximately 17 minutes at a 1 MHz laser repetition rate. The high-absorption ND lter ensures a low-ux regime where pileup can be ignored and where the histogram counts are Poisson distributed, as described in the Image Formation Model section. Moreover, for the extremely high numbers of shots, the Poisson distributed histogram counts become Gaussian distributed (the limiting form of the Poisson distribution is the Gaussian distribution). This allows us to rely on ecient conventional expectation- maximization algorithms for normal mixture models in the proposed calibration procedure. For further details on this algorithms, we refer the interested reader to Murphy [9]. Finally, Figure 2 shows calibration results for two dierent short-pulsed lasers used in this work. In both cases, tting K = 8 mixtures allows for a high-delity impulse response model. 12 450nm Laser Impulse Response 670nm Laser Impulse Response Measured Impulse Response Calibrated Model 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 1000 1500 Time [ps] 2000 2500 0 0 500 1000 1500 2000 2500 Time [ps] 0 0 500 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 NormalizedIntensity Supplementary Results Evaluation in Simulation In the following, we evaluate the proposed method using simulated measurements with known ground-truth depth and reectivity. Moreover, in addition to quantifying reconstruction error on a wide variety of scenes, synthesizing observations allows us also to analyze the eect of acquisition parameters on the accuracy of the proposed and of previously proposed reconstruction methods. Depth and Reectivity Error Evaluation To evaluate the performance of the proposed method in representative real-world scenarios, we use ground-truth depth and intensity images from the Middlebury Stereo dataset [10] which includes scenes of highly varying shape and relative distances. However, the intensity images in this data set are captured under the respective scene illumination, not under active illumination comparable to the proposed setup. To account for this, we estimate scene albedo value for the scene using the approach from Chen and Koltun [11]. Given a ground-truth depth/reectance map pair, we simulate histogram measurements using the full probabilistic image formation model for every pixel with N = 10, 000 shots. In the following, we show depth and albedo reconstruction error results for three dierent laser impulse response functions: the calibrated blue Alphalas LD-450-50 and red Alphalas LD-670-50 lasers from Fig. 2, and a synthetic, idealized Gaussian impulse response (FWHM of 50 ps). Figs. 3-8 show the depth and albedo error per scene for all impulse response functions. Here, we compare the proposed method against the conventional log-matched lter estimate [2], Gaussian t [12], rst-photon imaging [4, 1], and Coates method [3] followed by a Gaussian t. The logarithmic scale in the depth errors reveals that the proposed method outperforms all competing methods by two orders of magnitude in depth accuracy. Similarly, the re- ectance estimates show an order of magnitude in improvement over all existing methods. Hence, the proposed method opens the space of sub-picosecond accurate imaging which has not been possible with previous approaches. Tables 1, 2, and 3 show the average depth error and reectance PSNR per scene for all three laser impulse response functions, the calib- rated blue and red laser responses from Fig. 2, and the synthetic Gaussian impulse response. For compactness, we only include the conventional log-matched lter method, Coates [3] pileup correction followed by a Gaussian t, and the rst-photon imaging method [4, 1]. All error tables consistently verify that, independent of the scene and laser impulse response, we achieve two orders of magnitude improved depth accuracy and an order of magnitude lower reectance error compared with the best existing method. 13 Supplementary Figure 3: Depth Error Maps for the Alphalas LD-670-50 laser (FWHM of 50 ps). We compare the conventional log-matched lter estimate [2], Gaussian t [12], rst-photon imaging [4, 1], and Coates [3] followed by Gaussian t. 14 Supplementary Figure 4: Reectance Error Maps for the Alphalas LD-670-50 laser. Scenes from the Middlebury Stereo dataset [10] are shown. 15 Supplementary Figure 5: Depth Error Maps for the Alphalas LD-450-50 laser (FWHM of 90 ps). Scenes from the Middlebury Stereo dataset [10] are shown. 16 Supplementary Figure 6: Reectance Error Maps for the Alphalas LD-450-50 laser. Scenes from the Middlebury Stereo dataset [10] are shown. 17 Supplementary Figure 7: Depth Error Maps for an ideal Gaussian laser impulse response (FWHM of 50 ps). Scenes from the Middlebury Stereo dataset [10]. 18 Supplementary Figure 8: Reectance Error Maps an ideal Gaussian laser impulse response. Scenes from the Middlebury Stereo dataset [10] are shown. 19 Supplementary Table 1: Errors in depth and reectance estimation for blue Al- phalas LD-450-50 laser PSF (FWHM of 90 ps) are shown. Ground-truth depth and reectance from each scene are taken from the Middlebury Stereo dataset [10]. The best depth and albedo result for each scene is highlighted. We compare here re- constructions for the conventional log-matched lter estimate [2], Coates method [3] followed by a Gaussian t, rst photon imaging [4]m and the proposed method Dataset Scene Log-matched Filter Avg. Depth Error [mm] Coates [3] Avg. Depth Error [mm] First Photon [4] Avg. Depth Error [mm] Proposed Avg. Depth Error [mm] Log-matched Filter Reectance PSNR [dB] Coates [3] Reectance PSNR [dB] First Photon [4] Reectance PSNR [dB] Proposed Reectance PSNR [dB] Motorcycle Adirondack Jadeplant Piano Pipes Playroom Playtable Recycle Shelves Vintage Bicycle1 Average 13.95 mm 3.32 mm 13.71 mm 0.04 mm 11.24 dB 16.78 dB 12.99 dB 32.29 dB 10.56 12.81 1.85 2.58 6.28 16.39 0.03 0.04 15.24 13.06 19.94 18.55 23.59 12.80 29.02 33.22 14.07 3.27 7.92 0.04 10.33 15.79 11.78 27.88 14.40 3.50 19.69 0.04 10.43 16.41 8.71 30.59 14.57 3.43 17.38 0.04 11.13 17.02 9.66 32.47 14.66 4.16 10.06 0.03 9.35 15.04 10.33 23.63 13.29 2.96 12.47 0.04 11.28 17.08 12.21 36.04 15.44 3.73 21.94 0.04 9.73 15.59 8.25 41.09 10.90 1.71 11.27 0.04 16.10 20.53 18.13 30.64 15.08 3.83 4.87 0.04 9.53 15.21 19.33 37.73 17.61 5.45 22.57 0.04 7.42 13.40 8.07 32.42 Supplementary Table 2: Errors in depth and reectance estimation for red Al- phalas LD-670-50 laser PSF (FWHM of 50 ps) are shown. Ground-truth depth and reectance from each scene are taken from the Middlebury Stereo dataset [10]. The best depth and albedo result for each scene is highlighted. Dataset Scene Log-matched Filter Avg. Depth Error [mm] Coates [3] Avg. Depth Error [mm] First Photon [4] Avg. Depth Error [mm] Proposed Avg. Depth Error [mm] Log-matched Filter Reectance PSNR [dB] Coates [3] Reectance PSNR [dB] First Photon [4] Reectance PSNR [dB] Proposed Reectance PSNR [dB] Motorcycle Adirondack Jadeplant Piano Pipes Playroom Playtable Recycle Shelves Vintage Bicycle1 Average 5.67 mm 2.53 mm 8.27 mm 0.02 mm 12.85 dB 7.40 dB 12.30 dB 32.24 dB 4.14 5.21 1.28 2.05 10.27 4.99 0.02 0.02 11.70 15.45 9.73 6.70 12.06 21.67 29.00 33.23 5.74 2.50 6.27 0.02 12.92 6.81 9.01 27.87 5.87 2.68 12.45 0.02 12.88 7.06 8.55 30.58 6.00 2.70 9.68 0.02 15.20 5.87 9.47 32.45 6.08 3.19 4.60 0.02 12.97 6.25 10.32 23.62 5.34 2.35 8.52 0.02 12.22 8.41 11.07 36.02 6.31 2.82 13.38 0.02 12.80 6.48 8.28 41.11 4.28 1.48 8.11 0.02 12.72 9.74 18.01 30.64 6.16 2.76 3.20 0.02 12.00 7.50 18.75 37.70 7.26 4.01 9.47 0.02 10.53 6.86 8.16 32.43 Supplementary Table 3: Errors in depth and reectance estimation for an ideal Gaussian impulse response (FWHM of 50 ps) are shown. Ground-truth depth and reectance from each scene are taken from the Middlebury Stereo dataset [10]. The best depth and albedo result for each scene is highlighted. Dataset Scene Log-matched Filter Avg. Depth Error [mm] Coates [3] Avg. Depth Error [mm] First Photon [4] Avg. Depth Error [mm] Proposed Avg. Depth Error [mm] Log-matched Filter Reectance PSNR [dB] Coates [3] Reectance PSNR [dB] First Photon [4] Reectance PSNR [dB] Proposed Reectance PSNR [dB] Motorcycle Adirondack Jadeplant Piano Pipes Playroom Playtable Recycle Shelves Vintage Bicycle1 Average 6.89 mm 1.77 mm 4.54 mm 0.02 mm 17.22 dB 17.08 dB 13.41 dB 32.09 dB 5.62 6.65 0.66 1.22 3.58 2.79 0.02 0.02 14.40 17.37 17.71 18.90 15.22 23.80 28.99 33.19 7.31 1.81 6.16 0.02 19.24 17.53 10.46 32.35 6.99 2.40 4.27 0.02 19.39 15.56 10.48 23.61 7.91 3.37 7.03 0.02 19.08 14.52 8.59 32.24 6.63 1.52 2.59 0.02 16.02 17.22 12.56 35.82 7.08 1.90 5.10 0.02 17.56 17.11 9.42 30.52 5.86 0.58 2.67 0.02 14.65 18.54 19.52 30.62 7.28 2.17 2.94 0.02 16.85 16.17 19.38 37.29 6.93 1.77 5.58 0.02 16.92 17.40 9.43 27.85 7.50 2.08 7.28 0.02 17.99 17.27 8.66 40.49 Low-Flux Evaluation The rst-photon imaging method [4, 1] assumes a low-ux regime with negligible pileup and therefore performs only on-par with the conventional log-matched lter results in the quantitative evaluations from Tables 1-3. In addition to the pileup experiments from above, we have evaluated the proposed method in a low-ux regime at average photons 10−4 per-pulse and N = 10, 000 shots. Table 4 shows depth and reectance reconstruction performance in this scenario. Here, the rst-photon imaging method achieves substantial improvements over the conventional log-matched lter method and Coates [3] 20 Supplementary Table 4: Low-Flux Evaluation. Errors in depth and reectance estimation for red Alphalas LD-670-50 laser PSF (FWHM of 50 ps) are shown. Ground-truth depth and reectance from each scene are taken from the Middlebury Stereo dataset [10]. The best depth and albedo result for each scene is highlighted. Dataset Scene Log-matched Filter Avg. Depth Error [mm] Coates [3] Avg. Depth Error [mm] First Photon [4] Depth Error [mm] Proposed Avg. Depth Error [mm] Log-matched Filter Reectance PSNR [dB] Coates [3] Reectance PSNR [dB] First Photon [4] Reectance PSNR [dB] Proposed Reectance PSNR [dB] 18.28 18.28 6.87 0.89 10.04 10.04 18.09 26.89 20.08 20.07 8.63 0.67 9.91 9.91 18.67 26.99 Motorcycle Adirondack Jadeplant Piano Pipes Playroom Playtable Recycle Shelves Vintage Bicycle1 Average 18.57 mm 18.56 mm 7.25 mm 0.75 mm 10.24 dB 10.24 dB 19.55 dB 26.85 dB 12.94 17.81 12.94 17.78 11.51 8.16 1.11 0.73 8.79 11.83 8.79 11.83 20.65 19.99 25.86 28.35 18.50 18.53 4.34 0.65 10.51 10.51 18.67 23.98 19.20 19.22 3.16 0.71 10.42 10.42 19.61 27.05 20.25 20.21 5.57 0.65 10.27 10.27 18.75 31.25 19.72 19.65 8.09 0.65 11.40 11.40 19.55 29.39 19.17 19.21 9.01 0.54 10.33 10.33 19.68 21.32 16.51 16.49 1.89 1.12 10.40 10.40 23.54 27.88 21.81 21.81 12.55 0.55 8.72 8.72 17.90 26.36 pileup correction method. The latter performs on-par with the log-matched lter approach as measurement distortions due to pileup are minimal. The proposed method outperforms all competing approaches even in this low-ux scenario, however, it does not achieve sub- picosecond accuracy for such noisy measurement. Existing imaging methods tailored to this scenario, such as [2, 4, 1], perform denoising, deconvolution, peak-tting, and histogram correction, as separate stages of a pipeline, for both reectance and depth. In contrast, the proposed method performs these tasks jointly as part of the probabilistic inference framework from Supplementary Materials and Methods section, leading to the substantially improved performance even in this low-ux imaging scenario. Analysis In this section, we analyze the eects of acquisition parameters on the reconstruction per- formance. To evaluate the probabilistic reconstruction method independent of the spatial priors, we consider a single-pixel histogram. Fig. 9 shows plots of the logarithmic depth estimation error with varying incident intensity, which is determined by the laser intensity, scene reectance, and distance fallo. The eect of varying the number of shots N per his- togram acquisition is shown in individual plots for each of the two calibrated laser impulse response functions (cf. Fig. 2). The accuracy follows a similar trend across all plots. The conventional log-matched reconstruction reaches an optimum for a relatively low number of counts that is dropping quickly as the number of incident detections becomes larger than 1 expected photon detection per pulse, creating substantial histogram pileup. Existing sequen- tial pileup correction methods, such as Coates [3], alleviate pileup and eectively extend the range where optimal performance can be reached beyond 1 expected photon detection per pulse. In this experiment, the proposed probabilistic reconstruction accuracy is identical to previous methods for the low-ux regime for a low number of shots since we only consider a single pixel and no spatial priors can be exploited. As the intensity increases the proposed approach substantially improves on all existing methods with the optimal performance by two orders of magnitudes for a higher number of shots. Note that even in low-ux regimes, the discussed method improves on competing approaches when pileup becomes negligible for a large number of shots. Also note that the optimal performance in each setting is achieved, 21 Supplementary Figure 9: We show the depth reconstruction accuracy for a vary- ing number of shots N at varying incident intensities which depend on distance of the reector, reectance and laser intensity. The rst row depth accuracy results for the blue Alphalas LD-450-50 laser (FWHM of 90 ps) , and the second row shows the red Alphalas LD-670-50 laser (FWHM of 50 ps). Results are shown for the conventional log-matched lter method, Coates [3], and the proposed methods. in fact, in a dierent intensity region around 1 photon detection per pulse, in contrast to existing methods which perform best between 10−2 to 10−1 average photons detections per pulse. Hence, the proposed approach argues not only for a joint probabilistic reconstruction Supplementary Figure 10: Eect of Histogram Bin Width on Estimation Error. The left plot shows the depth reconstruction error of the proposed method for varying histogram bin sizes and incident intensities. We can see that the optimal intensity region stays around an average of 1 photon per pulse independent of the bin-size. The right plot shows reconstruction quality versus bin width for a xed intensity of 1 photon per pulse. The proposed method achieves low depth reconstruction error across a broad range of bin widths, and only fails for bin widths larger than 80 ps. 22 Log-matched Coates Proposed 10-2 100 102 104 10-4 10-2 100 102 104 10-4 10-2 100 102 104 10-4 10-2 100 102 104 10-2 100 102 104 10-4 10-2 100 102 104 10-4 10-2 100 102 104 10-4 10-2 100 102 104 Intensity [Expected Photon Detections] Intensity [Expected Photon Detections] Intensity [Expected Photon Detections] Intensity [Expected Photon Detections] 102 101 100 10-1 10-2 10-4 102 101 100 10-1 10-2 10-4 450nm Laser - Log Error [mm] 670nm Laser - Log Error [mm] Number of Trials: 40 30 80 Histogram Bin Resolution [ps] 50 70 60 90 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 AbsoluteError[mm] 0.5 10 0 10 -0.5 10 -1 10 10-4 10-1.5 Logarithmic Error [mm] 1 0 10 20 10-1 10-2 10-3 100 101 Expected Photon Detections 103 102 10 [ p s ] 0 20 t i o n 30 o l u 40 s R e 0.5 0 -0.5 -1 -1.5 Error[mm] Logarithmic 100 t o g r 80 90 H i s 60 50 70 a m B i n but also for a dierent operating mode far outside of the conventional range. Fig. 10 analyzes the eect of the histogram bin width on depth estimation accuracy in the same single-pixel scenario as considered above for Fig. 9. The accuracy of the proposed approach is consistent across a broad range of histogram bin-width from sub-2 picoseconds up to 80 picoseconds, as shown in the right plot of Fig. 10. This result veries that the proposed method exploits the large temporal support of the laser impulse response. Only for very large bin-widths above 80 picoseconds, reconstruction accuracy begins to drop substantially. This means that the proposed approach, besides improving state-of-the-art accuracy by orders of magnitude, also reduces the required timing accuracy on the detector side. Moreover, the left plot in Fig. 10 veries that the optimal intensity region lies around 1 photon per pulse independent of the histogram bin widths, demonstrating that the proposed method enables this unconventional operating regime across detectors with dierent temporal resolutions. Experimental Evaluation In the following, we evaluate the proposed method on measurements experimentally ac- quired with the setup illustrated in Fig. 1 of the main manuscript. First, we validate the sub-picosecond accuracy of the proposed approach without using spatial priors, i.e. rely- ing solely on the probabilistic pileup model for individual histogram measurements. Next, we assess the depth and albedo reconstruction quality on a variety of scenes with varying object shape and albedo. Both the per-histogram evaluations and the full spatio-temporal results demonstrate an order of magnitude improvement in accuracy compared to existing reconstruction methods. Temporal Resolution Evaluation without Spatial Priors To validate the sub-picosecond accuracy of the described approach, we acquire a horizontal scanlines of histogram measurements. Three scene objects with dierent vertical depth proles are used in the following experiments. Photographs of these scenes are shown in the left column of Fig. 11 along with an overlay illustration of the respective laser scanline path. All targets have been placed at a distance of 1 m to the detector. The rst scene is a planar foam board target placed perpendicularly to the optical axis of the detector's objective lens. The second scene is a 3D-printed step target which was covered with a thin layer of white primer to eliminate severe subsurface scattering in the acrylic plastic printing material. With the third scene, shown in the bottom left of Fig. 11, we evaluate the performance of the proposed reconstruction method in the absence of spatial priors. Specically, we acquire a sequence of single-spot measurements of a planar foam board target which is moved back and forth along the optical axis of the detector optics. For all scenes, we acquire ground truth measurements by placing a 5% Neutral Density lter in the laser path which eliminates pileup distortion but leads to substantial increase in noise for a xed number of shots. To eliminate these uctuations in the ground truth reference, we acquire long sequences of 10 s length per spot at 5 MHz laser repetition rate. For these reference measurements, we use a histogram bin size of 4 ps. We calibrate the temporal 23 impulse response from measurements of a planar target with the identical capture settings. The ground truth depth is then extracted from the clean, long-exposure measurements by log- matched ltering [2]. We acquire histogram measurements for each scene from N = 100, 000 trials without additional ltering in the optical path. Comparing the best approaches from Depth and Reectivity Error Evaluation section, we compare the proposed method against the conventional log-matched lter estimate [2], and Coates pileup correction method [3] followed by a Gaussian t. Figs. 11 and 12 show the ground truth and reconstructions for all scenes and for dierent laser impulse responses from the 450 nm Alphalas LD-450-50 laser and the 670 nm Alphalas LD-670-50 laser, see the Additional Calibration Details section for a comparison of the two temporal impulse response functions. We have removed the calibration oset due to the baseline between camera and sensor for the planar target for better visualization of the reconstruction error. The measurements show varying pileup based on laser spot position and response function due to the change in incident intensity based on distance-fallo and laser pulse energy. To illustrate the depth dependency, we have acquired the linear stage measurements at two dierent distances in Fig. 11. The measurement at 0.5 m distance suers here from signicantly more pileup distortion than the measurement at 1 m distance. Table 5 shows the average absolute error in depth and round-trip time for all of these measurements. The proposed method outperforms all methods by more than an order of magnitude in improvement. In addition to the conventional log-matched lter estimate [2] and Coates method [3], we also show here Shin [4] applied on the pileup-corrected histogram data output by Coates method, which adds censoring and background signal suppression to the basic Coates method. The results validate that the proposed method achieves sub- picosecond accuracy with relatively long laser pulses larger than 50 ps independent of the laser pulse shape and independent of the scene shape. Note that the motion stage recon- structions have a higher average error than the planar target measurements due to the lack of spatial priors. This validates the performance independent of the scene (sub-picosecond accuracy is achieved for all scenes), but also demonstrates the performance of the spatial prior even in this high-intensity scenario, leading to an 2× reduction in average error. The depth proles for the log-step in Figs. 11 and 12 reveal that the increased average error compared to the planar target occur at two positions where the target contains sharp edge transitions. At these positions, the depth of two surface patches at dierent depths are mixed in the impulse response, making this an ambiguous point with uncertainty in the log-matched ground-truth measurement. Note that the proposed approach solves here for a single depth estimate consistent with spatial prior information nearby. Aside from such corner-cases, the proposed method achieves low reconstruction error with sub-picosecond accuracy for all scanline measurements and translation measurements. Depth and Reectivity Error Evaluation In this section, we evaluate the performance of the proposed method for a variety of chal- lenging real-world scenes with highly varying reectance and depth proles. Photographs of the test scenes are shown in Fig. 13. The individual scene objects contain complex geomet- 24 Supplementary Figure 11: Experimental depth evaluation for the 450 nm Al- phalas LD-450-50 laser PSF (FWHM of 90 ps). Depth proles for three dierent scenes with photographs in the left are shown in the rows. The rst two rows show scanline measurements for a planar and step target, while the last two rows show a sequence of spot measurements for a motion stage aligned with the optical axis at the distance of 1 m and 0.5 m. The center column shows depth prole reconstruc- tions for the conventional log-matched lter estimate [2], Coates method [3] followed by a Gaussian t, and the proposed probabilistic reconstruction method. The right column shows the corresponding absolute error for each measurement. See Table. 5 for the corresponding error analysis. 25 Setup Depth Reconstruction Reconstruction Error Log-matched Coates Proposed 0 5 0 5 15 10 Horizontal Laser Spot Position 25 30 35 40 20 45 50 Log-matched Coates Proposed 15 10 40 Horizontal Laser Spot Position 20 25 35 30 45 50 Log-matched Coates Proposed 0 10 20 30 40 50 60 70 80 90 100 Log-matched Coates Proposed 0 10 20 40 70 30 Motion Stage Position 60 50 80 90 100 10 0123456789 10 0123456789 10 0123456789 10 0123456789 Absolute Error [mm] Absolute Error [mm] Log-matched Coates Proposed Ground Truth 15 10 Horizontal Laser Spot Position 20 30 35 40 25 45 50 Log-matched Coates Proposed Ground Truth 15 40 10 Horizontal Laser Spot Position 35 20 30 25 45 50 20 18 16 14 12 10 02468 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 02468 Depth Profile [mm] Depth Profile [mm] 0 5 0 5 Laser scanline highlighted in Blue Laser scanline highlighted in Blue Absolute Error [mm] Log-matched Coates Proposed Ground Truth 0 10 20 30 40 50 60 70 80 90 100 Absolute Error [mm] Log-matched Coates Proposed Ground Truth 80 90 100 0 10 20 40 70 30 Motion Stage Position 60 50 60 58 56 54 52 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 2468 0 60 58 56 54 52 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 2468 0 Depth Profile [mm] Depth Profile [mm] Laser spot highlighed in Blue Planar Target Log-Step Target Linear Translation Stage at 1m Linear Translation Stage at 0.5m Supplementary Table 5: Validation of sub-picosecond accuracy on experimental data. The average depth and round-trip time error for the scenes from Figs. 11 and 12 is shown, for the 450 nm Alphalas LD-450-50 laser (FWHM of 90 ps) and the 670 nm Alphalas LD-670-50 laser (FWHM of 50 ps), respectively. We com- pare here reconstructions for the conventional log-matched lter estimate [2], Coates method [3] followed by a Gaussian t, Shin [4] on Coates-corrected measurements, and the proposed method. Scene Target Laser Wavelength Planar 450 nm Log-Step 450 nm Linear Stage Stage at 0.5 m 450 nm 450 nm Planar 670 nm Log-Step 670 nm Log-matched Filter Avg. Depth Error [mm] Coates [3] Avg. Depth Error [mm] First Photon [4] on Coates [3] Avg. Depth Error [mm] Proposed Avg. Depth Error [mm] Log-matched Filter Avg. Round-time Error [ps] Coates [3] Avg. Round-time Error [ps] First Photon [4] on Coates [3] Round-time Error [mm] Proposed Avg. Round-time Error [ps] 4.54 4.40 3.39 0.14 15.13 14.68 11.30 0.46 2.77 2.69 1.82 0.16 9.25 8.96 6.06 0.55 2.79 2.65 1.98 0.24 9.31 8.85 6.58 0.80 4.22 4.01 3.56 0.27 14.06 13.35 11.87 0.91 3.19 3.00 2.37 0.16 10.64 9.99 7.91 0.52 2.65 2.49 2.07 0.22 8.82 8.29 6.89 0.75 Linear Stage 670 nm 8.13 mm 8.09 mm 8.00 mm 0.23 mm 27.11 ps 26.97 ps 26.68 ps 0.76 ps ries, varying materials and reectance properties, including highly specular and Lambertian reectors with both spatially varying and uniform albedo. Hence, the intensity-driven pileup distortions occur with high spatial uctuation for the given test scenes. We have acquired histogram measurements using the setup illustrated in Fig. 1 of the main manuscript using the 670 nm Alphalas LD-670-50 laser. The scene objects are placed at a distance of 1 m to the detector. For all scenes, we capture a ground truth reference measurement with a 5% Neutral Density lter in the laser path which eliminates pileup distortions by damping the source intensity. To minimize shot noise uctuations at the low count rates resulting from the ND lter, we acquire very long sequences of 6 s length per spot at 4 MHz laser repetition rate. We scan every scene at a spatial resolution of 150 × 150 laser spots. Hence, a full reference measurement is acquired in 150 · 150 · 6 s = 37.5 h per scene. The ground truth depth is extracted from this long-exposure measurements using log-matched ltering [2] with the impulse response calibrated using a planar target captured under the same acquisition settings, and hence unaected by pileup. The target measurement, which serves as input for the proposed method, is acquired without any ND lter in the optical path. In addition, to validate the performance of the proposed method in the low-ux regime as in the Low-Flux Evaluation section, we acquire a third low-ux measurement using a 1% Neutral Density lter resulting in a very low per-spot count, e.g. less than 5 for Fig. 14. For each scene, depth and albedo reconstructions, and the corresponding error maps in the high-ux and low-ux regime are visualized in Figs. 14 - 18. Note that, in contrast to the previous section which validates the performance of the proposed method in the absence of spatial priors, we show here reconstructions using the proposed depth and reectance priors. We show perspective renderings of the point cloud reconstructions along with error maps for the ground truth depth measured as discussed above. These results verify that the proposed approach achieves high-quality depth and albedo reconstructions for a wide range 26 Supplementary Figure 12: Experimental depth evaluation for the 670 nm Al- phalas LD-670-50 laser PSF (FWHM of 50 ps). The same scene as in Fig. 11 are shown here. See Table. 5 for the corresponding error analysis. of scenes despite the substantial pileup distortion present in the measurements. The eect of pileup becomes apparent in the error map for the log-matched lter estimate [2] with more than 10 mm error depending on the scene. In contrast, the proposed method achieves sub-picosecond accuracy independent of scene depth and reectance. Fig. 14 shows a scene with varying albedo and depth, Fig. 15 contains depth variations and uniform albedo, the scene in Fig. 16 is composed of cluttered scene geometry, Fig. 17 shows reconstructions for a highly specular copper coin surface, and Fig. 18 demonstrates results for the back side of the coin painted with matte primer. In all of these scenarios, the proposed approach recovers extremely ne features. The ne text on the bottom of the coin in Fig. 17 has a depth of 27 Reconstruction Error Log-matched Coates Proposed 15 20 10 Horizontal Laser Spot Position 25 40 30 35 45 50 Log-matched Coates Proposed 15 10 Horizontal Laser Spot Position 35 20 40 30 25 45 50 Log-matched Coates Proposed 80 90 100 40 70 30 Motion Stage Position 50 60 0 10 20 10 0123456789 Absolute Error [mm] Log-matched Coates Proposed Ground Truth 80 90 100 40 70 30 Motion Stage Position 50 60 0 10 20 62 60 58 56 54 52 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 02468 Depth Profile [mm] Laser scanline highlighted in Red Laser spot highlighed in Red 10 0123456789 0 5 Absolute Error [mm] Log-matched Coates Proposed Ground Truth 0 5 15 40 10 Horizontal Laser Spot Position 30 20 25 35 45 50 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 2468 0 Depth Profile [mm] 0 5 10 0123456789 Absolute Error [mm] Log-matched Coates Proposed Ground Truth Depth Reconstruction 10 15 40 Horizontal Laser Spot Position 30 25 20 35 45 50 20 18 16 14 12 10 2468 Depth Profile [mm] Setup Laser scanline highlighted in Red 0 0 5 Planar Target Log-Step Target Linear Translation Stage less than 0.2 mm, measured with a coordinate measurement system, which is recovered by the proposed method. The depth prole of the Lincoln bust on the matte backside of the coin in Fig. 18 ranges between 0.05 mm and 0.3 mm and is faithfully recovered as visualized in the color-coded depth visualization in the center column of Fig. 18. For all scenes, we have compared the proposed method against the best approaches identied in the Depth and Reectivity Error Evaluation section, for both the high-ux and low-ux scenario. Specically, we compare the proposed method against the conventional log-matched lter estimate [2] and Coates pileup correction method [3] followed by a Gaussian t. The experimental error map results validate that the proposed method outperforms all existing methods by an order of magnitude in accuracy. We achieve sub-picosecond temporal resolution with relatively long laser pulses larger than 50 ps and independent of the scene. Note that this is in contrast to existing methods which do not account for pileup and hence suer from strongly scene-dependent resolution. In addition, we have also experimentally validated the proposed method in the low- ux regime in Figs. 14 - 18, previously evaluated in simulation in the Low-Flux Evaluation section. Specically, we show error maps for the low-ux measurement for the rst-photon imaging method [4] and the proposed method applied to the same low-ux measurement data. While the average error is substantially larger than in the high-ux range, matching the simulation results from the Analysis Section, the proposed method reduces the error by up to an order of magnitude for scene points, such as in Fig. 15. For areas with counts so low that the method only relies on the spatial prior, such as a few low-albedo regions in Fig. 14, the residual error is on par with the rst photon imaging approach. However, across all test scenes, temporal resolution increases by a factor between 2× and 10× in the low-ux regime. While existing imaging methods tailored to this scenario, including [4], perform denoising, deconvolution, peak-tting, and histogram correction, as separate stages of a pipeline, these experimental results validate that jointly solving these tasks in the proposed probabilistic Supplementary Figure 13: Photographs of the scenes used in the experimental validation of the proposed approach. The scenes contain objects with complex shapes, e.g. synthetic ower scene, varying reectance properties, e.g. metallic coin with highly specular reectance, and varying albedos, e.g. painted bas-relief. 28 ScenesAforAExperimentalAAssessment Bas-relief SyntheticAFlowerABouquet StatueAofADavid CopperACoinAReplica DiffuselyAPaintedACoin inference framework improves resolution by an order of magnitude in both, the high-ux as well as low-ux measurement scenario. 29 Supplementary Figure 14: Experimental results and error analysis for the Bas- relief scene from Fig. 13. Perspective point cloud renderings of depth and albedo are shown for two dierent camera positions in the top two rows. While the recovered reectance is shown as point-intensity in the left plots, the right plots additionally color-code the error according to the color map in the bottom right. The bottom row visualizes the corresponding orthographic depth error maps comparing: conventional log-matched lter [2], rst-photon imaging [4, 1] on a low-ux measurement of the scene, the proposed method on identical low-ux measurement, Coates method [3] followed by Gaussian t (on high-ux measurement), and the proposed method. 30 Supplementary Figure 15: Experimental results and error analysis for the Statue of David scene from Fig. 13. Perspective point cloud renderings and orthographic error maps are shown analogue to Fig. 14. 31 Supplementary Figure 16: Experimental results and error analysis for the Syn- thetic Flower Bouquet scene from Fig. 13. Perspective point cloud renderings and orthographic error maps are shown analogue to Fig. 14. 32 Supplementary Figure 17: Experimental results and error analysis for the Cop- per Coin Replica scene from Fig. 13. Perspective point cloud renderings and ortho- graphic error maps are shown analogue to Fig. 14. 33 Supplementary Figure 18: Experimental results and error analysis for the Dif- fusely Painted Coin scene from Fig. 13. Perspective point cloud renderings and orthographic error maps are shown analogue to Fig. 14. In addition, point cloud renderings with the depth color-coded are shown in the center column of the top two rows, which highlight the ne depth prole of the Lincoln bust prole recovered in the center of the coin. 34 References [1] Kirmani, A. et al. First-photon imaging. Science 343, 5861 (2014). [2] Shin, D., Kirmani, A., Goyal, V. K. & Shapiro, J. H. Photon-ecient computational 3-d and reectivity imaging with single-photon detectors. IEEE Transactions on Com- putational Imaging 1, 112125 (2015). [3] Coates, P. The correction for photon `pile-up' in the measurement of radiative lifetimes. Journal of Physics E: Scientic Instruments 1, 878 (1968). [4] Shin, D. et al. Photon-ecient imaging with a single-photon camera. Nature Commu- nications 7 (2016). [5] Parikh, N. & Boyd, S. Proximal algorithms. Foundations and Trends in Optimization 1, 123231 (2013). [6] Boyd, S., Parikh, N., Chu, E., Peleato, B. & Eckstein, J. Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends in Machine Learning 3, 1122 (2011). [7] Chambolle, A. & Pock, T. A rst-order primal-dual algorithm for convex problems with applications to imaging. Journal of mathematical imaging and vision 40, 120145 (2011). [8] Polyakov, S. V. Single-Photon Detector Calibration (Elsevier, 2015). [9] Murphy, K. P. Machine learning: a probabilistic perspective (The MIT Press, Cam- bridge, MA, 2012). [10] Scharstein, D. et al. High-resolution stereo datasets with subpixel-accurate ground truth. In German Conference on Pattern Recognition, 3142 (Springer, 2014). [11] Chen, Q. & Koltun, V. A simple model for intrinsic image decomposition with depth cues. In Proceedings of the IEEE International Conference on Computer Vision, 241248 (2013). [12] Pawlikowska, A. M., Halimi, A., Lamb, R. A. & Buller, G. S. Single-photon three- dimensional imaging at up to 10 kilometers range. Opt. Express 25, 1191911931 (2017). 35
1903.07567	1	1903	2019-03-07T20:35:49	Low-loss waveguides on Y-cut thin film lithium niobate: towards acousto-optic applications	[ "physics.app-ph", "physics.optics" ]	We investigate the dependence of photonic waveguide propagation loss on the thickness of the buried oxide layer in Y-cut lithium niobate on insulator substrate to identify trade-offs between optical losses and electromechanical coupling of surface acoustic wave (SAW) devices for acousto-optic applications. Simulations show that a thicker oxide layer reduces the waveguide loss but lowers the electromechanical coupling coefficient of the SAW device. Optical racetrack resonators with different lengths were fabricated by argon plasma etching to experimentally extract waveguide losses. By increasing the thickness of the oxide layer from 1 {\mu}m to 2 {\mu}m, propagation loss of 2 {\mu}m (1 {\mu}m) wide waveguide was reduced from 1.85 dB/cm (3 dB/cm) to as low as 0.37 dB/cm (0.77 dB/cm), and, resonators with quality factor greater than 1 million were demonstrated. An oxide thickness of approximately 1.5 {\mu}m is sufficient to significantly reduce propagation loss due to leakage into the substrate and simultaneously attain good electromechanical coupling in acoustic devices. This work not only provides insights on the design and realization of low-loss photonic waveguides in lithium niobate, but most importantly offers experimental evidence on how the oxide thickness directly impacts losses and guides its selection for the synthesis of high-performance acousto-optic devices in Y-cut lithium niobate on insulator.	physics.app-ph	physics	Low-loss waveguides on Y-cut thin film lithium niobate: towards acousto-optic applications LUTONG CAI *, ASHRAF MAHMOUD AND GIANLUCA PIAZZA Department of Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, PA, 15213, USA Abstract: We investigate the dependence of photonic waveguide propagation loss on the thickness of the buried oxide layer in Y-cut lithium niobate on insulator substrate to identify trade-offs between optical losses and electromechanical coupling of surface acoustic wave (SAW) devices for acousto-optic applications. Simulations show that a thicker oxide layer reduces the waveguide loss but lowers the electromechanical coupling coefficient of the SAW device. Optical racetrack resonators with different lengths were fabricated by argon plasma etching to experimentally extract waveguide losses. By increasing the thickness of the oxide layer from 1 µm to 2 µm, propagation loss of 2 µm (1 µm) wide waveguide was reduced from 1.85 dB/cm (3 dB/cm) to as low as 0.37 dB/cm (0.77 dB/cm), and, resonators with quality factor greater than 1 million were demonstrated. An oxide thickness of approximately 1.5 µm is sufficient to significantly reduce propagation loss due to leakage into the substrate and simultaneously attain good electromechanical coupling in acoustic devices. This work not only provides insights on the design and realization of low-loss photonic waveguides in lithium niobate, but most importantly offers experimental evidence on how the oxide thickness directly impacts losses and guides its selection for the synthesis of high- performance acousto-optic devices in Y-cut lithium niobate on insulator. 1. Introduction Lithium niobite (LN), a versatile optical material which possesses outstanding properties including nonlinear optical, electro-optical, piezoelectric, and acousto-optical (AO) effects and a wide transparency range (350 nm ~ 5.2 μm) [1], has wide application in telecommunication networks, sensors, frequency conversions, quantum optics and microelectromechanical systems (MEMS) [2-5]. The emergence of lithium niobite on insulator (LNOI, thin film of LN bonded on low refractive index material) in the last two decades has led to rapid growth of LN photonics since its high index contrast waveguide structure enables dramatic reduction of the device footprint and enhancement of optical effects with respect to implementations in the bulk material [6]. Substantial cutting-edge research activities have flourished based on compact thin film LN photonic devices for a variety of applications [7-15]. For AO applications, LN features high elasto-optic coefficients like GaAs, but comes also with very strong piezoelectricity over many other materials (e.g., GaAs, ZnO, quartz…) [16], which results in acoustic devices that have a substantially higher electromechanical coupling coefficient [17] However, compared to more mature material platforms like silicon on insulator (SOI) and other semiconductor materials, producing low loss waveguides in LNOI is extremely challenging. Dry etched waveguides have exhibited high propagation loss due to rough sidewalls or byproduct formed by chemical reaction [18,19]. Hybrid waveguides consisting of other material loaded on top of the LN thin film avoid etch issues but confine only a portion of the light in the active LN layer, hence they do not fully harness the material capabilities [20]. Only recently, low-loss subwavelength photonic waveguides have been attained by optimizing plasma etching conditions and coating over-cladding layers on X-cut and Z-cut LNOI [21,22]. Despite the significant progress, none of these prior demonstrations directly relates the achievable low loss in LNOI waveguides to the thickness of the buried oxide layer (BOX), but rather focus on the etch method to define the waveguides. It is important to note that, for AO devices, the buried oxide layer (BOX) has a direct impact on the 2) and thicker films of oxide negatively impact the electromechanical coupling coefficient (kt performance of MHz devices such as those used for inertial sensing applications [13]. Therefore, it is well worth studying the dependence of the waveguide loss on the BOX thickness on LNOI as that would also impact the design of high-coupling AO devices. Differently from prior demonstrations, we study this dependence in Y-cut LN as it is one of the most appropriate cuts for the implementation of acousto-optic (AO) devices. In fact, in this cut, surface acoustic waves (SAW) can be efficiently excited by taking advantage of the high kt 2 of the film and result in resonators with high quality factor [17]. Both simulations and experiments were conducted to study the dependence of waveguide losses on the thickness of the BOX layer. As predicted by simulations, the waveguide loss extracted from racetrack (RT) resonators decreased with increasing the BOX thickness. Low- loss waveguides (0.33 dB/cm) and high Q resonators (>1,000,000) were fabricated by argon plasma etching followed by RCA cleaning. Most importantly, the results show that a thickness of 1.5 µm could be used as a good compromise between photonic and acoustic performance. 2. Simulations 2.1 Electromechanical coupling coefficient (kt 2) as a function of oxide thickness 2) for various TBOX and acoustic Fig. 1. Simulated electromechanical coupling coefficient (kt wavelengths (Λ) on a YZ cut LNOI (see inset for wafer stack). The presence of oxide clearly 2. It is interesting to note how the impact changes with has a deleterious impact on the device kt the specific wavelength. This is due to changes in penetration of the acoustic and electric fields in the thin films of LN, oxide and the thick LN substrate. At larger acoustic wavelengths, the oxide thickness is a small fraction of the acoustic devices. At intermediate wavelengths, the 2 is more dramatic. At oxide becomes a dominant part of the active SAW and its impact on kt the smallest wavelength, the acoustic and electric fields are almost entirely confined in the thin film of LN and the oxide has a lower impact on kt 2. 2, broadly defined as the ratio of the conversion between electrical and acoustic energy in a kt piezoelectric transducer, is an important figure of merit that determines the selection of a 2 is also of paramount importance in AO piezoelectric material for acoustic applications. kt 2) and the devices, since it ultimately determines the device size (inversely proportional to kt 2 of SAW devices in bulk LN effectiveness with which an acoustic wave interacts with light. kt can be readily computed via numerical methods [17]. We extended the same methodology to 2 of SAW devices for LNOI by using finite element methods in COMSOL derive the kt 2 as a function of different BOX Multiphysics. We investigated the dependence of kt thicknesses (TBOX) and acoustic wavelengths (Λ) for a Y-cut LNOI substrate. Y-cut LN was 2 and selected because it is one of the preferred cuts for SAW applications given the larger kt lower acoustic losses that can be attained [17]. In this analysis, a z-propagating SAW wave was simulated in a thin film of LN having a thickness of 500 nm, a variable oxide thickness and a thick LN substrate (see Fig. 1 inset for material stack). The thickness of the top LN layer was set at 500 nm as the preferred value for the making of photonic waveguides with a well-confined quasi-TE00 mode around 1550 nm. It is important to note that the presence of a thick LN substrate below the oxide is important for the effective excitation of large acoustic wavelength devices. In fact, if such substrate was not present or substituted with silicon, then large wavelength SAWs would not be effectively excited. The results of this analysis are 2 is plotted versus TBOX for different Λ. Although TBOX has a different shown in Fig. 1 where kt 2 depending on the specific wavelength (see Fig. 1 caption for further impact on the kt explanation), it is clear that the presence of TBOX has a deleterious effect on the electromechanical coupling of the SAW device. Therefore, depending on the specific application, the oxide thickness should be carefully selected and should not generally be made to exceed 2 µm except for particular acoustic wavelengths. In the following sections, we detail the impact of this same oxide thickness on photonic losses for the same Y-cut LNOI, so as to provide guidance in the design of high-performance AO devices. 2.2 Propagation losses as a function of oxide thickness The schematic cross-section of a partially etched waveguide and the corresponding mode profile are shown in Fig. 2(a) and (b), respectively. The waveguide cross-section is the y-x crystal plane of the film and light propagates in the z-direction. We performed a full-vectorial finite difference simulation using a commercial software, Lumerical, to compute the propagation losses in this waveguide structure [23]. A perfectly matched layer (PML) boundary was used to enclose the waveguide region and absorb any incident electromagnetic field at its boundary. Both LN and BOX were defined as dielectric materials without intrinsic loss, hence the only accountable losses in the simulations were coming from the optical field decaying through the BOX and coupling to a radiation mode in the substrate (absorbed by the PML in the simulation framework). The contour plot in Fig. 2(c) shows that loss can be reduced by either increasing TBOX or the width of the rib (W). The rationale behind this behavior can be explained by looking at the optical field exponential decay in the substrate direction, ~ exp(-αL), where α is the decay rate and L is the decay length. As TBOX increases, the decay length, L, increases, while as W increases a higher effective refractive index (neff) waveguide is formed thus enhancing the decay rate, α. Losses of 0.01 dB/cm, 0.1 dB/cm, 1 dB/cm and 5 dB/cm are marked in the contour plot of Fig. 2(c) to roughly highlight what minimum oxide thickness is required in order to achieve such level of losses for a given waveguide width. We also calculated the dependence of losses on Tbox and W for bent waveguides with a bending radius of 20 μm as shown in Fig. 2 (d). Clearly, because of the additional radiation losses the bent waveguides exhibit higher losses but similar trends. For large radius, e.g., the 100 μm we used in the devices reported in this work, the dependence of losses on Tbox and W is practically the same as that for straight waveguides as illustrated in Fig. 2(c). Fig. 2. (a) Schematic cross-section of the Y-cut LNOI rib waveguide. PML boundary condition was applied in the simulation to represent radiation losses into the substrate. (b) E intensity distribution of the fundamental TE-like mode of the waveguide with W = 1 μm and Tetch = 300 nm. Calculated dependence of loss on W and TBOX for (c) straight waveguide and (d) bent waveguide (radius = 20 μm). 3. Experimental demonstration of low loss waveguides in Y-cut lithium niobate Methods such as the cut-back and Fabry-Perot (FP) interference [24,25] are commonly used to extract waveguide losses. The cut-back method needs consistent coupling efficiency from the fiber to the waveguide to accurately extract loss, while the FP method needs a perfectly polished end-face to avoid the impact of coupling losses on the extraction process. An alternative way to extract loss that can cancel out the role of input/output coupling efficiency is measuring the losses (extracted from the measured Q factor) of optical racetrack (RT) resonators with different dimensions. Waveguide losses can then be extrapolated from the propagation loss of the straight waveguide forming the RT [21]. In the following two sections, we will discuss the details on the fabrication and optical characterization of RT resonators and the extraction of waveguide losses in Y-cut LNOI substrates. 3.1 Fabrication The fabrication flow is shown in Fig. 3(a). We used three Y-cut LNOI samples with 1 μm, 1.5 μm and 2 μm thick BOX respectively, to explore the dependence of waveguide loss on BOX thickness. First, photonic patterns including grating couplers (GC), feeding waveguides ("U" shaped) and RT resonators, were patterned on CSAR 62 positive resist by electron-beam lithography. Among all the techniques to produce photonic devices on LNOI, physical etching using Ar plasma features high anisotropic etching profile. Therefore, we etched the LNOI film by inductively coupled plasma reactive ion etching (ICP-RIE) using the following parameters: Ar flow of 30 sccm, bias power of 100 W, ICP power of 600 W and pressure of 5 mT. The etching depth (Tetch) was about 300 nm. This recipe exhibited a selectivity to CSAR 62 resist of 1:1. Fig. 3. (a) Fabrication flow of LNOI photonic devices. The LNOI wafer was purchased from NGK Insulators, LTD. (b) SEM pictures of GC and waveguide before RCA cleaning. The redeposition of LN along the sidewalls of the patterned features is clearly visible. (c) Microscope picture of one of the fabricated RT resonators (upper right) and SEM pictures of various photonic components after RCA cleaning. The waveguide is 2 μm wide and the coupling region (lower right) between the waveguides features a gap of 200 nm. The sidewall is very smooth after RCA cleaning. θ is the angle between the propagation direction of the straight waveguide and the crystalline z-axis (θ = 0° for the RT shown in the microscope picture). The electron-beam resist left on the sample was removed by oxygen plasma cleaning right after etching. The final step included cleaning the etched patterns by RCA (NH4OH, H2O2 and H2O mixed by the volume ratio of 1:1:5) at 60°C for 30 min. This RCA step is critical as it removes the organic residues and insoluble particles left behind by the etch step by changing their zeta potentials [26]. From the SEM images of the etching profile before and after RCA cleaning (Fig. 3(b) and (c)), it was obvious that the RCA cleaning effectively stripped the byproducts of the LN etch process and any residual resist that was stuck to the waveguide sidewalls. 3.2 Characterization To characterize RT resonators, we coupled light in/out of the device by GCs shown in Fig. 3(c). The insertion loss of each GC was measured to be 7 dB at the wavelength of 1550 nm around which we characterized the RT resonators. By sweeping the wavelength of the tunable laser, the transmission spectra of RT resonators were recorded by a power-meter. The RT resonators had fixed bend radius (R = 100 μm) but varying straight waveguide lengths (L = 300 μm, 700 μm, 2000 μm and 4000 μm. Total length of RT being Ltot = 2πR+2L) and widths (W = 2 μm and 1 μm). First, we measured the transmission spectra of these RTs with different lengths and then extracted their intrinsic losses (loss in a single-pass trip besides coupling loss), α, by fitting the experimental curves to a Lorentzian function. The transmission of RT resonators (L = 4000 μm) fabricated on a 2 μm thick BOX (TBOX = 2 μm) with W = 2 μm and 1 μm are shown in Fig. 4(a) and (b), respectively. The unloaded Q (QU) for these devices are higher than 1,000,000 for the case in which W = 2 μm. By comparing α from RTs with different L (or Ltot), we could eliminate the impact of bent waveguide loss and get the straight waveguide losses by simply looking at the slope of the fitted curve plotting loss vs. RT length (see Fig. 4(e) and (f)). The propagation loss of the straight waveguide with W = 2 μm (1 μm) was as low as 0.37 ± 0.02 dB/cm (0.77 ± 0.07 dB/cm), which was among the lowest propagation loss reported so far for dry-etched LNOI waveguides with similar dimensions and without any over-cladding oxide layer [27]. The bending loss (100 μm radius) for W = 2 μm (1 μm) is extracted to be 1.1 ± 0.1 dB/cm (1.55 ± 0.3 dB/cm). Using an over- cladding layer could further reduce the scattering loss due to the lower index contrast between LN and the surroundings but requires additional steps (e.g. lithography and etching) to open windows to enable electrical contacts to external signal sources if such waveguides are to be used in practical applications for electro-optic (EO) or acousto-optic (AO) modulators. Fig. 4. (a) and (b): Measured (black dots) and fitted (red curves) transmission of RT resonator (L = 4000 μm) with W = 2 μm and W = 1 μm, respectively. (c) and (d): Loaded and unloaded Q factors in RT resonator as a function of the total length (Ltot = 2πR+2L) for 2 μm and 1 μm wide straight waveguides respectively. (e) and (f): Intrinsic loss in RT resonator as a function of the total length. Average propagation losses of 0.37 dB/cm and 0.77 dB/cm for 2 μm and 1 μm wide straight waveguides were extracted by linearly fitting the experimental data. The variations in the value of the average propagation losses are set equal to one standard deviation of the losses extracted from several resonances around 1550 nm. Using the same method reported above for extracting propagation losses, we measured the dependence of waveguide losses on different thicknesses of the BOX layer (TBOX = 1 μm, 1.5 μm). The measurement results are summarized in Table 1. It is clear that waveguide loss decreases with increasing TBOX independently of the width of the waveguide as predicted by simulations. Note that, for small TBOX, i.e. TBOX = 1 μm, the leakage loss dominates over all other loss sources in the waveguides because the simulated value closely matches the experimental data. As TBOX increases, the energy leakage into the substrate becomes negligible and other loss factors like scattering loss become dominant. In particular, for the cases of TBOX = 2 μm and 1.5 μm, losses displayed in the table can be almost exclusively attributed to scattering losses since the leakage obtained from simulation is much less than the measured loss. This is a very important finding as it points out that TBOX of ~ 1.5 µm is sufficient the electromechanical performance of SAW devices. low propagation to ensure loss without significantly impacting Since the design of high-performance EO or AO devices requires rotation of the waveguide with respect to the X-axis of the Y-cut crystal [17], we also measured the loss of the waveguide with TBOX = 2 μm for other two propagation directions (θ = 45° and 90°). We summarize these results in Table 1. For the case of W = 1 μm, it appears that the losses increase as θ increases from 0° to 90°. We speculate that since the value of neff lowers as θ increases due to the birefringence of LN, then the smaller waveguides tend to suffer more from scatting losses as light experiences greater interaction with the waveguide sidewall. Table 1. Summary of the waveguide propagation losses (Unit: dB/cm) W = 1μm TBOX = 1μm θ = 0° 3 ± 0.5 W = 2μm 1.85 ± 0.4 4. Conclusions TBOX = 1.5μm θ = 0° 0.94 ± 0.01 TBOX = 2μm θ = 0° θ = 45° θ = 90° 0.77 ± 0.07 0.83 ± 0.08 1.09 ± 0.07 0.59 ± 0.03 0.37 ± 0.02 0.23 ± 0.01 0.34 ± 0.02 In conclusion, we experimentally investigated the dependence of waveguide propagation loss on BOX thickness, waveguide width and in-plane orientation for Y-cut LNOI. As predicted by FDTD simulations, the measured loss decreased with increasing TBOX and reached values < 0.4 dB/cm when TBOX was selected to be 2 μm. By comparing the simulated and measured losses, we can conclude that leakage losses are dominant when TBOX is small, while, other loss sources (i.e. scattering loss) are more relevant when TBOX is large. Effectively, this work shows that for oxide thicknesses exceeding 1.5 µm, propagation losses due to substrate 2 of leakage can be dramatically reduced. Since, as evidenced by COMSOL simulations, the kt SAW devices tends to deteriorate as TBOX increases, 1.5 µm of oxide could be considered as a good compromise for the making of high-performance AO devices. Ultimately, the selection of the oxide thickness will depend on the specific application and acoustic frequency for which the AO device is used. Nonetheless, this work reports important experimental insights on how propagation losses are affected by BOX thickness in Y-cut LNOI and offers general guidelines in the selection of the BOX thickness for high performance AO devices such as the inertial sensor reported in [13]. Funding This material is based upon work supported by the DARPA PRIGM-AIMS program under Award No. N66001-16-1-4025. Any opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of DARPA. Acknowledgments Authors acknowledge S. Hiramatsu and Dr. Yuji Hori from NGK Insulators, LTD. for providing the LNOI wafers. References 1. L. Arizmendi, "Photonic applications of lithium niobate crystals," Phys. Status Solidi A 201(2), 253-283 (2004). 2. P. Rabiei and W. H. Steier, "Lithium niobate ridge waveguides and modulators fabricated using smart guide," Appl. Phys. Lett. 86(16), 161115(2005). 3. K. R. Parameswaran, R. K. Route, J. R. Kurz, R. V. Roussev M. M. Fejer, "Highly efficient second-harmonic generation in buried waveguides formed by annealed and reverse proton exchange in periodically poled lithium niobate," Opt. Lett. 27(3), 179-181 (2002). 4. H. Jin, F. M. Liu, P. Xu, J. L. Xia, M. L. Zhong, Y. Yuan, J. W. Zhou, Y. X. Gong, W. Wang, and S. N. Zhu, "On-Chip Generation and Manipulation of Entangled Photons Based on Reconfigurable Lithium-Niobate Waveguide Circuits," Phys. Rev. Lett. 113(10), 113601 (2014). 5. Y. Song and S. Gong, "Wideband Spurious-Free Lithium Niobate RF-MEMS Filters," J. Microelectromech. Syst. 26(4), 820-828 (2017). 6. G. Poberaj, H. Hu, W. Sohler, and P. Günter, "Lithium niobate on insulator (LNOI) for micro-photonic devices," Laser Photonics Rev. 6(4), 488-503 (2012). 7. A. Mercante, S. Shi, P. Yao, L. Xie, R. Weikle, and D. Prather, "Thin film lithium niobate electro-optic 8. modulator with terahertz operating bandwidth," Opt. Express 26(11), 14810 -- 14816 (2018). P. Weigel, J. Zhao, K. Fang, H. Al-Rubaye, D. Trotter, D. Hood, J. Mudrick, C. Dallo, A. Pomerene, A. Starbuck, C. DeRose, A. Lentine, G. Rebeiz, and S. Mookherjea, "Bonded thin film lithium niobate modulator on a silicon photonics platform exceeding 100 GHz 3-dB electrical modulation bandwidth" Opt. Express 26(18), 23728 -- 23739 (2018). 9. R. Wu, J. Zhang, N. Yao, W. Fang, L. Qiao, Z. Chai, J. Lin, and Y. Cheng, "Lithium niobate micro-disk resonators of quality factors above 107," Opt. Lett. 43(17), 4116-4119 (2018). 10. H. Lu, B. Sadani, N. Courjal, G. Ulliac, N. Smith, V. Stenger, M. Collet, F. I. Baida, and M.-P. Bernal, "Enhanced electro-optical lithium niobate photonic crystal wire waveguide on a smart-cut thin film," Opt. Express 20(3), 2974-2981 (2012). 11. L. Chen, Q. Xu, M. G. Wood, and R. M. Reano, "Hybrid silicon and lithium niobate electro-optical ring modulator," Optica 1(2), 112 -- 118 (2014). 12. A. Rao, J. Chiles, S. Khan, S. Toroghi, M. Malinowski, G. F. Camacho-González, and S. Fathpour, "Second- harmonic generation in single-mode integrated waveguides based on mode-shape modulation," Appl. Phys. Lett. 110(11), 111109 (2017). 13. M. Mahmoud, A. Mahmoud, L. Cai, M. Khan, T. Mukherjee, J. Bain, and G. Piazza, "Novel on chip rotation detection based on the acousto-optic effect in surface acoustic wave gyroscopes," Opt. Express 26(19), 25060- 25075 (2018). 14. L. Cai, A. Gorbach, Y. Wang, H. Hu, and W. Ding, "Highly efficient broadband second harmonic generation mediated by mode hybridization and nonlinearity patterning in compact fiber-integrated lithium niobate nano- waveguides," Sci. Rep. 8, 12478 (2018). 15. Z. Hao, J. Wang, S. Ma, W. Mao, F. Bo, F. Gao, G. Zhang, and J. Xu, "Sum-frequency generation in on-chip lithium niobate microdisk resonators," Photon. Res. 5(6), 623 -- 628 (2017). 16. J. Xu and R. Stroud, Acousto-optic devices: principles, design, and applications (Wiley, 1992). 17. J. J. Campbell and W. R. Jones, "A method for estimating optimal crystal cuts and propagation direction for excitation of piezoelectric surface waves," IEEE Trans. Sonics Ultrason. SU15 (4), 209-217 (1968). 18. S. Y. Siew, E.J.H. Cheung,H. Liang,A. Bettiol, N. Toyoda, B. Alshehri, E. Dogheche, and A.J. Danner, "Ultra- low loss ridge waveguides on lithium niobate via argon ion milling and gas clustered ion beam smoothening," Opt. Express, 26(4), 4421 -- 4430 (2018). 19. H. Nagata, N. Mitrugi, K. Shima, M. Tamai, and E.M. Haga, "Growth of crystalline LiF on CF4 plasma etched LiNbO3 substrates," J. Cryst. Growth 187(3-4), 573-576 (1998). 20. L. Chang, Y. F. Li, N. Volet, L. Wang, J. Peters, and J. E. Bowers, "Thin film wavelength converters for photonic integrated circuits," Optica 3(5), 531-535 (2016). 21. M. Zhang, C. Wang, R. Cheng, A. Shams-Ansari, and M. Lončar, "Monolithic ultra-high-Q lithium niobate 22. microring resonator," Optica 4(12), 1536 -- 1537 (2017). I. Krasnokutska, J.-L. J. Tambasco, X. Li, and A. Peruzzo, "Ultra-low loss photonic circuits in lithium niobate on insulator," Opt. Express 26(2), 887 -- 894 (2018). 23. Lumerical Solutions: http://www.lumerical.com/ 24. B. Yang, L. Yang, R. Hu, Z. Sheng, D. Dai, Q. Liu, and S. He, "Fabrication and Characterization of Small Optical Ridge Waveguides Based on SU-8 Polymer," J. Lightw. Tech. 27(18), 4091 -- 4096 (2009). 25. L. Cai, R. Kong, Y. Wang, and H. Hu, "Channel waveguides and y-junctions in x-cut single-crystal lithium niobate thin film," Opt. Express 23(22), 29211-29221 (2015). 26. M. Itano, F. W. Kern, M. Miyashita, and T. Ohmi, "Particle removal from silicon wafer surface in wet cleaning process," IEEE Trans. Semicond. Manuf. 6(3), 258 -- 267 (1993). 27. R. Luo, Y. He, H. Liang, M. Li, and Q. Lin, "Highly tunable efficient second-harmonic generation in a lithium niobate nanophotonic waveguide," Optica 5(8), 1006-1011 (2018).
1909.11053	1	1909	2019-09-24T17:10:45	Cavity receiver designs for parabolic trough collector	[ "physics.app-ph" ]	Solar parabolic trough collector technology has received significant attention among researchers for its economic potential to meet electrical and thermal energy demands. Furthermore, parabolic trough collector power plants can be hybridized with fossil fuels or other renewable power plants. Improving the efficiency of the parabolic trough collector leads to increase the electricity production and allows the parabolic trough collector power plant to be built in a more compact size, at reduced capital investment. This review focuses on parabolic trough collector receivers, which uses cavity technology to improve efficiency. The cavity receiver has shown the potential to overcome shortages of the conventional receiver. It was seen that the thermal losses of the cavity receiver are affected by various parameters such as aperture sizes, emissivities, and working fluid temperature as well as mass flow rate and wind velocity. Different cavity designs have a different effect on the solar flux distribution along the circumference of the receiver. An efficient cavity receiver still has the potential to minimize the losses and hence, improve the overall efficiency. Some of the design gaps have been identified as a guide for future work.	physics.app-ph	physics	Cavity receiver designs for parabolic trough collector aSchool of Physics, Materials for Energy Research Group, University of the Witwatersrand, Johannesburg, 2001, South Africa. bSchool of Physics, Mandelstam institute for Theoretical Physics, University of the Witwatersrand, Johannesburg, 2001, South Africa. Khaled Mohamada, P. Ferrera,b Solar parabolic trough collector technology has received significant attention among researchers for its economic potential to meet electrical and thermal energy demands. Furthermore, parabolic trough collector power plants can be hybridized with fossil fuels or other renewable power plants. Improving the efficiency of the parabolic trough collector leads to increase the electricity production and allows the parabolic trough collector power plant to be built in a more compact size, at reduced capital investment. This review focuses on parabolic trough collector receivers, which uses cavity technology to improve efficiency. The cavity receiver has shown potential to overcome shortages of the conventional receiver. It was seen that the thermal losses of the cavity receiver are affected by various parameters such as aperture sizes, emissivities, and working fluid temperature as well as mass flow rate and wind velocity. Different cavity designs have a different effect on the solar flux distribution along the circumference of the receiver. An efficient cavity receiver still has the potential to minimize the losses and hence, improve the overall efficiency. Some of the design gaps have been identified as a guide for future work. Keywords: Cavity absorber; Heat transfer fluid; Parabolic trough collector; Receiver unit. 1. Introduction Among the concentrated solar power technologies, the Parabolic Trough Collector (PTC) is the earliest and most widely accepted design for harnessing solar energy for the dispatchable source (it refers to the source of energy that can be used on demand or at the request of power grid operators) of electricity or industrial and chemical, due to its low cost and reliability [1]. In this technology, solar radiation is focused onto a focal line aimed at the receiver unit. The receiver heats up and in turn, imparts a significant portion of its heat to a Heat Transfer Fluid (HTF) circulating within. The hot HTF can be used to generate electricity through a steam cycle or in thermochemical applications. The receiver unit is one of the most complex parts of the Parabolic Trough Collector (PTC). Its optical and thermal properties directly influence the performance of the entire system, and the efficiency of the whole system largely depends on it [2]. Therefore, the receiver unit has to be carefully designed in such a way to minimize energy losses. The conventional type of the PTC receiver unit for PTC is evacuated glass metal. It consists of a blackened absorber pipe encapsulated by the glass cover. An evacuated region in between minimizes convective heat losses, see Figure 1. Figure 1: Structure of a parabolic trough receiver [3]. The thermal contacts between the receiver pipe and the glass cover are kept to a minimum to reduce conduction heat losses. The conventional receiver has advantages such as high efficiency and low heat loss. However, it has some deficiencies, such as: • The breakage of the glass and the metal glass seal, which leads to vacuum loss in the annular region between the absorber pipe and the glass cover [4]. • The permeation of hydrogen gas into the annular region is due to the thermal decomposition of the organic heat transfer fluids (HTF) such as Therminol VP-1. The hydrogen gas in the annular region is found to increase the heat losses to four-time those in the evacuated receiver [5]. • Lower efficiency at high temperatures [6][7]. • Non-uniform thermal distribution and thermal stress problems [7][8]. • High capital outlay and high maintenance cost [9][10]. Increasing the outlet HTF temperature is one of the main challenges to improve the overall efficiency of the PTC power plant and to reduce the Levelized Cost of energy and the combined thermal storage. For example, thermal storage quantity can be reduced to one third, if the HTF temperature increases from 350 °C to 600 °C [5]. Currently, most of the power plants used oil-based fluids, which operate to a limited temperature up to 400 °C. The dominant heat losses mechanism at high temperatures are due to the thermal emission (IR) from the receiver pipe. This loss is conventionally minimized by painting the receiver pipe with a spectrally selective coating. It is a material that absorbs well in the visible region of the solar spectrum and emits poorly in the IR region. Much work has been published on selective coatings and improving their properties [11]. Dudley et al. [12] investigated the performance of black chrome and cermet selective coatings, and found that the Cermet had lower emissivity values and thus reduced thermal losses and improved efficiency. Forristal [13] compared six different selective coating materials to evaluate receiver unit performance. He showed that the receiver performance is susceptible to the optical properties of the selective coatings, and improving the coatings could result in significant efficiency gain. Cheryl et al.[14] Investigated spectrally selective coating materials for concentrated solar power applications, and concluded that the ideal selective coating material should be easy to manufacture, low-cost, chemically and thermally stable in air at operating temperature of 500 °C. Because the selective coatings are placed on absorber pipe of the receiver, the receiver operation temperature is limited and hence thermal efficiency. Kennedy et al. [15] were able to successfully model a solar-selective coating composed of materials stable to 500 °C using computer- aided design software. Archimede Solar Energy (ASE) [16] manufactured the world's most advanced solar receiver tube with selective coating. It operates at temperature up to 580 °C with molten salts as HTF. Further, the solar receiver SCHOTT PTR 70 is designed for solar thermal power plants operating with oil-based HTF at a temperature up to 400 °C [17]. The heat loss measurements for SCHOTT PTR 70 are carried out in a round-robin test performed by SCHOTT Solar in cooperation with Deutsches Zentrum für Luft-und Raumfahrt (DLR) and US National Renewable Energy Laboratory (NREL). The tests confirmed that the heat loss is less than 250 W/m at working temperature 400 °C [17]. Some complications start to appear at a high temperature (> 300 °C) and with a non-uniform solar flux distribution such as the thermal performance and thermal stress of the receiver unit. When a non-uniform heat flux profile is incident on the receiver unit, the temperature across the circumference varies, and peaks/hot spots in the receiver start to increase with temperature. It leads to bending of the absorber pipe and breaking of the glass cover. P Wang et al. [8] found that the maximum temperature gradient for the safe operation of receiver tubes is about 50 K. These complications have been addressed to increase the life span of the receive absorber. Some recent research focused on improving both thermal transfer and uniformity of the thermal distribution [7], but sacrifice pressure drop of the receiver unit or increase the quality of the absorber pipe and other components, adding cost [6]. Some of these studies suggest applying inserts into the absorber pipe such as metal foam, porous discs, perforated plates or coiled wire turbulators inserts. A metal foam inserted into the absorber pipe facing the concentrates sunlight reduces the thermal stress, decreases the temperature difference on the outer surface of the absorber pipe by about 45%, but increases flow resistance [8]. Experimental [18] and theoretical work [19] was conducted for the porous disc insert application (a disc perforated with holes inserted into the pipe), increasing the thermal efficiency between 1.2% and 8% according to the numerical study [20]. The coiled wire turbulators insert application has been examined experimentally and numerically [21]. At the pitch distance 30 mm of coiled wire turbulator (a coil-shaped wire inserted into the absorber pipe), the heat transfer enhancement is approximately twice that of the smooth tube [21]. Furthermore, Eliamasa-ard, Thianpong and Eiamsa-ard [22] studied the effects of three different twisted tapes inserts on thermal enhancement, Nusselt number (Nu) and friction factor (f), which are a single twisted tape, twin-counter (co- twisted) tape and counter (co-swirled) tape. The experiments of four different twist ratios (2.5, 3, 3.5, and 4) were conducted. The twist ratio is the ratio of the axial length for 180° turn and the inner diameter of the absorber tube. The results showed that the thermal enhancement, Nu and f were increased with decreased the twist ratio, and co-swirled tape has significant heat transfer enhancement compared to co-twisted tape. Other research works investigated the heat transfer performance and pressure drops of the absorber pipe with wire-coil inserts. It was found that wire coil inserts increased the turbulence inside the absorber pipe and Nu increased to 330% [23][24]. In addition, the effect of the wire-coil inserts material types (aluminum, stainless steel and copper) on f and the heat transfer rate were studied by Shashank and Choudhari [25]. They found that the copper wire coil inserts have improved the heat transfer rate compared to stainless steel and aluminum. Nanan, Thianpong and Pimsarn [26] conducted experimental and numerical studies to investigate tube inserts with baffle turbulators (straight baffles, straight cross-baffles, straight alternate-baffles, twisted baffles, alternate twisted-baffles and twisted cross-baffles). The twisted cross-baffles showed the highest thermal performance, while the straight cross-baffles showed the lowest thermal performance. Yuxiang Hong, Du J and Wang S [27] performed experimental investigation of the thermal and fluid flow characteristics of a spiral grooved tube (SGT) fitted with twin overlapped twisted tapes in counter large/small combinations (TOTT-CL/S). The experiments were conducted for plain tube (PT), SGT and SGT fitted with TOTT-CL/S. The heat transfer rate of SGT fitted with TOTT-CL/S was higher than that of the SGT and PT but larger pressure drop. Other studies focus on geometrical structure improvement for the absorber pipe of the receiver unit such as a dimpled tube, unilateral milt-longitudinal vortex-enhanced tube as well as symmetric and asymmetric outward convex corrugated tubes. These inserts manipulate the Reynolds number, substantially improving the "mixing" of different temperature layers of fluid. A numerical study showed improved performance of a dimpled absorber pipe under non-uniform heat flux over uniform heat flux [28]. Similar improvements were found for the unilateral milt-longitudinal vortex-enhanced tube, with better heat transfer performance than the smooth pipe under a wide range of working conditions [29]. The introduction of symmetric [30] and asymmetric [31] outward convex corrugated tubes, regular outwards "bulges" in the absorber pipe, effectively decrease the thermal strain and enhance the heat transfer performance. Further, researchers have investigated Nanofluids (with suspended nanoparticles) to enhance the heat transfer. The most used nanofluids contain nanoparticles such as Al, Fe2O3, Al2O3, Cu, TiO2, and SiO2 [32][33]. Nanofluids tend to have more significant thermal conductivity than normal heat transfer fluids. The thermal conductivity increases by decreasing the particle size and increasing the volume fraction and temperature [33]. E. Bellos found that the use of the nanofluids increases the efficiency of the collector by 4.25% [32]. Some of the limitations associated with evacuated receivers have been overcome. Nevertheless, the monopoly over this technology and the cost of the receivers put barriers for solar projects, especially in the developing countries. Accordingly, finding alternatives to evacuated receivers that can compete in terms of efficiency and cost-effectiveness is a prime target. Such alternatives are expected to have a significant impact on the already long-standing industry. An alternative to the evacuated receiver is the cavity receiver, where the concept of the cavity receiver comes from the blackbody principle. The blackbody is an ideal object, absorbs all incident radiation, regardless of direction and wavelength [34]. The object that most closely resembles a blackbody is a large cavity with a small opening. The radiation that is incident through the opening has very little chance to escape, it is either absorbed or undergoes multiple reflections before being absorbed [34]. The cavity receiver that can achieve the blackbody principle could make up the shortages of the evacuated receiver, as mentioned earlier. Although early studies of the cavity receiver lacked detail in terms of optimizing, efficiency, and thermal losses, the recent researches are looking to improve and come up with innovative designs. In this paper, we review different cavity receiver designs. This review is intended for the receiver of a parabolic trough collector, but some designs are suitable for the receiver of the linear Fresnel reflector systems. This work aims to study different cavity designs. It is worthwhile to gather innovative ideas that help to enhance the overall efficiency and identify the gaps in these designs for future research work. In the following discussion, we divided the cavity receiver designs into cylindrical receiver units and non-cylindrical receiver units. 2. Cylindrical receivers The base design of the cavity receiver consists of a cylindrical metal tube with a cavity opening for the incident solar radiation. In most of the cavity receivers that have been studied, the space inside the cavity was at atmospheric pressure. The inner cavity surface opposite to the aperture window was mirrored. The aperture window of the cavity was at the bottom, facing the parabolic mirror collector and it was closed by a transparent cover to reduce convection and re-radiation heat loss. Moreover, the outer cavity surface was thermally insulated to minimize the effect of heat loss by convection and radiation to surroundings. A. Mirrored glass cover with uncoated aperture Ramchandra et al. [5] studied and optimized the cavity receiver shown in Figure 2 for minimum heat loss. The study intended to evaluate and compare different PTC cavity receiver alternatives using a validated numerical model. Their cavity receiver, shown in Figure 2, consists of a mirrored cavity surface facing the absorber pipe. The cavity aperture is closed by a transparent glass cover to reduce re-radiation and convection heat losses. The annulus between them is separated by an air gap at atmospheric pressure. An annular ring of microtherm between the absorber and glass envelope at both ends is placed to suspend the absorber [5]. The dimension of the cavity aperture was carefully selected alongside with the focal line of the collector to ensure that all reflected radiation enters the cavity through the aperture and to minimize inaccuracies in both tracking and directional errors due to the suns shape. The optimum annulus dimension between the absorber pipe and the cavity envelope was selected to reduce convection and conduction heat loss. The optimum dimension of the annulus largely depended on the diameter of the receiver components [35]. Insulation Air gap at atmospheric pressure Glass cover Aperture Absorber pipe Heat Transfer Fluid Figure 2: Schematic representation of the linear cavity receiver[5]. Although this cavity receiver with a parabolic trough of small rim angle (around 45°) was capable of being a suitable alternative to the conventional receiver, it had some limitations. First, the intercept factor (the incident energy that enters the aperture and reaches the absorber) was lower than that of a conventional receiver. This problem was addressed by coupling this cavity design with a lower rim angle (around 45°) for the parabolic trough. The conventional receiver had a maximum concentration at rim angle = 90°, whereas the concentration ratio for this design was maximum at rim angle = 45° [5]. Second, the selective coating failed because the air in the annular gap oxidized the coating. The oxidation of the selective coating in the presence of the air in the annulus was expected to be resolved with the progress in developing selective coatings [36]. Third, at a higher temperature, the thermal conductivity of the insulation material increased, thus the heat loss also increased [35]. This issue was solved by selecting a better-suited insulation material. B. Hot-mirror coated glass cover A similar philosophy to the above mention design by Ramchandra et al. [5] of a cavity receiver that evolved from applying a reflective mirror coating on the inner side of the glass cover to trap the radiation inside the receiver. Ferrer et al. [37] [38] suggested using a hot mirror coating, a dielectric material that is transparent to the visible region of the solar spectrum and reflected well in the IR region. The study intended to reduce the heat losses, especially at a high temperature and reduce the thermal stress of the glass cover as well as introducing an alternative to the conventional receiver with a selective coating. In this suggested design, the annulus between the absorber pipe and the coated glass cover was under vacuum, see Figure 3. Hot mirror coating was first implemented for energy-efficient windows in automobiles and buildings [39] and for applications related to concentrating photovoltaics and thermophotovoltaics [40][41]. There are two general types of hot mirror films: a semiconducting oxide with a high doping level and a very thin metal film sandwiched between two dielectric layers (see [42][43][39] for more details). The thin metal film coating shows some unavoidable losses, while the highly doped semiconducting oxide shows more advantages, i.e., Indium-Tin-Oxide (ITO). The hot mirror coating for a solar collector must meet some performance specifications. It needs to be highly transparent in the visible region and have high reflectivity in the IR region of the solar spectrum. Granqvist et al. [46] and Lampert et al. [44] focused on improving the transparency in the visible region and the reflectivity in the IR. For PTCs, the effects of the hot mirror film have been modeled and studied previously, see Figure 3. Grena [43] simulated the system, including heat reflection using hot mirror films with simplifying assumptions, and his results showed an increase in overall efficiency tested over a year by 4%. Also, a 2D simulation in this regard showed the possibility of increasing the working fluid's temperature to over 400 °C [45]. Other efforts of this type used a three-dimensional model to take into account the radiation exchange by using different segmented surfaces inside the receiver along the pipe's length. This study showed that the hot mirror receiver effectively reduced the IR losses at higher temperatures, reduced the thermal stress on the glass cover, and suggested the use in a hybrid system [37]. Incoming concentrated solar radiation vacuum Glass cover with hot mirror coating from the inside (HM) Absorber pipe (AP) Heat Transfer Fluid (HTF) Figure 3: The cross-section of the receiver unit with a hot mirror coating on the glass cove envelop [37]. Practically, current hot mirror coatings have not been found to improve on the overall efficiency of solar plants, but have the capacity of reaching much higher temperatures than their selective coating counterpart. Hot mirror type systems are currently investigated, which may, in the near future, surpass selective coating technology [37]. If a hot mirror coating was found with better visible transmission/IR reflection properties than a selective coating, then the hot mirror coating would have some advantages. C. Mirrored cavity receiver with a hot mirror application on the aperture K Mohamad and P Ferrer filed a patent on a cavity receiver for PTC [46][6]. This cavity receiver design combined the use of the mirrored cavity receiver with a hot mirror coating in a novel way [6]. The design aims to make up the shortages of the conventional receiver and achieve higher efficiency, higher HTF temperature, and significantly cut the costs per unit PTC. The receiver consists of a highly reflective hot mirror coating on the inside of the borosilicate glass cover on the cavity aperture. The inner cavity surface was coated with a highly IR reflective material, such as polished aluminum. A vacuum in between minimized convective losses see Figure 4. The highly polished inner cavity surface reflects thermal radiation onto the absorber much more effectively than the hot mirror coating. The authors showed that the design was able to achieve the highest thermal and optical efficiencies, especially if the hot mirror coating has a higher transmission for the incident solar radiation. Novel aspects of the background theory for this design related to IR reflectivity in the receiver annulus were added and implemented in a simulation code [6]. The model of the IR reflectivity inside the receiver annulus was validated in [47][38]. The simulation results indicated that the proposed design could exceed the HTF temperature ceiling compared to existing alternatives and could hence potentially increase the efficiency of the system. Further, the cavity geometry and a hot mirror coating at the aperture enable increased retention of thermal radiation of the receiver. Glass cover with hot mirror coating (HM) Incoming concentrated solar radiation vacuum Opaque and highly reflective IR mirror (IRM) Absorber pipe (AP) Heat Transfer Fluid (HTF) Figure 4: the cross-section of the cavity receiver with a hot mirror coating on the cavity aperture [48]. The cavity system was simulated and studied in many aspects using the operating conditions and design parameters for SEGS (Solar Electric Generating System) LS2. It is one of three generations of parabolic troughs installed in the nine SEGS power plants in California [12]. Different aperture sizes were simulated. The smallest aperture reached the highest temperature and thermal efficiency. The maximum HTF temperature for this design rose close to 1300 K if the aperture opening size was 50% of the total size of the outer receiver circumference and 1490 K if the aperture opening size was 15% of the outer receiver circumference size. The receiver efficiency of the smallest aperture was at 33% compared to 27% for the largest opening, at receiver length of 400 m. Furthermore, different reflectivities for the inner surface of the cavity were studied (92%, 95%, and 98%). It was seen that the highest reflectivity reached the highest temperature, at receiver length of 500 m (about 200 K higher than the lowest reflectivity), and had the best receiver efficiency (33% compared to 27% for the lowest, at receiver length of 400 m). The effect of the hot mirror coating was studied by comparing the system with and without the hot mirror application. At the aperture opening size of 15% of the total size of the outer receiver circumference size and the reflectivity of the inner cavity mirror 98%, the stagnation temperature was found to be∼250 K higher with coating. The cavity system without hot mirror coating initially performed better at a lower temperature, but after 775 K, the hot mirror coated system dominated performance. Furthermore, a comparison between this cavity design (in case of a cavity opening with a hot mirror coating, an opening size of 15% of the total outer receiver circumference, and the cavity mirror reflectivity of 98%) and conventional receivers with a selective coating and without coating (bare) was conducted. The results indicated that the selective coating performed slightly better in terms of efficiency at a lower temperature, but the cavity system dominated at higher temperatures, see Figure 5 and Figure 6. Figure 5: HTF outlet temp for 375 K inlet temperature of different designs [6]. Figure 6: The total efficiency of the system as a function of the length of the receiver with different designs [6]. Although the cavity system seemed theoretically capable of reaching very high temperatures, it is unlikely that it will be used as such. A very high reflectivity (> 99%) inside the cavity walls, which is well within the realm of possibility, will make the system very efficient. Higher efficiency will likely allow the HTF to achieve optimum plant temperature in a shorter length, thus reducing the needed AP length. This system is likely to reduce costs and the overall size of the system [6]. D. Cavity receiver with asymmetric compound parabolic concentrator Roman Bader et al. [49] has proposed a cavity receiver, presented in Figure 7, with the aim to significantly cut the costs per unit PTC through a decrease in the PTC size and use of low-cost materials as well as using the air as the heat transfer fluid. The cavity design consists of an absorber tube enclosed by an insulated stainless steel cylindrical cavity with an asymmetric compound parabolic concentrator (CPC) at the cavity aperture, show in Figure 7. This design was tested with a 43 m long prototype installed and a 9 m aperture solar trough concentrator to study the efficiency of receiver under different operating conditions and validate the heat transfer model of the receiver that based on Monte Carlo ray-tracing and finite-volume methods [49]. Figure 7: Schematic representation of the cavity receiver design proposed by[49] At HTF inlet temperature of 120 °C, HTF outlet temperature was expected in the range of 250-450 °C, at summer solstice solar noon (input of 280 kW), while the efficiency of the receiver ranged from 45% to 29%. The optical losses were one- third of the total incident radiation on the receiver, due to the spillage at the aperture and reflection inside the cavity. The thermal losses due to the natural convection from the cavity insulator were 5.6 - 9.1%. Another heat loss was due to the re-radiation through the cavity aperture (6.1 - 17.6%). Moreover, HTF pumping work had an associated energy penalty of 0.6 -- 24.4% of the total power generated [50]. The authors suggested a modified cavity receiver design that evolved from the initial design, which is discussed next. E. Corrugated cavity receiver Roman Bader et al. [49] suggested modifications to their previous cavity design. The modified design is shown in Figure 8. It consists of a cylindrical cavity with a smooth or corrugated black inner surface with a single or double-glazed aperture window. The modification can be summarized as follows: The absorber pipe was eliminated to allow the HTF (air) to flow in a sufficiently large cross-section through the cavity to compensate for the lower volumetric heat capacity. The cavity from the inside (surface 1, Figure 8) was enhanced with V-corrugations to increase the heat transfer surface area. The cavity aperture was made from glass to reduce the convective heat loss at the aperture, where the glass is almost opaque for the radiation emitted from a blackbody at < 600 °C [49]. The two window panes of a double-glazed window with air in between were used to trap the emitted radiation of surface 1 and the inner window to reduce the heat conduction through the window. The authors studied this design with four different receiver configurations: smooth, V-corrugated absorber tube, single, and double-glazed aperture window. Figure 8: The modified design [42]. Using the air as a HTF in this design is to avoid the chemical instability during the operation and to allow the direct coupling of the solar collector with a packed bed thermal storage [51][52]. However, the air has a lower volumetric heat capacity, which leads to lower convective heat transfer compared to conventional HTFs, such as molten salt and thermal oils. This problem could be solved by designing a receiver with a larger diameter and higher heat transfer area than that of the conventional receiver [53]. In the modified design, V- corrugations were used to increase the heat transfer of the surface area. The simulation was validated with the experimental work for the design in Figure 7, and then used to analyze the modified design. The modified design Figure 8 was simulated during the summer solstice at solar noon in Sevilla, Spain, with direct normal solar irradiance equals 847 W/m2. It had a collector efficiencies between 60% and 65% at HTF temperature of 125 °C and between 37% and 43% at 500 °C. The largest source of energy loss was the optical loss, which was more than 30% of the incident solar radiation. It was mainly because of the absorption by the concentrators and reflection at the receiver's aperture window [49]. Moreover, the required pumping power for the HTF through a 200 m long receiver operated between 300 °C and 500 °C was between 11 and 17 kW. Furthermore, increasing the HTF mass flow rate would cause a decrease in heat loss with the corrugated absorber pipe, which led to an increase in receiver efficiency. At the expense of additional reflection loss, the double-glazed aperture window significantly reduced the re-radiation loss from the receiver's aperture compared to the single-glazed window, where the double-glazed acted as an effective radiation trap. However, the single-glazed receiver led to higher collector efficiencies at low HTF temperature (< 300 °C), the double-glazed receiver led to higher collector efficiency at high HTF temperatures. The author suggested using a material with a high reflectivity on the aperture window to improve the overall efficiency of the PTC [49]. F. Cavity receiver with copper pipes and copper annulus Barra et al. [54] proposed a different design of the cavity receiver shown in Figure 9. The motivation of their study was to design a blackbody cavity receiver more efficient and less expensive than the conventional receiver units. The structure of that cavity receiver was made from iron and copper pipes, which were selected for economic and stability reasons. The iron oxidation was to induce high visible absorption in the cavity. The V-shape Pyrex glass in the cavity opening was removable and selected for low-cost commercial availability and to reduce the convection loss. This design was tested with 50 m2 parabolic trough prototype, and the experiment parameters and results used to build a mathematical model to simulate the system. The working conditions of this cavity receiver were not optimum, but the performance of that design appeared promising compared to those obtained from a more expensive receiver. The study showed less solar interception with the cavity and more thermal losses, which were strongly depending on wind intensity and direction. Furthermore, the cavity receiver without a vacuum and selective treatments reached a good performance, but still inferior to the current receivers with highly selective coating and vacuum. The author suggested that the cavity receiver needed a design improvement and proposed the substitution of copper pipes with a copper annulus, as shown in Figure 10. This design was first proposed by Boyd et al. [55] to achieve better performance without requiring advanced materials and coating. The inside of the annular tube was coated with black paint. The entrance aperture with the insulation on the wall side was cut in V- shape to limit the thermal radiation loss from the collector aperture. It was assumed to have a diffuse surface to reduce the effect of radiative field view of the aperture to the surrounding [55]. In addition, the natural convection loss could be restricted by adjusting the entrance angle of the aperture because the narrow passage causes flow restrictions [55]. The receiver efficiency has been calculated for the hot end of the receiver, where it is lowest, and also averaged over the length of the collector, as a measure of overall performance [55]. The hot end efficiency was 53%, and the average was estimated to be 76% at 370 °C. Furthermore, the conduction through the insulator was dominant at a lower temperature (at 130 °C) and radiation loss was dominant at a higher temperature (at 370 °C). Figure 9: Vertical cross-section of the cavity receiver[54]. Figure 10: Vertical cross-section of the suggested optimized cavity receiver[54]. G. The arc-shaped cavity receiver Xueling Li et al. [56] studied an arc-shaped linear cavity receiver, where the absorber has a crescent shaped channel, as seen in Figure 11, with the aims to raise the HTF temperature and reducing the cost of production and maintenance as well as having a similar shape and size to the conventional receiver [56]. The crescent shaped channel was made from a copper and the outer surface of the channel covered with opaque insulation. The cavity aperture window was fabricated from borosilicate glass, see Figure 11. Figure 11: Arc-shaped linear cavity receiver with a lunate channel [56]. The thermal performance of this design was studied theoretically using a numerical model [56]. The effects of the HTF temperature, surface emissivity, inclination angle, and aperture width were analyzed and displayed some of the following characteristics: the total heat loss of the receiver decreased from 394.5 W/m to 335.8 W/m, when the inclination angles increased from 0° to 90°. The heat loss of the receiver increased with the aperture width of the cavity at the same collecting temperature. A reasonable aperture width of this cavity design is about 50 - 70 mm, where the dimensions of the cavity are shown in Figure 11. Generally, at a larger aperture width, the heat loss increased, and the optical loss (which could be from installation and tracking errors, mirror roughness and manufacturing error) decreased. At a high temperature ( > 400 K), a comparison between the proposed design and an evacuated receiver (Solel's UVAC) showed that the heat loss of the proposed design was less and slower than that of the evacuated receiver [56]. 3. Non-cylindrical receiver unit geometries A. Elliptical cavity receiver Fei Cao et al. [57] studied the elliptical cavity receiver, as shown in Figure 12, with the aim of having a receiver with high thermal performance, low cost, and not frangible. The outer cover was elliptic with an open inlet towards the parabolic trough mirror. The incoming concentrated solar radiation was incident on the receiver and entered into the absorber pipe through the opening. Figure 12: Elliptical cavity receiver [57]. Figure 13: Heat and light flux distributions around the receiver cavity [57]. The position of the absorber pipe was on the opposite focus from the opening. The design was studied using a ray-tracing model and a heat transfer model. The study focused on analyzing the heat transfer and the heat flux distribution in a 1m cavity receiver tube as well as the thermal stress distributions of the absorber pipe. The geometry of the design and its dimension were mentioned in [57]. The light and heat flux distribution around the absorber pipe outer surface is summarized in Figure 13. There was no light and heat flux from 0° to 10° and from 125° to 215°, see Figure 13. Most of the other light entered the cavity and reflected from the elliptic inner surface to the absorber surface. This lead to a non- uniform heat flux absorption along the circumference of the absorber tube. The thermal stress was found to cause a maximum deformation of the receiver tube along the fluid direction of 3.1 mm at 0.82 m. The thermal stress was generated by the temperature difference between the tube inner/outer wall and the heat transfer fluid pressure and phase. This, in turn was caused by the fluid and steel characteristic, solar heating flux and the characteristics of the heat transfer fluid and the absorber material. Besides that, equivalent stress along the absorber pipe was found at higher fluid mass flow rates. B. Elliptical cavity receiver with optical funnel Fei Cao et al. [58] proposed a modified design that evolved from their initial design in Figure 12 to improve the performance of the elliptical cavity. The authors modified the design by adding a flat plate reflector at the cavity aperture, see Figure 14, to diminish the effect of changing the PTC focal length on the cavity opening length and the effect of tracking error angle on the cavity performance. Figure 14: Elliptical cavity with flat plate reflector [58] a) Structure schematic b) light distribution around the receiver. Figure 15 shows the effects of different tracking error angles and PTC focal distances on the cavity darkness of the modified cavity receiver. Cavity darkness is the percentage of the sunlight on the absorber surface to the total incident sunlight. It was found that introducing the flat plate reflector can significantly increases the cavity darkness, where the flat plate breaks the monotonic relationship of the cavity darkness under different focal distances. Further, at different tracking error angles the incident sun radiation is reflected by the flat plate, which causes different multi-reflections inside the cavity, which leads to the curves in Figure 15 [58]. Figure 15: Darkness of the modified elliptical cavity receiver under different tracking angle and PTC focal distances [58]. C. Triangular and V cavity receivers F Chen et al. [59] studied the triangle cavity receiver experimentally and theoretically with the aims to provide high efficiency and make up the shortages of the conventional receivers such as high cost, leakage during long-term running, and challenging technology. The cavity was made from aluminum, and the absorber surface was a triangle or V-shape structure with fins in the dorsal side to enhance the thermal performance between the absorber surface and the HTF, see Figure 16. Figure 16: Cross-section of the triangle cavity receiver [59]. A radiation shield was placed at the aperture to reduce the view factor of the absorber pipe to the surrounding ambient. Experimental and theoretical work was conducted to study the heat loss due to the absorber inclination and the ambient wind speed. The results showed that the ambient wind speed had a relatively significant effect on heat loss, while the impact of the absorber inclination was relatively small. The heat loss was found to be 51.2 W/m, 53.2 W/m, 59.7 W/m and 61.8 W/m, when the inclination angles were 0°, 30°, 60° and 90°, respectively, at temperature differences between the working fluid and ambient of 150 ± 3 °C. At the inclination of 60° and temperature differences of 27 °C, 84.5 °C, 126.7 °C and 176.3 °C, the heat losses were 11.5 W/m, 30.4 W/m, 48.25 W/m and 71.55 W/m, respectively. Furthermore, the heat losses of the cavity receiver were 76.6 W/m, 85.5 W/m, 96 W/m and 102.6 W/m, when the wind speeds were 1 m/s, 2 m/s, 3 m/s and 4 m/s, respectively, at the temperature difference of 150 ± 3 °C. At the wind speed of 3 m/s the temperature differences were 27.6 °C, 85.6 °C, 124.4 °C and 172.6 °C, and the heat losses were 17.3 W/m, 63 W/m, 83.85 W/m and 104 W/m, respectively. In a windless case, the heat loss of this design was equivalent to that of UVAC3 evacuated receiver and the new-generation (UVAC2008)[60]. Two studies investigated the same design with an additional glass cover on the aperture [61][62]. Their objective was to reduce the heat loss from the aperture. The first study focused on heat transfer performance, where the investigation was theoretically and experimentally. The optical performance was studied using Monte Carlo ray-tracing method [61]. The design showed high optical efficiency of about 99% because the concentrated sunlight repeatedly reflected by the triangular shape with almost no escape [61]. Furthermore, the heat flux distribution of the heating surface of the design was heterogeneous, which could cause thermal stress at higher temperature [61]. The second study focused on thermal performance, also from a theoretical and experimental perspective [62]. The design had a good thermal performance in the medium temperature range, and it was comparable to that of the evacuated tube in that temperature range. The study involved the effect of the glass cover and the fins. These additions showed an improvement in thermal performance. Moreover, the heat transfer fluid temperature in this cavity design could exceed 570 K [62]. Fei Chen et al. [2] studied the optical properties of the triangular cavity absorber using a theoretical method. They found that the cavity absorber's aperture width, depth to width of the triangular shape, and the offset distance from the focus of the triangular cavity were important parameters to improve the optical performance, where the cavity optical efficiency was 89.23%. It was recommended to select the depth to width ratio of 0.8 to 1, the aperture width of 70 mm, and the offset distance of 15 mm [2]. Figure 17 shows another design that is similar to the V- shape or triangular cavity. It consists of a center tube as absorber and two inclined fins which acted as the inner cavity surface with the glass cover over the aperture. A rectangular shell separated the cavity from the surrounding. The space between the shell and the absorber was filled with aluminum silicate fiber and asbestos rubber sheet between the glass cover and the end of the inclined fins. At the sides of the shell, there was a fixed axle to make the whole cavity movable. It rotated the system when solar irradiance was high enough to collect the concentrated solar energy [63][64]. The movable cavity mechanism was a novel design to prevent overheating while reducing heat loss. The study was based on experimental and theoretical investigations. Figure 17: Cross-section of the V-shape cavity receiver [63]. Regarding the effect of the on-off state of the movable mechanism of the cavity on the heat loss, it was found that the heat loss of turning off the movable cover was less than that of turning on [64], where the heat loss reduction rate varied from 6.36% to 13.55%. The author found that the movable mechanism should be optimized for thermal insulation performance and operation control strategy [64]. The collector efficiency was tested at different inlet temperatures ranging from 80.6 °C to 160.5 °C and the mass flow rate from 170 to 181 g/s. The efficiency was in the range of 34.2% to 48.5% [63]. Further, the thermal conductivity of insulation materials was significantly improved the thermal performance of the design, i.e., the collector efficiency increased by 1.47 times, if the thermal conductivity of the insulation materials changed from 0.1 to 0.02 W. K/m [63]. Besides, the collector efficiency could increase by decreasing the emittance of the absorber and glass cover, i.e., the collector efficiency increased from 34.45% to 38.49% if the emittance of the absorber 𝜀𝜀𝑎𝑎𝑎𝑎 and the glass cover 𝜀𝜀𝐺𝐺 drop down from 0.9 and 0.95 to 0.1 and 0.15, respectively. The efficiency of this design was comparable to the efficiency of the metal glass evacuated tube (64.25%). If we use the following optimized parameters: 𝜀𝜀𝑎𝑎𝑎𝑎=0.15, 𝜀𝜀𝐺𝐺=0.1, • • • • • absorption coefficient of the absorber in visible region = 0.935, thermal conductivity of insulation material 1 (Figure 16) = 0.02 W/m.K, thermal conductivity for insulation material 2 = 0.1W/m.K [63]. Zhai H et al. [65] studied a triangle cavity at a lower temperature (< 200 °C). The optical efficiency simulated by using a light tracking method and the thermal performance was tested experimentally under temperature levels of 90 °C and 150 °C [65]. This study found that the triangle shape cavity receiver optical efficiency was 99%, and thermal losses were 20 W and 41 W (measured at 0.5 m of the receiver length) at the inlet temperature of 90 °C and 150 °C, respectively. Moreover, the solar conversion efficiency could be beyond 67% for the triangle cavity [65]. D. Trapezoidal cavity receiver Singh et al. [66][67] studied the effects of various design parameters of a trapezoidal cavity absorber on the thermal performance. The trapezoidal cavity absorber with a round pipes is shown in Figure 18. The absorber pipes were made of six mild steel round tubes brazed together in a single layer. The absorber pipe was at the upper portion of the cavity. Glass wool insulation was provided at the top and the sides of the pipes. At sidewalls of the cavity, ceramic tiles plates were provided. At the bottom part of the cavity, a glass plane was provided as a window for transmitting the solar radiation. Also, this study proposed another trapezoidal cavity receiver with a rectangular pipe absorber instead of round pipes to compare and evaluate their performance, see Figure 19. Figure 18: Cross-sectional of the trapezoidal cavity with round pipe absorber [66]. Figure 19: Cross-sectional of the trapezoidal cavity with rectangular pipe absorber [66]. The dimensions of the designs were mentioned in detail in [66]. The thermal performance was measured for eight sets of identical designs with round pipe and rectangular pipe absorbers. The trapezoidal cavity with round pipe absorbers was tested in four different setups. The first two, the round pipe absorbers, were painted with ordinary matt black paint and black nickel coating (selective coating) with emissivity 0.91 and 0.17 at 100 °C, respectively. The other two, the cavity was fabricated with double (10 mm spacing) and single glass cover. Similar to the above mentioned scenarios, the trapezoidal cavity with the rectangular pipe absorbers was tested with four different setups. The experimental results showed that the difference between the heat loss coefficient of rectangular and round pipe absorbers in the trapezoidal cavity were not significantly different -- they differed by a factor of 3.3 to 8.2 W/m2, respectively [66]. The selective coating on the absorbers had a remarkable reduction of overall heat loss coefficient by 20% to 30 % compared to ordinary black paint. In addition, using double glass cover reduced the overall heat loss by 10% to 15 % compared to single glass cover [66]. Manikumar et al. [68] analyzed trapezoidal cavity numerically and experimentally. The cavity had a multi- tube absorber with a plate and without plate underneath, see Figure 20 and Figure 21, for various values of gaps between the tubes and depths of the cavity. The values of the overall heat loss coefficient and convective heat transfer coefficient were observed to increase with gaps between the tubes and the tube temperature. The thermal efficiency of the cavity with a plate was higher than without. Figure 20: The arrangement of multi-tube absorber with plate[68]. Figure 21: The arrangement of multi-tube absorber with plate [68]. Oliveira et al. [69] analyzed and optimized a trapezoidal cavity receiver via ray-trace and computational fluid dynamic (CFD) simulations. It was found that the cavity with six absorber tubes of 1/2" / 5/8" inner/outer diameters collects had higher optical efficiency. Further, the maximum inclination of 50° of the lateral cavity wall with respect to the bottom base was found to be optically acceptable. CFD simulation was used to optimize the cavity depth and rock wool insulation thickness. The lowest heat transfer coefficient was observed at the cavity depth of 45 mm. The insulation thickness of 35 mm of rock wool showed a good compromise between shading and insulations. Reynold et al. [70] studied the heat transfer rate and the heat loss of the trapezoidal cavity experimentally with the aim of optimizing the cavity design to achieve maximum thermal efficiency and to develop a numerical model. Furthermore, the cavity was modeled by using a computational fluid dynamic (CFD) software package. The heat transfer rate results and flow patterns showed reasonable agreement between the computational and experimental works, but the heat loss that was measured by CFD was underestimated by about 40% as compared to the experimental results. This discrepancy could not be explained. The heat transfer rate was compared between a uniform and non-uniform heating of the bottom wall with natural convection flows in a trapezoidal cavity using a finite element analysis with bi-quadratic elements (a method for evaluating the nonlinear coupled partial differential equations for flow and temperature fields) [71]. It was found that for all Raleigh numbers, the non-uniform heating of the bottom wall has a significant heat transfer rate as compared to uniform heating [71]. Moreover, the trapezoidal cavity receiver has been studied optically by Liang et al. [72]. The optical efficiency (it is the ratio between the solar radiation reaching the receiver absorber and the solar radiation coming from the concentrator mirror) of the total configuration was around 85% while the optical efficiency of the absorber tube was about 45%. Natarajan et al. [73] studied the effect of the Grashof number, absorber angle, aspect ratio (ratio of width and depth of cavity), surface emissivity and temperature ratio (ratio of the bottom and top surface temperature of the cavity) of the trapezoidal cavity. In their model, radiation and convection heat transfer was included. Consequently, the Grashof number has been included in the Nusselt number correlations. It was found that the effect of Grashof number on the combined heat loss (natural convection and surface radiation) was negligible. Further, the combined Nusselt number on the absorber angle was also negligible. The Nusselt number was decreased by increasing the aspect ratio and the temperature ratio, while an increase was observed in surface emissivity. Beyond the temperature ratio of 0.6 and aspect ratio 2.5, the combined heat loss variation in the cavity was not notable. 4. Conclusion The alternative to the evacuated receiver unit of the Parabolic Trough Collector (PTC) that can overcome its shortages and compete in terms of efficiency, performance, and cost-effective is the objective of the research of the PTC. Cavity receivers have the potential to overcome the shortages of the evacuated receiver. This work is a comprehensive study of different cavities receiver designs. It was found that the effect of the cavity aperture size and the mirrored cavity reflectivity on the cavity performance are essential, where the smaller the cavity aperture and the higher cavity mirror reflectivity yield higher efficiency. Further, the reflectivity of the cavity mirror enhances the optical performance and solar flux distribution inside the receiver as well as reducing the temperature gradient and thermal stresses due to the effect of the multiple reflections in the cavity annulus. However, the dimension of the cavity aperture alongside with the focal line of the collector should be carefully selected to ensure all the reflected radiation enters the cavity receiver through the aperture. It is necessary to keep the concentration ratio maximum (it could require to change the parabolic trough rim angle) and to minimize inaccuracies in both tracking and directional errors due to the sun shape. To diminish the effect of changing the focal length on the cavity opening and the error in the tracking angle, the cavity aperture could have a mirrored V shape or adding flat plate reflector at the cavity aperture (optical funnel). The main deficiencies of most of the existing cavity designs are keeping the cavity annulus under atmospheric pressure and leaving the cavity aperture open. This often leads to significant losses. Generally, the thermal losses of the cavity are affected by the following: surface emissivity, working fluid temperature, mass flow rate, and wind velocity. Also, cavity geometry could affects both optical and thermal efficiency. In which case the cylindrical geometry seems to be more effective, especially in terms of thermal and optical flux distribution. In non-cylindrical geometry cavity receivers, for the case of the elliptical cavity receiver, which the absorber pipe was positioned on the opposite focus from the cavity opening, was prone to deformation due to non-uniform heat flux absorption across the pipe circumference. The triangular and V-shape cavity receivers had the best optical efficiency among the non-cylindrical cavity receivers followed by the elliptical cavity with a flat plate reflector. However, their heat flux distribution of the heating surface were heterogeneous, which could cause thermal stress at a higher temperature. In addition, their thermal performance was comparable to the evacuated receiver at the medium temperature range and the heat loss was equivalent to the evacuated receiver in a windless condition. Moreover, the optical performance was affected by the aperture width, the depth to width and the offset distance from the focus of the triangular shape. For the trapezoidal cavity receiver, the optical efficiency of the absorber tube was about 45% and for the total configuration was around 85%. The overall heat loss of the trapezoidal cavity aperture with double-glazed cover reduced by up to 15% compared to the single glass cover. For cylindrical cavity receivers, where there is no vacuum in the annulus, the convection and conduction heat loss in the annulus can be minimized by selecting the optimum annulus dimension between the absorber pipe and the cavity envelope. Although this largely depends on the diameters of the receiver components. The conventional designs that was developed by coating the inner side of the glass cover with a mirrored or hot mirror coatings had a deficiency of being breakable. The receiver unit with hot mirror over the glass cover with a vacuum in the annulus was found to be more effective than the conventional receiver at a higher temperature. The hot mirror receiver could be better alternative to the conventional receiver with improving hot mirror materials for hot mirror applications especially, the transmissivity in the visible region. In the current receiver unit with hot mirror application, the hybrid system is usually favored. Those receivers that were utilized air as a working fluid had overcome the lower volumetric heat transfer capacity of the air by increasing the heat transfer area. This was done by eliminating the absorber pipe and using corrugate structure inside the cavity wall to allow the air to flow in a sufficient large cross section, but a significant energy power was required for pumping the working fluid. In cylindrical geometry designs of the absorber pipe for the cavity receiver such as copper pipes, copper annulus and crescent shaped channel had the aims of raising the HTF and reducing the cost of production and maintenance. It was found that, the heat loss of the cavity receiver with the crescent shaped channel with a borosilicate glass on the cavity aperture was less and slower than that of the evacuated receiver at a high temperature (> 400 °C). The design of the mirrored cavity receiver with a hot mirror application on the aperture showed that it was able to achieve the highest thermal and optical efficiencies, especially if the hot mirror coating has a higher transmission coefficient for the incident solar radiation. Further, this cavity exceeded the HTF temperature ceiling compared to existing alternatives and hence increase the efficiency of the system. Acknowledgments The authors would like to thank the following entities for their kind support: MERG (Materials for Energy Research Group) and MITP (Mandelstam Institute for Theoretical Research), at the University of the Witwatersrand. 5. References [1] [2] Chaanaoui M, Vaudreuil S, Bounahmidi T. Benchmark of Concentrating Solar Power Plants: Historical, Current and Future Technical and Economic Development. Procedia Comput Sci 2016; 83: 782 -- 789. Chen F, Li M, Hassanien Emam Hassanien R, et al. Study on the optical properties of triangular cavity absorber for parabolic trough solar concentrator. Int J Photoenergy; 2015. [3] Moya EZ. 7 - Parabolic-trough concentrating solar power (CSP) systems. In: Lovegrove K, Stein W (eds) Concentrating Solar Power Technology. Woodhead Publishing, pp. 197 -- 239. [4] Forristall R. Heat transfer analysis and modeling of a parabolic trough solar receiver implemented in engineering CO.(US), equation http://large.stanford.edu/publications/coal/references/troughnet/solarfield/docs/34169.pdf (2003, accessed 13 August 2017). Renewable National Golden, Energy solver. Lab., [5] Ramchandra G. P, Sudhir V. P, Jyeshtharaj B. J, et al. Alternative designs of evacuated receiver for parabolic trough collector. Energy; 155: 66 -- 76. [6] Mohamad K, Ferrer P. Parabolic trough efficiency gain through use of a cavity absorber with a hot mirror. Appl Energy 2019; 238: 1250 -- 1257. [7] Fuqiang W, Ziming C. Progress in concentrated solar power technology with parabolic trough collector system: A comprehensive review. Renew Sustain Energy Rev 2017; 79: 1314 -- 1328. [8] Wang P, Liu DY, Xu C. Numerical study of heat transfer enhancement in the receiver tube of direct steam generation with parabolic trough by inserting metal foams. Appl Energy 2013; 102: 449 -- 460. [9] Sargent, Lundy LLC Consulting Group. Assessment of Parabolic Trough and Power Tower Solar Technology Publishing, Cost https://scholar.google.co.za/scholar?hl=en&as_sdt=0%2C5&q=Assessment+of+Parabolic+Trough+and+Power+ Tower+Solar+Technology+Cost+and+Performance+Forecasts&btnG= (2003). Performance Forecasts. Chicago, Illinois: DIANE and [10] International Renewable Energy Agency. Renewable energy technologies: cost analysis series: concentrating solar power; https://www.irena.org/DocumentDownloads/Publications/RE_Technologies_Cost_Analysis- CSP.pdf (accessed 13 August 2017). 2012., [11] Antonaia A, Castaldo A, Addonizio ML, et al. Stability of W-Al 2 O 3 cermet based solar coating for receiver tube operating at high temperature. Sol Energy Mater Sol Cells 2010; 94: 1604 -- 1611. [12] Dudley VE, Kolb GJ, Mahoney AR, et al. Test results: SEGS LS-2 solar collector. Nasa Stirecon Tech Rep N; 96. [13] Forristall R. EES heat transfer model for solar receiver performance. Proc ISEC Sol 2004; 11 -- 14. [14] Kennedy CE. Review of mid-to high-temperature solar selective absorber materials. National Renewable Energy Lab., CO.(US), http://large.stanford.edu/publications/power/references/troughnet/solarfield/docs/31267.pdf (2002, accessed 8 August 2017). Golden, [15] Kennedy CE, Price H. Progress in Development of High-Temperature Solar-Selective Coating. ASME Int Sol Energy Conf Sol Energy 2005; 749 -- 755. [16] ASE. Archimede Solar Energy, http://www.archimedesolarenergy.it/ (accessed 8 December 2017). [17] SCHOTT. SCHOTT c0b5c78b01ae/1.0/schott_ptr70_4th_generation_brocure.pdf (accessed 27 February 2019). PTR 70, https://www.schott.com/d/csp/370a8801-3271-4b2a-a3e6- [18] Reddy KS, Ravi Kumar K, Ajay CS. Experimental investigation of porous disc enhanced receiver for solar parabolic trough collector. Renew Energy 2015; 77: 308 -- 319. [19] Ghasemi S, Ranjbar A. Numerical thermal study on effect of porous rings on performance of solar parabolic trough collector. Appl Therm Eng 2017; 118: 807 -- 816. [20] Mwesigye A, Bello-Ochende T, Meyer JP. Heat transfer and thermodynamic performance of a parabolic trough receiver with centrally placed perforated plate inserts. Appl Energy 2014; 136: 989 -- 1003. [21] Sahin H, Baysal E. Investigation of heat transfer enhancement in a new type heat exchanger using solar parabolic trough systems. Int J Hydrog Energy 2015; 40: 15254 -- 15266. [22] Eiamsa-ard S, Thianpong C, Eiamsa-ard P. Turbulent heat transfer enhancement by counter/co-swirling flow in a tube fitted with twin twisted tapes. Exp Therm Fluid Sci 2010; 34: 53 -- 62. [23] Diwan K, Soni MS. Heat transfer enhancement in absorber tube of parabolic trough concentrators using wire-coils inserts. Univers J Mech Eng 2015; 3: 107 -- 112. [24] Martín RH, Pérez-García J, García A, et al. Simulation of an enhanced flat-plate solar liquid collector with wire- coil insert devices. Sol Energy 2011; 85: 455 -- 469. [25] Shashank SC, Taji SG. Experimental studies on effect of coil wire insert on heat transfer enhancement and friction factor of double pipe heat exchanger. Int J Comput Eng Res; 3. [26] Nanan K, Thianpong C, Pimsarn M, et al. Flow and thermal mechanisms in a heat exchanger tube inserted with twisted cross-baffle turbulators. Appl Therm Eng 2017; 114: 130 -- 147. [27] Hong Y, Du J, Wang S. Experimental heat transfer and flow characteristics in a spiral grooved tube with overlapped large/small twin twisted tapes. Int J Heat Mass Transf 2017; 106: 1178 -- 1190. [28] Huang Z, LI Z. Numerical investigations on fully-developed mixed turbulent convection in dimpled parabolic trough receiver tubes. Appl Therm Eng 2017; 114: 1287 -- 1299. [29] Cheng Z, He YL. Numerical study of heat transfer enhancement by unilateral longitudinal vortex generators inside parabolic trough solar receivers. Int J Heat Mass Transf 2012; 55: 5631 -- 5641. [30] Fuqiang W, Qingzhi L, Huaizhi H. Parabolic trough receiver with corrugated tube for improving heat transfer and thermal deformation characteristics. Appl Energy 2016; 164: 411 -- 424. [31] Fuqiang W, Zhexiang T, Xiangtao G. Heat transfer performance enhancement and thermal strain restrain of tube receiver for parabolic trough solar collector by using asymmetric outward convex corrugated tube. Energy 2016; 114: 275 -- 292. [32] Bellos E, Tzivanidis C, Antonopoulos KA. Thermal enhancement of solar parabolic trough collectors by using nanofluids and converging-diverging absorber tube. Renew Energy 2016; 94: 213 -- 222. [33] Hassani S, Saidur R, Mekhilef S, et al. A new correlation for predicting the thermal conductivity of nanofluids; using dimensional analysis. Int J Heat Mass Transf 2015; 90: 121 -- 130. [34] Yunus CA, Afshin JG. Heat and mass transfer: fundamentals and applications. 5 Ed. 2015. [35] Patil RG, Panse SV, Joshi JB. Optimization of non-evacuated receiver of solar collector having non-uniform temperature distribution for minimum heat loss. Energy Convers Manag 2014; 85: 70 -- 84. [36] Stettenheim J, McBride TO, Brambles OJ. Cavity Receivers for Parabolic Solar Troughs. 2013. [37] Kaluba VS, Ferrer P. A model for hot mirror coating on solar parabolic trough receivers. J Renew Sustain Energy 2016; 8: 053703. [38] Kaluba VS, Mohamad K, Ferrer P. Experimental and simulated Performance of Heat Mirror Coatings in a Parabolic Trough Receiver. Submitt Appl Energy J ArXiv190800866 Physicsapp-Ph. [39] Granqvist CG. Spectrally Selective Coatings for Energy Efficiency and Solar Applications. Phys Scr 1985; 32: 401. [40] Canan K. Performance analysis of a novel concentrating photovoltaic combined system. Energy Convers Manag 2013; 67: 186 -- 196. [41] Miller DC, Khonkar HI, Herrero R, et al. An end of service life assessment of PMMA lenses from veteran concentrator photovoltaic systems. Sol Energy Mater Sol Cells 2017; 167: 7 -- 21. [42] Lampert CM. Coatings for enhanced photothermal energy collection II. Non-selective and energy control films. Sol Energy Mater 1979; 2: 1 -- 17. [43] Grena R. Efficiency Gain of a Solar Trough Collector Due to an IR-Reflective Film on the Non-Irradiated Part of the Receiver. Int J Green Energy 2011; 8: 715 -- 733. [44] Granqvist CG. Radiative heating and cooling with spectrally selective surfaces. Appl Opt 1981; 20: 2606 -- 2615. [45] Cyulinyana MC, Ferrer P. Heat efficiency of a solar trough receiver with a hot mirror compared to a selective coating. South Afr J Sci 2011; 107: 01 -- 07. [46] Mohamad K, Ferrer P. [Patent] Thermal radiation loss reduction in a parabolic trough receiver by the application of a cavity mirror and a hot mirror coating. PCT/IB2019/053531, South Africa, Johannesburg. [47] Mohamad K, Ferrer P. Experimental and numerical measurement of the thermal performance for parabolic trough solar concentrators. Proc 63th Annu Conf South Afr Inst Phys SAIP2018 ArXiv190800515 Physicsapp-Ph. [48] Mohamad K, Ferrer P. Computational comparison of a novel cavity absorber for parabolic trough solar concentrators. Proc 62th Annu Conf South Afr Inst Phys SAIP2017 ArXiv190713066 Physicsapp-Ph. [49] Bader R, Pedretti A, Barbato M, et al. An air-based corrugated cavity-receiver for solar parabolic trough concentrators. Appl Energy 2015; 138: 337 -- 345. [50] Bader R, Pedretti A, Steinfeld A. Experimental and numerical heat transfer analysis of an air-based cavity-receiver for solar trough concentrators. J Sol Energy Eng 2012; 134: 021002. [51] Hänchen M, Brückner S, Steinfeld A. High-temperature thermal storage using a packed bed of rocks -- Heat transfer analysis and experimental validation. Appl Therm Eng 2011; 31: 1798 -- 1806. [52] G. Z, A. P, A. H, et al. Design of packed bed thermal energy storage systems for high-temperature industrial process heat. Appl Energy; 137: 812 -- 822. [53] Burkholder F, Kutscher C. Heat Loss Testing of Schott's 2008 PTR70 Parabolic Trough Receiver. NREL/TP-- 550-45633, 1369635. Epub ahead of print 1 May 2009. DOI: 10.2172/1369635. [54] Barra OA, Franceschi L. The parabolic trough plants using black body receivers: experimental and theoretical analyses. Sol Energy 1982; 28: 163 -- 171. [55] Boyd DA, Gajewski R, Swift R. A cylindrical blackbody solar energy receiver. Sol Energy 1976; 18: 395 -- 401. [56] Li X, Chang H, Duan C, et al. Thermal performance analysis of a novel linear cavity receiver for parabolic trough solar collectors. Appl Energy 2019; 237: 431 -- 439. [57] Cao F, Li Y, Wang L, et al. Thermal performance and stress analyses of the cavity receiver tube in the parabolic trough solar collector. In: IOP Conference Series: Earth and Environmental Science. IOP Publishing, 2016, p. 012067. [58] Fei C, Lei W, Tianyu Z. Design and Optimization of Elliptical Cavity Tube Receivers in the Parabolic Trough Solar Collector. Int J Photoenergy; 2017: 7. [59] Chen F, Li M, Xu C, et al. Study on heat loss performance of triangle cavity absorber for parabolic trough concentrators. In: Materials for Renewable Energy and Environment (ICMREE), 2013 International Conference on. IEEE, 2014, pp. 683 -- 689. [60] Xiong Y, Wu Y, Ma C, et al. Numerical investigation of thermal performance of heat loss of parabolic trough receiver. Sci China Technol Sci 2010; 53: 444 -- 452. [61] Xiao X, Zhang P, Shao DD, et al. Experimental and numerical heat transfer analysis of a V-cavity absorber for linear parabolic trough solar collector. Energy Convers Manag 2014; 86: 49 -- 59. [62] Chen F, Li M, Zhang P, et al. Thermal performance of a novel linear cavity absorber for parabolic trough solar concentrator. Energy Convers Manag 2015; 90: 292 -- 299. [63] Hongbo L, Chunguang Z, Man F. Study on the thermal performance of a novel cavity receiver for parabolic trough solar collectors. Appl Energy 2018; 222: 790 -- 798. [64] Liang H, Fan M, You S, et al. An analysis of the heat loss and overheating protection of a cavity receiver with a novel movable cover for parabolic trough solar collectors. Energy 2018; 158: 719 -- 729. [65] Zhai H, Dai Y, Wu J, et al. Study on trough receiver for linear concentrating solar collector. In: Proceedings of ISES World Congress 2007 (Vol. I -- Vol. V). Springer, 2008, pp. 711 -- 715. [66] Singh PL, Sarviya RM, Bhagoria JL. Heat loss study of trapezoidal cavity absorbers for linear solar concentrating collector. Energy Convers Manag 2010; 51: 329 -- 337. [67] Singh PL, Sarviya RM, Bhagoria JL. Thermal performance of linear Fresnel reflecting solar concentrator with trapezoidal cavity absorbers. Appl Energy 2010; 87: 541 -- 550. [68] Manikumar R, Palanichamy R, Arasu AV. Heat transfer analysis of an elevated linear absorber with trapezoidal cavity in the linear Fresnel reflector solar concentrator system. J Therm Sci 2015; 24: 90 -- 98. [69] Facão J, Oliveira AC. Numerical simulation of a trapezoidal cavity receiver for a linear Fresnel solar collector concentrator. Renew Energy 2011; 36: 90 -- 96. [70] Reynolds DJ, Jance MJ, Behnia M, et al. An experimental and computational study of the heat loss characteristics of a trapezoidal cavity absorber. Sol Energy 2004; 76: 229 -- 234. [71] Natarajan E, Basak T, Roy S. Natural convection flows in a trapezoidal enclosure with uniform and non-uniform heating of bottom wall. Int J Heat Mass Transf 2008; 51: 747 -- 756. [72] Liang H, Fan M, You S, et al. A Monte Carlo method and finite volume method coupled optical simulation method for parabolic trough solar collectors. Appl Energy 2017; 201: 60 -- 68. [73] Natarajan SK, Reddy KS, Mallick TK. Heat loss characteristics of trapezoidal cavity receiver for solar linear concentrating system. Appl Energy 2012; 93: 523 -- 531.
1711.00684	1	1711	2017-11-02T11:12:46	Landau levels with magnetic tunnelling in Weyl semimetal and magnetoconductance of ballistic $p$-$n$ junction	[ "physics.app-ph", "cond-mat.mes-hall" ]	We study Landau levels (LLs) of Weyl semimetal (WSM) with two adjacent Weyl nodes. We consider different orientations $\eta=\angle(\mathbf{B},\mathbf{k}_0)$ of magnetic field $\mathbf{B}$ with respect to $\mathbf{k}_0$, the vector of Weyl nodes splitting. Magnetic field facilitates the tunneling between the nodes giving rise to a gap in the transverse energy of the zeroth LL. We show how the spectrum is rearranged at different $\eta$ and how this manifests itself in the change of behavior of differential magnetoconductance $dG(B)/dB$ of a ballistic $p$-$n$ junction. Unlike the single-cone model where Klein tunneling reveals itself in positive $dG(B)/dB$, in the two-cone case $G(B)$ is non-monotonic with maximum at $B_c \propto \Phi_0k_0^2/\ln(k_0l_E)$ for large $k_0l_E$, where $l_E=\sqrt{\hbar v/\|e\|E}$ with $E$ for built in electric field and $\Phi_0$ for magnetic flux quantum.	physics.app-ph	physics	Landau levels with magnetic tunnelling in Weyl semimetal and magnetoconductance of ballistic p−n junction D. R. Saykin,1, 2 K. S. Tikhonov,2, 3 and Ya. I. Rodionov4, 3 1Moscow Institute of Physics and Technology, Russia 2Landau Institute for Theoretical Physics, Russia 3National University of Science and Technology MISIS, Moscow, 119049 Russia 4Institute for Theoretical and Applied Electrodynamics, Moscow, 125412 Russia (Dated: November 3, 2017) We study Landau levels (LLs) of Weyl semimetal (WSM) with two adjacent Weyl nodes. We consider different orientations η = ∠(B, k0) of magnetic field B with respect to k0, the vector of Weyl nodes splitting. Magnetic field facilitates the tunneling between the nodes giving rise to a gap in the transverse energy of the zeroth LL. We show how the spectrum is rearranged at different η and how this manifests itself in the change of behavior of differential magnetoconductance dG(B)/dB of a ballistic p−n junction. Unlike the single-cone model where Klein tunneling reveals itself in positive dG(B)/dB, in the two-cone case G(B) is non-monotonic with maximum at Bc ∝ Φ0k2 0/ ln(k0lE) for large k0lE, where lE = (cid:112)v/eE with E for built in electric field and Φ0 for magnetic flux quantum. I. INTRODUCTION Since the discovery of time-reversal invariant topologi- cal insulators (see Ref. 1 and references therein) topolog- ical properties of the electronic band structure of crys- talline materials have been enjoying a lot of attention. More recently it was demonstrated that topological prop- erties are shared by accidental point band touchings so that they may become nontrivial and robust under bro- ken either time reversal or inversion symmetry. The band structure near these points can be described by a mass- less two-component Dirac or Weyl Hamiltonian. After Ref. 2 indicated the possibility of a WSM state for py- rochlore iridates, the quest for model Hamiltonians and material candidates ensued3–6. Experimentally, the WSM state was first discovered in TaAs7,8 and TaP9, materials with the broken inversion symmetry. First principle calculations10 revealed that in both materials all Weyl nodes form a set of closely posi- tioned pairs of opposite chirality in momentum space, this prediction was confirmed by experimental obser- vation. Recently, the active experimental research11–20 shifted from the initial band-structure study to the sur- face and transport phenomena: a significant amount of attention was devoted to magnetotransport, which was addressed theoretically21–24 and experimentally25–28. One of the most impressive manifestations of gapless band structure is Klein tunneling, which reveals itself in transport through a p−n junction. Such junctions have been investigated theoretically for graphene29–31, carbon nanotubes32,33 and surface of topological insulators34. The recent study Ref. 23 was devoted to magnetoconduc- tance of a p−n junction realized in WSM. The authors showed that in the case of a longitudinally aligned exter- nal magnetic field: B (cid:107) E where E is a junction's built- in electric field, the differential magnetoconductance dG(B)/dB is positive. This situation is different from the ordinary semiconductor p−n junction35, where the real tunneling between valence and conductance bands is involved and magnetoconductance is negative36,37. The authors of Ref. 23 argued that the positivity of magne- toconductance is attributed to the existence of a zero mode in electron LLs with degeneracy ∝ B and field- independent reflectionless Klein tunneling. 0 of motion lB =(cid:112)c/eB becomes of order k−1 The treatment of Ref. 23 was done in the approxima- tion of well-separated (in the momentum space) Weyl nodes. This is a standard approximation which holds well in many situations. Usually, the influence of pairwise38 structure of WSM nodes on transport phenomena is ac- counted for by simple multiplication of a single point con- tribution by the number of cones in the spectrum. This approach however, breaks down in strong magnetic fields. When cyclotron radius of a particle Rc ∼ ck0/eB with the momentum k0 becomes of the order of its coordinate uncertainty k−1 or, in other words, characteristic length 0 , intern- ode coupling must be taken into account. For example, in TaAs8 the momentum distance between the Weyl nodes in a pair is 2k0 = 0.0183 A−1 and it happens at fields of order B (cid:39) Φ0k2 0 (cid:39) 17 T. Indeed, such field-induced tunneling between two nodes in a pair has already been observed experimentally39. Therefore, it is not sufficient to consider the problem in a single Weyl cone approxi- mation in such relatively high fields. On the one hand, taking into account of the full spectrum of WSM which has 12 pairs of Weyl nodes (like in TaAs) is an intractable problem. On the other hand, the distance between pairs of Weyl nodes in momentum space fortunately happens to be much larger than the distance between the nodes in a pair7. Therefore, the correct treatment is to consider the nodes pairwise. The problem of field-induced inter-node tunneling was addressed in a semiclassical approximation40, where the behavior of the carrier density of states for high-lying LLs was explored. Recently, the same problem was studied numerically39,41 for all LLs for the perpendicular orien- 2 FIG. 2: Magnetoconductance G(B) G(0) vs Magnetic field ζ vB cE with parameter ζk0lE = 2. Dashed line represents result of Ref. 23 and black line depicts first term of the sum (23). where lE =(cid:112)v/eE is the electric field length. The paper is organized as follows: section II explores the structure of LLs in the two–conical system of Eq. (1), section III is dedicated to the magnetoconductance of a ballistic p−n junction realized in such system and the conclusions are drawn in section IV. II. LANDAU LEVELS We begin our analysis with search for energy disper- sion law in the presence of the magnetic field starting from Hamiltonian (1). In the presence of magnetic field, Hamiltonian (1) can't be split into two independent parts of opposite chirality (only effectively, at k0 → ∞) so the nodes should be treated simultaneously. We ori- ent the coordinates so that x axis points in the direc- tion of Weyl node separation k0. The magnetic field B = B(cos η, 0, sin η) is inclined at the angle η with re- spect to x axis (see. FIG. 1), so that the field is described by potential A = B(−y sin η, 0, y cos η). At first, we solve an eigenvalue problem in two limiting cases η = 0, π 2 an- alytically and then provide numerical solutions for arbi- trary angles. Field parallel to nodes splitting. We start from the case η = ∠(B, k0) = 0. After the shift of the variable y (cid:55)→ y − kzl2 (1 + iσy)ψ the Hamiltonian transforms to x − k2 2mv (k2 0) yl−2 B + ∂y B, and unitary rotation ψ (cid:55)→ 1√ B − ∂y yl−2 x − k2 2mv (k2 0) (cid:18)− H = v (cid:19) (4) 2 . Eigenfunctions are expressed through Hermite functions ψosc n (y) = (2nn! πlB)−1/2e−y2/2Hn(y) as √ (cid:32) nψoscn (yl−1 c1 B ) nψoscn−1(yl−1 c2 B ) (cid:33) (cid:18)ψosc 0 (yl−1 B ) 0 (cid:19) , ψ0 = , (5) FIG. 1: The orientation of Weyl spectrum and magnetic field. tation of field with respect to nodes splitting k0. It was discovered that the magnetic-induced tunneling opens a gap in a LL zero mode. The numerical analysis shows that the gap is non-perturbative in external magnetic field B. We present here analytical theory of the spectrum of the LLs and its dependence on angle η between mag- netic field B and k0. In particular, we show that for η = π/2 the problem of LLs is reduced to the supersym- metric quantum mechanics of a particle in quadratic su- perpotential. Let us suppose the spectrum consists of two cones separated by the distance 2k0. Generic low–energy Hamiltonian for such system was derived in Ref. 42: (cid:16)kyσy + kzσz (cid:17) , (1) H = ∆ + 2 2m x − k2 0 σx + v (cid:16)k2 (cid:114) B (k0)2 where ∆ is Weyl node energy measured from the chemical potential and v stands for Fermi velocity. In what follows, we drop the energy offset ∆ from most of the equations (it matters only in estimation of heterojunction's built– in potential, see in Ref. 23). In the framework of a model described by Hamiltonian Eq. (1), our results are as follows. The lowest LL energy is indeed exponentially close to zero and non-perturbative in magnetic field. The exact formula for transversal energy reads Φ0k2 0 , π vy = m (2) exp , B0 = ζ ε0 = πB0 where ζ ≡ vx k0 mv is anisotropy parameter. Numer- ically computed dependence of LLs on angle η between magnetic field B and k0 is shown on the FIG. 3. We have also studied analytically the magnetoconduc- tance of WSM-based p−n junction for E (cid:107) B ⊥ k0 and found that the magnetoconductance becomes a non- monotonic function of B (see. FIG. 2). We found the field corresponding to the maximum of G(B) to be equal to (cid:19) − 2B0 3B (cid:17) (cid:18) Bc ∼ 2 3 B0 ln(ζk0lE) ζk0lE (cid:29) 1. , (3) ψn(cid:54)=0 = 3 FIG. 3: Energy versus momentum along the magnetic field k (cid:107) B dependence (computed numerically) with parameters ζ = 1.6 and k0lB = 1.4 for different angles η = π/6, π/3, π/2. Material parameters are taken for TaAs 0/2m (cid:39) 12 meV and magnetic field is set to B (cid:39) 34 T to make the gap visible at η = π/2. 2k2 (cid:112)2n x − k2 0) (cid:115) where n ∈ Z and coefficients c1,2 malized solutions of eigenproblem (cid:32)− 2 x − k2 (cid:112)2n 2m (k2 vl−1 B 0) vl−1 2 2m (k2 B n are determined as nor- (cid:33)(cid:18)c1 (cid:19) n c2 n (cid:18)c1 (cid:19) n c2 n = εn As a result, we find εn(cid:54)=0(kx) = v · sgn(n) ε0(kx) = − 2 (cid:0)k2 2n l2 B x − k2 (cid:1) , 0 2m (cid:19)2 , (cid:18)(k2 x − k2 0) 2mv n ∈ Z. + . (6) (7) Here we observe that spectrum (7) possesses electron- hole symmetry for all non zero modes n → −n. How- ever, the symmetry is violated for zero mode n = 0 where only the hole state with negative energy exists. To make the whole picture more intelligible we present the plot of LLs on FIG. 4. The mentioned asymmetry, as the reader might have already guessed, is a remnant of the chiral anomaly. It is important to note that the zeroth LL is independent of magnetic field. It leads to linear in B magnetoconductance G(B) for large B in such ori- entation, see Section III. A similar scheme of LLs was obtained in paper Ref. 43 for η = 0 and almost identical Hamiltonian. Here we go further and retrieve LLs for any orientation of magnetic field with respect to k0 vector. η = ∠(B, k0) = π Hamiltonian becomes Field perpendicular to nodes splitting. For the case B the 2 after the shift y (cid:55)→ y + kxl2 H = v (cid:18) kz Q (cid:19) Q = W − i(cid:98)py, W = Q† −kz , (cid:16) y2 l4 B 2mv (cid:17) . − k2 0 (8) (cid:18) c1 nχ1n( y ) nχ2n(− y lB c2 lB (cid:19) . (9) ) To find the eigenvectors one may factorize y–dependent part of ψ–function as the solution of Qχ2 Q†χ1 n = −l−1 n = −l−1 B nχ1 n B nχ2 n , ψn = When such vector χ1,2 n (y) is found, eigenfunctions can be determined from the following linear equation (cid:18)kz − εn(kz) −n (cid:19)(cid:18)c1 (cid:19) n c2 n −n −kz − εn(kz) = 0. (10) which simultaneously determines εn(kz). As a result, we find that the energy disperses with the moment parallel to the magnetic field according to εn(kz) = v · sgn(n) z, n = ±0,±1, . . . (11) where ±0 denotes the ground states for electrons and holes respectively with convention sgn(±0) = ±1. + k2 2n l2 B We still have to solve the eigenproblem (9) to deter- mine n. Let us decouple equations via (cid:115) (cid:18) n (cid:19)2 lB (cid:18) n (cid:19)2 lB Q†Qχ1 n = n, QQ†χ2 χ1 n = χ2 n. (12) Operators H + = Q†Q and H− = QQ† in (12) form a hermitian couple. Written explicitly H± = p2 y + W (y)2 ∓ 2∂yW (y), (13) each of them can be associated with the Hamiltonian of a system. For description of the systems with Hamil- tonians given by (13) one can introduce supersymmetry generators where W (y) plays the role of superpotential. The problem can thus be reformulated in terms of super- symmetric quantum mechanics. Theoretical grounds of supersymmetric quantum me- chanics are well developed. In some cases they even allow to find the full exact spectrum of a system, us- ing Q± in a way analogous to creation and annihilation operators are used in ordinary quantum mechanics, see Refs 44 and 45. Here we need the two important state- ments of the theory: i) if the supersymmetry is unbroken, the ground state of the system is non degenerate and is exactly zero and ii) the supersymmetry is broken when W (+∞) · W (−∞) > 0. This breakdown leads to the ground state with positive energy. 4 From the one hand, it satisfies (15) with = 0. From the other hand, it can be expanded in powers of g to produce n–th order perturbation theory result for (15) implying that perturbatively (n) 0 = 0. In order to find correct non-perturbative energy levels, we resort to stan- dard WKB technique (see Appendix for details) and ob- tain for the zeroth level √ (cid:114) 0 = e−1/6g. ζ π (18) From here we recover the ground state energy (2). Re- sult (2) is valid up to g < 1 4 (i. e. for magnetic field up to B0) with error less then 3%. Higher levels are shifted by anharmonicity of the potential and splitted according to (cid:114) (cid:18) 2 (cid:19)k e−1/6g g k! ζ π 2k − 2k−1 = . (19) In Eq. (18) and (19) only the leading term in semiclassical expansion is retained. Finally, let us consider intermediate values of angle η. Since the spatial variables do not separate in this case, we computed the LL dependence on longitudinal (along magnetic field) momentum ε(k(cid:107)) numerically (see FIG. 3). One observes an interesting crossover from a field-independent level at η = 0, Eq. (7) to a level, weakly dependening on magnetic field, Eq. (11). The field dependence in the latter case is due to the gap be- tween ±0 levels, resulting from supersymmetry breaking of the underlying Hamiltonian at η = π/2. Interestingly, this gap turns into the gap between zeroth and first LL as η evolves from π/2 to 0. III. MAGNETOCONDUCTANCE We now evaluate the conductance in the presence of magnetic field perpendicular to p−n junction. We will make use of Landauer formalism and solve the scattering problem for electrons moving from conductance to va- lence band through p−n junction. As our discussion of the LLs suggests, the result will be qualitatively different for two orientations of the junction with respect to the nodes splitting: parallel η = 0 and perpendicular η = π 2 . In both cases, the spatial variables can be separated for longitudinal and transversal motion in the magnetic field. For η = 0 in the transversal motion there exists a (7). After substitu- (1 + iσy)ψ and separation of variables 2 n (y)φ1,2(x) with functions ψn given by field-independent mode, see Eq. tion ψ (cid:55)→ 1√ ψ1,2(x, y) = ψ1,2 (5), the scattering problems reads (cid:34)2(k2 x − k2 0) 2m √ v lB σz + 2nσx − eEa th (cid:17)(cid:35) (cid:16) x a φ = εφ. (20) FIG. 4: Landau levels for η = ∠(B, k0) = 0 plotted from Eq. (7). Magnetic field is supposed to satisfy sides of the junction are noted as µ± = µ(±∞). E and electrochemical potentials from different lBa < l2 We see that our function W (y), see Eq. (8) gives rise to a system with broken supersymmetry and non-zero de- generate ground state. To see this explicitly one can try to construct an exact ground state of the Hamiltonians H±. E.g. for H + we have: Qχ0(y) = 0 ⇔ ∂yχ0 = W (y)χ0 ⇒ χ0(y) = exp (y3 − 3k2 0l2 By) . (14) (cid:18) 6mvlB (cid:19) This solution although formally of zero energy, is not normalizable (the zero mode of H− suffers from the same disease). The spectrum of the problem can still be found analyt- ically for k0lB (cid:29) 1. In the limit of k0 → +∞ with ζ fixed, the Weyl cones become separated and the ψ–function √ near each Weyl point is given23 by an appropriate combi- nation of Hermite functions ψosc 2ζn. To n (y) with n = √ solve the present problem, we introduce a coupling con- stant g ≡ (4ζk2 B)−1 (cid:28) 1 and rescale y (cid:55)→ y/ 0l2 ζ. Using Eq. (12) we write down the corresponding Schrodinger equation (cid:34) (cid:18) −∂2 y + g y2 − 1 4g (cid:35) (cid:19)2 ∓ 2 √ gy χ1,2 = 2νχ1,2, (15) n/ζ. For g → 0 we have two where we denoted 2ν ≡ 2 (cid:17) independent harmonic oscillators and eigenfunctions are approximately n(cid:54)=0(y) ∼ ψosc χ1 0(y) ∼ ψosc χ1 (cid:17) − (−1)(cid:98) n (cid:17) (cid:16) √ y + 1 2 n(−y), y − 1 √ 2 y − 1 √ 2 χ2 n(y) = χ1 2 (cid:99)ψosc n−1 (cid:16) (cid:16) g g n 0 , , g with eigenvalues νn = (cid:100) n 1, ν2 = 1 etc.) 2(cid:101), n ∈ N0 (so that ν0 = 0, ν1 = Let us first discuss the zeroth level. Non-normalizable quasi-solution at zero energy is (16) χ=0(y+) = e− y2 2 −√ + g y3 + 3 , y+ = y − 1 √ 2 g . (17) Transmission coefficient for the zeroth level n = 0 is field-independent and for k0a (cid:29) 1 is T0 ≈ 1 (slight sup- pression from unity is due to the scattering between the nodes induced by built-in electric field). After account- ing for LL degeneracy this results in linear contribution to magnetoconductance, G(B) ∝ (e2/h)(BS/Φ0)T0, lB (cid:28) lE. Thus, for η = 0 the presence of a second nearby Weyl node does not change magnetoconductance qualitatively, as long as built-in electric field does not transfer particles between the nodes. The situation is very different for the junction perpen- dicular to k0. Longitudinal (z) and transverse (x,y) vari- ables can be separated in Landau gauge A = (−By, 0, 0) H v (cid:19)2 − k2 (cid:32)(cid:18) ky + σz σx + σy (cid:33) kx − y l2 B ζ 2k0 kz + z l2 E = 0 . Substitution ψ1,2 = eikxxχ1,2(yl−1 B − kxlB)φ1,2(zl−1 E ) (cid:20) leads to transverse equations (cid:0)y2 − k2 0l2 B (cid:1) σx − i∂yσy ζ 2k0lB (cid:21) TABLE I: Numerical values, taken from Ref. 8 for TaAs, and Refs. 9 and 39 for TaP. 5 2k0, A−1 ∆, meV ζ = vx/vy lE, A ζk0lE B0, T Bc, T TaAs (W2) TaP (W1) 0.0183 2 1.65 470 7 9 3 0.021 (cid:46) 2 1.6 720 12 11 3.7 and TaP9,39 and estimate critical parameters. We sup- pose doping is weak enough to estimate E (cid:39) ∆2/(v)3/2 according to Ref. 23. The numbers presented in TABLE I show that the sit- uation we consider is indeed possible in an experimental setup. As Bc < B0, effective coupling constant g corre- sponding to position of the maximum is indeed small and our WKB calculation is valid for such fields. χn(y) = nχn(y), (21) IV. CONCLUSIONS which is the same as (12), and scattering problem [(−i∂z)σz + (lE/lB)nσx + z] φ(z) = 0, which is equivalent the Landau-Zener one and the trans- (cid:1) is deter- mission coefficient T = exp(cid:0)−π(lE/lB)22 (cid:21) Producing summation over Landau levels mined by energies n. (cid:20) (22) n G(B) = e2 h S 2πl2 B exp −π l2 E l2 B 2 n (cid:88) n (23) and taking into account Eq. (2), we obtain dependence G(B) depicted at FIG. 2: (cid:20) (cid:16)−be− 4 3 (cid:17) a b G(B) = 2πG0b exp (cid:21) exp(−πb) 2 sinh(πb) + , (24) where we have introduced G0 ≡ 2 ζ e2 h S (2πlE)2 , a ≡ (ζk0lE)2, b ≡ ζ l2 E l2 B . (25) In the two–cone model magnetoconductance has a max- imum at magnetic field Bc ∼ 4ζ 3π Φ0k2 0 3 (ζk0lE)2 ln 4 , ζk0lE (cid:29) 1. (26) Eqs (24) and (26) are the main predictions of our pa- per. For a plot of magnetoconductance as a function of magnetic field G(B) see FIG. 2. To test feasibility of the found results we take numerical values for TaAs8 To conclude, we have studied the LL structure in WSM analytically (η = 0, π/2) and numerically (0 < η < π/2). Our analytical results are summarized in Eqs. (2) and (3) as well as FIG. 3. We believe that the gap in the LL spectrum predicted by this equation at n = 0 has al- ready been observed in experiment Ref. 39 (the authors explained it via numerical solution of the Schroedinger equation). Our study completes these findings via ana- lytical solution and numerical description of the crossover from η = 0 to η = π/2 with rotating magnetic field. We have also described how the tunnelling between Weyl nodes leads to the change in the behavior of magne- toconductance of p−n junction in WSM. We found that this tunnelling leads to the appearance of the character- istic field Bc, Eq. (26), at which the differential mag- netoconductance changes its sign. We believe the same feature would exist at intermediate angles, but due to the absence of separation of longitudinal and transversal motion at 0 < η < π/2 in the two-cone approximation we were not able to study this problem in more detail. In our treatment, we have completely discarded the in- fluence of disorder and interaction. It means that char- acteristic traversal time through a pn-junction should be smaller then the quasiparticle relaxation time. The transport relaxation time in TaAs was estimated in i.e. Ref. 46, τ = 7· 10−13 s and vF ≈ 0.5· 106 m/s. Then the width of the p−n junction should be less than ∼ 1µm. We have also neglected the Zeeman splitting which is neg- ligible as long as magnetic field is smaller than spin-orbit interaction scale which produces the spin-orbit splitting of the quasiparticle bands. For TaAs the correspond- ing magnetic field is estimated47,48 to be around 50 T. Therefore there exists plenty of space for purely orbital magnetic-induced tunnelling in the framework of low- energy Hamiltonian Eq. (1). Overall, we are positive that the undertaken analysis helps to shed some light on the structure of a realistic WSM in moderate and strong magnetic fields and hope that the predicted behavior of magnetoconductance of p−n junctions is going to be measured in the coming experiments. Appendix: Semiclassical computation of ground state energy of tilted supersymmetric double–well potential Eigenlevels of the Schrodinger equation (cid:34) (cid:18) (cid:35) (cid:19)2 − √ − 1 2 ∂2 + g 2 y2 − 1 4g gy χ(y) = νχ(y) (A.1) can be studied in the limit of g (cid:28) 1 via semiclassical ap- proximation. Proper quantization condition taking into account both perturbative and non-perturbative correc- tions in small g can be derived via uniform WKB49. Non- perturbative corrections to zeroth energy level (where they are the only ones), as well as to higher levels (fully determining the LL level splitting) can be found in the following way. g potential is quadratic Near the minimum y± = y∓ 1 √ 2 (ν+ ≡ ν, ν− ≡ ν − 1) (cid:2)−∂2 + y2± − (2ν± + 1)(cid:3) χ(y±) ≈ 0, (2y)νe− y2 cos πν(−2y)νe− y2 2 + √ 2 , πe ν ∼ ψosc Decaying at y → +∞ solution is given by Hermite func- tion ψosc ν (y) which asymptotes y± (cid:28) 1√ g . y2 2 Γ(−ν)(−y)ν+1 , y → +∞, y → −∞. We have to match ψosc the potential hump y ∼ const√ g ν with WKB solution, valid under ´ (cid:112)k(y) 0 k(z)dz y exp− ´ (cid:112)k(y) y 0 k(z)dz , (A.2) + C− exp χ = C+ (cid:18) y2 − 1 4g (cid:19)2 − 2 √ k(y)2 = g 6 gy − 2ν. (A.3) To this end, we expand semiclassical action at y (cid:46) 1 √ 2 g √ gy (cid:19) (cid:112)(1 − z2)2 − 16g(z + 2ν) dz ∼ (cid:12)(cid:12)(cid:12)(cid:12)(cid:21)(cid:12)(cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)(cid:12) 1 − z ln(cid:12)(cid:12)1 − z2(cid:12)(cid:12) + ν ln 1 + z 0 √ gy S(y) = (cid:20) 1 (cid:18) 2 0 1 8g z − z3 3 + 8g 1 2 ∼ and near the minimum y± (cid:28) 1√ √ ln [−4 + g S(y) ∼ 1 12g ∼ − 1 12g − y2 + 2 y2− 2 + 1 2 1 2 gy+] + ν ln [−√ √ √ gy+] − ν ln [ ln [4 gy+] , (cid:16) g (cid:17)2ν 2 + where we neglected terms O(cid:0)y3±(cid:1). Matching (A.2) with gy−] asymptotics of Hermite functions we derive the quanti- zation condition Γ(ν)Γ(1 + ν) tan2 πν = − 1 3g , e π 2 (A.4) 3g /2π and for ν → N+ which for ν → 0 becomes ν0 = e − 1 gives energy level splittings due to inter-well tunnelling. These results have been derived via instanton technique in Refs. 50 and 51. FIG. 5: Tilted supersymmetric double–well potential (A.1) and its energy levels νn with coupling constant g = 2−6. 1 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 4 A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. B (2005). 2 X. Wan, A. M. Turner, A. Vishwanath, Savrasov, Phys. Rev. B 83, 205101 (2011). and S. Y. 5 G. B. Hal´asz and L. Balents, Phys. Rev. B 85, 035103 84, 235126 (2011). (2012). 3 A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 6 O. Vafek and A. Vishwanath, Annual Review of Condensed (2011). Matter Physics 5, 83 (2014). 7 7 B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, Z. Fang, X. Dai, T. Qian, and H. Ding, Phys. Rev. X 5, 031013 (2015). 8 B. Q. Lv, N. Xu, H. M. Weng, J. Z. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, C. E. Matt, F. Bisti, V. N. Strocov, J. Mesot, Z. Fang, X. Dai, T. Qian, M. Shi, and H. Ding, Nature Physics 11, 724 (2015). 9 N. Xu, H. M. Weng, B. Q. Lv, C. E. Matt, J. Park, F. Bisti, V. N. Strocov, D. Gawryluk, E. Pomjakushina, K. Conder, N. C. Plumb, M. Radovic, G. Auts, O. V. Yazyev, Z. Fang, X. Dai, T. Qian, J. Mesot, H. Ding, and M. Shi, Nature Communications 7, 11006 (2016). (2014). 26 Y. Zhao, H. Liu, C. Zhang, H. Wang, J. Wang, Z. Lin, Y. Xing, H. Lu, J. Liu, Y. Wang, S. M. Brombosz, Z. Xiao, S. Jia, X. C. Xie, and J. Wang, Phys. Rev. X 5, 031037 (2015). 27 M. Novak, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. B 91, 041203 (2015). 28 J. Du, H. Wang, Q. Chen, R. Mao, QianHuiand Khan, B. Xu, Y. Zhou, Y. Zhang, J. Yang, B. Chen, C. Feng, and M. Fang, Science China Physics, Mechanics & Astronomy 59, 657406 (2016). 29 V. V. Cheianov and V. I. Fal'ko, Phys. Rev. B 74, 041403 (2006). 10 H. Weng, C. Fang, Z. Fang, A. Bernevig, and X. Dai, 30 A. V. Shytov, N. Gu, and L. S. Levitov, arXiv:0708.3081 Phys. Rev. X 5, 011029 (2015). 11 S. Jeon, B. B. Zhou, A. Gyenis, B. E. Feldman, I. Kimchi, A. C. Potter, Q. D. Gibson, R. J. Cava, A. Vishwanath, and A. Yazdani, Nature Materials 13, 851 (2014). 12 J. Xiong, S. K. Kushwaha, T. Liang, J. W. Krizan, M. Hirschberger, W. Wang, R. J. Cava, and N. P. Ong, Science 350, 413 (2015). 13 S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, B. Wang, N. Alidoust, G. Bian, M. Neupane, C. Zhang, S. Jia, A. Bansil, H. Lin, and M. Z. Hasan, Nature Com- munications 6, 7373 (2015). 14 Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosi´c, A. Fedorov, R. Zhong, J. Schneeloch, G. Gu, and T. Valla, Nature Physics 12, 550 (2016). 15 S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.-C. Lee, S.- M. Huang, H. Zheng, J. Ma, D. S. Sanchez, B. Wang, A. Bansil, F. Chou, P. P. Shibayev, H. Lin, S. Jia, and M. Z. Hasan, Science 349, 613 (2015). 16 S.-Y. Xu, N. Alidoust, I. Belopolski, Z. Yuan, G. Bian, T.-R. Chang, H. Zheng, V. N. Strocov, D. S. Sanchez, G. Chang, C. Zhang, D. Mou, Y. Wu, L. Huang, C.-C. Lee, S.-M. Huang, B. Wang, A. Bansil, H.-T. Jeng, T. Neupert, A. Kaminski, H. Lin, S. Jia, and M. Zahid Hasan, Nature Physics 11, 748 (2015). 17 J. Xiong, S. K. Kushwaha, T. Liang, J. W. Krizan, W. Wang, R. Cava, and N. Ong, arXiv:1503.08179 (2015). 18 F. Arnold, C. Shekhar, S.-C. Wu, Y. Sun, R. D. Dos Reis, N. Kumar, M. Naumann, M. O. Ajeesh, M. Schmidt, A. G. Grushin, J. H. Bardarson, M. Baenitz, D. Sokolov, H. Bor- rmann, M. Nicklas, C. Felser, E. Hassinger, and B. Yan, Nature Communications 7 (2016). 19 H. Murakawa, M. S. Bahramy, M. Tokunaga, Y. Kohama, and C. Bell, Y. Kaneko, N. Nagaosa, H. Y. Hwang, Y. Tokura, Science 342, 1490 (2013). 20 C. M. Wang, H.-Z. Lu, and S.-Q. Shen, Phys. Rev. Lett. 117, 077201 (2016). 21 D. T. Son and B. Z. Spivak, Phys. Rev. B 88, 104412 (2013). 22 J. Klier, I. V. Gornyi, and A. D. Mirlin, Phys. Rev. B 92, 205113 (2015). (2007). 31 L. M. Zhang and M. M. Fogler, Phys. Rev. Lett. 100, 116804 (2008). 32 A. Andreev, Phys. Rev. Lett. 99, 247204 (2007). 33 W. Chen, A. V. Andreev, E. G. Mishchenko, and L. I. Glazman, Phys. Rev. B 82, 115444 (2010). 34 J. Wang, X. Chen, B.-F. Zhu, and S.-C. Zhang, Phys. Rev. B 85, 235131 (2012). 35 L. V. Keldysh, Sov. Phys. JETP 6, 763 (1958). 36 A. G. Aronov and G. E. Pikus, Sov. Phys. JETP 24, 188 (1967). 37 L. Esaki and R. R. Haering, Journal of Applied Physics 33, 2106 (1962). 38 H. Nielsen and M. Ninomiya, Physics Letters B 130, 389 (1983). 39 C.-L. Zhang, S.-Y. Xu, C. M. Wang, Z. Lin, Z. Z. Du, C. Guo, C.-C. Lee, H. Lu, Y. Feng, S.-M. Huang, G. Chang, C.-H. Hsu, H. Liu, H. Lin, L. Li, C. Zhang, J. Zhang, X.-C. Xie, T. Neupert, M. Z. Hasan, H.-Z. Lu, J. Wang, and S. Jia, Nature Physics 13, 979 (2017). 40 T. E. O'Brien, M. Diez, and C. W. J. Beenakker, Phys. Rev. Lett. 116, 236401 (2016). 41 C.-K. Chan and P. A. Lee, arXiv:1708.06472 (2017). 42 R. Okugawa and S. Murakami, Phys. Rev. B 89, 235315 (2014). 43 H.-Z. Lu, S.-B. Zhang, and S.-Q. Shen, Phys. Rev. B 92, 045203 (2015). 44 L. E. Gendenshtein, Pis'ma Zh. Eksp. Teor. Fiz. 38, 299 (1983). 45 L. E. Gendenshtein and I. V. Krive, Usp. Fiz. Nauk 146, 553 (1985). 46 C.-L. Zhang, S.-Y. Xu, I. Belopolski, Z. Yuan, Z. Lin, B. Tong, N. Alidoust, C.-C. Lee, S.-M. Huang, T.-R. Chang, et al., Nature Commun. 7, 1073 (2016). 47 F. Arnold, M. Naumann, S.-C. Wu, Y. Sun, M. Schmidt, H. Borrmann, C. Felser, B. Yan, and E. Hassinger, Phys- ical review letters 117, 146401 (2016). 48 B. Ramshaw, K. Modic, A. Shekhter, P. J. Moll, M. Chan, J. Betts, F. Balakirev, A. Migliori, N. Ghimire, E. Bauer, et al., arXiv:1704.06944 (2017). 49 G. V. Dunne and M. Unsal, Phys. Rev. D 89, 105009 23 S. Li, A. V. Andreev, and B. Z. Spivak, Phys. Rev. B 94, (2014). 081408 (2016). 50 I. I. Balitsky and A. V. Yung, Nuclear Physics B 274, 475 24 B. Z. Spivak and A. V. Andreev, Phys. Rev. B 93, 085107 (1986). (2016). 51 U. Jentschura and J. Zinn-Justin, Physics Letters B 596, 25 L. P. He, X. C. Hong, J. K. Dong, J. Pan, Z. Zhang, and S. Y. Li, Phys. Rev. Lett. 113, 246402 J. Zhang, 138 (2004).
1809.01938	1	1809	2018-09-06T12:18:43	Experimental demonstration and in-depth investigation of analytically designed anomalous reflection metagratings	[ "physics.app-ph", "physics.optics" ]	We present the design, fabrication and experimental investigation of a printed circuit board (PCB) metagrating (MG) for perfect anomalous reflection. The design follows our previously developed analytical formalism, resulting in a single-element MG capable of unitary coupling of the incident wave to the specified (first order) Floquet-Bloch (FB) mode while suppressing the specular reflection. We characterize the MG performance experimentally using a bistatic scattering pattern measurement, relying on an original beam-power integration approach for accurate evaluation of the coupling to the various modes across a wide frequency range. The results show that highly-efficient wide-angle anomalous reflection is achieved, as predicted by the theory, with a relatively broadband response. In addition, the MG is found to perform well when illuminated from different angles, acting as a multichannel reflector, and to scatter efficiently also to higher-order FB modes at other frequencies, exhibiting multifunctional capabilities. Importantly, the merits of the introduced beam-integration approach, namely, its improved resilience to measurement inaccuracies or noise effects, and the implicit accommodation of different effective aperture sizes, are emphasized, highly relevant in view of the numerous recent experimental reports on anomalous reflection metasurfaces. Finally, we discuss the source for losses associated with the MG; interestingly, we show that these correlate well with edge diffraction effects, rather than (commonly assumed) power dissipation. These experimental results verify our theoretical synthesis scheme, showing that highly-efficient anomalous reflection is achievable with a realistic fabricated MG, demonstrating the practical applicability and potential mutifunctionality of analytically designed MGs for future wavefront manipulating devices.	physics.app-ph	physics	Experimental demonstration and in-depth investigation of analytically designed anomalous reflection metagratings Oshri Rabinovich,1, 2 Ilan Kaplon,2 Jochanan Reis,2 and Ariel Epstein1, ∗ 1Andrew and Erna Viterbi Faculty of Electrical Engineering, Technion - Israel Institute of Technology, Haifa 3200003, Israel 2Rafael, Advanced Defense Systems, Ltd. Haifa 31021, Israel (Dated: September 7, 2018) We present the design, fabrication and experimental investigation of a printed circuit board (PCB) metagrating (MG) for perfect anomalous reflection. The design follows our previously developed an- alytical formalism, resulting in a single-element MG capable of unitary coupling of the incident wave to the specified (first order) Floquet-Bloch (FB) mode while suppressing the specular reflection. We characterize the MG performance experimentally using a bistatic scattering pattern measurement, relying on an original beam-power integration approach for accurate evaluation of the coupling to the various modes across a wide frequency range. The results show that highly-efficient wide-angle anomalous reflection is achieved, as predicted by the theory, with a relatively broadband response. In addition, the MG is found to perform well when illuminated from different angles, acting as a multichannel reflector, and to scatter efficiently also to higher-order FB modes at other frequencies, exhibiting multifunctional capabilities. Importantly, the merits of the introduced beam-integration approach, namely, its improved resilience to measurement inaccuracies or noise effects, and the im- plicit accommodation of different effective aperture sizes, are emphasized, highly relevant in view of the numerous recent experimental reports on anomalous reflection metasurfaces. Finally, we discuss the source for losses associated with the MG; interestingly, we show that these correlate well with edge diffraction effects, rather than (commonly assumed) power dissipation. These experimental results verify our theoretical synthesis scheme, showing that highly-efficient anomalous reflection is achievable with a realistic fabricated MG, demonstrating the practical applicability and potential mutifunctionality of analytically designed MGs for future wavefront manipulating devices. I. INTRODUCTION Metagratings (MGs), periodic structures comprised of one or a few polarizable meta-atoms per period, are at- tracting considerable attention lately [1 -- 7] , due to their ability to manipulate beams with very high efficiencies, usually difficult to achieve with gradient metasurfaces (MS) [8, 9]. Furthremore, they can also overcome real- ization difficulties related to conventional Huygens' and bianisotropic MSs, as the latter require microscopic de- sign of multiple closely-packed meta-atoms per period, typically corresponding to a significantly larger number of design degrees of freedom [10 -- 18]. In particular, the concept of metagratings for anoma- lous reflection was rigorously introduced and formulated in a recent paper by Ra'di et al. [5] using a repeat- ing magnetically-polarizable particle (conducting loop) above a perfect electric conductor (PEC) ground plane. In the wide-angle reflection scenarios considered therein, the transverse momentum difference between the incident wave and the desirable reflected wave defines a grating periodicity Λ that allows only two Floquet-Bloch (FB) modes to propagate: the specular (fundamental) and anomalous (first order) reflection modes; the rest of the modes are evanescent [19]. Subsequently, the authors showed that enforcing two physical constraints, namely, ∗ [email protected] cancellation of specular reflection and global power con- servation, allows synthesis of perfect anomalous reflection MGs using only two degrees of freedom. First, based on these two constraints, a nonlinear condition was analyt- ically formulated, resolving the required MG-PEC sep- aration distance h; next, once this parameter was set, the meta-atom geometry was tuned to optimize the cou- pling to the desirable first-order FB mode. This yielded an appealing theoretical scheme for designing simple yet highly efficient beam deflectors; nonetheless, it has yet to be experimentally tested. Lately, analytical models for electrically-polarizable MGs were developed, relying on arrays of dielectric rods [20] or capacitively-loaded wires [21] for achieving wide- angle anomalous reflection and beam splitting. In fact, the theoretical framework was augmented in [21] to al- low complete semianalytical design of the MG, up to the conductor geometry of the printed capacitors. Notwith- standing, although these reports marked a significant ad- vance towards an experimental demonstration by utiliz- ing meta-atom geometries more compatible with common 2D fabrication techniques, they still considered the MGs to be suspended in air above the ground plane, thus not suitable for practical realization. Aiming at such a realization, we have extended in a recent paper [22] the model of [21] as to consider a struc- ture fully-compatible with printed circuit board (PCB) implementation, namely, having a dielectric substrate be- tween the MG and the PEC (Fig. 1). This extension 2 FIG. 1. Physical configuration of the PCB MG realized and characterized herein. (a) Side view of the MG configuration (b) Top view of the MG, featuring the capacitively-loaded wires. The printed capacitor unit cells are repeating with a subwavelength period l (cid:28) λ along the x axis. comprised of loaded strips with period Λ, defined on a dielectric substrate backed by a metal (PEC) layer. (c) The MG printed capacitor unit cell, featuring a single degree of freedom, namely, the capacitor width W . The trace separation is s, its width is w, and the copper thickness is t. required accounting for multiple reflections within the Fabry-P´erot etalon formed by the PEC-backed dielectric slab, introducing modal reflection coefficients into the formulation; and incorporation of an effective dielectric constant when deriving the printed capacitor geometry, positioned at the interface between air and the dielectric [22]. The realistic analytical model produces detailed fabrication-ready design specifications of PCB MGs for perfect anomalous reflection, obtained without requiring even a single simulation in full-wave solvers. These re- sults naturally pave the path to experimental demonstra- tion of single-element MGs designed following this ana- lytical scheme, yet to be carried out. With the recent advances in the research of MSs for anomalous reflection [15, 16], experiments conducted in order to characterize these structures and verify their functionality at microwave frequencies were reported [6, 7, 23, 24]. In order to measure both the specular and anomalous reflections, the experimental setup often consisted of a single rotating element: either the MS was rotated while the transmitter and receiver antennas were fixed [23], or the receiver antenna was rotated while the transmitter and MS were fixed [6]. In both cases, the efficiency of anomalous reflection was deduced via com- parison to the scattering from a metallic plate of the same size. Furthermore, the power coupled to the various FB modes was inferred from the peaks of the corresponding beams (a single measurement point for each propagat- ing FB mode); hence, to achieve meaningful results, the authors had to factor out the differences in effective aper- ture sizes at different observation angles [25]. In this paper, we present the design, fabrication and experimental characterization of a MG that is capable of reflecting an incoming wave into a prescribed angle (anomalous reflection) with very high efficiency, validat- ing our analytical model [22]. In contrast to previous realizations of this functionality, our MG is exclusively designed, up to the complete PCB layout, following a rigorous analytical scheme [22], featuring only a single element with a single degree of freedom per period. The in-depth investigation we conduct examines the angular response of the MG at various frequencies, revealing its potential as a multifunctional device, and probes the MG performance when illuminated from non-designated an- gles, indicating its broad acceptance angle. Overall, our results verify experimentally that MGs can generally im- plement highly-efficient anomalous reflection over a rela- tively wide range of frequencies and angles of incidence, as pointed out in [5, 6, 21, 22]. In addition to providing experimental proof to the practical viability of analytically designed MGs, we sug- gest here an original approach to the characterization of anomalous reflection from planar scattering devices such as MSs and MGs, differing from the common method- ologies [6, 7, 23, 24]. Our proposed experimental scheme relies on measurements of the bistatic scattering from the MG, utilizing a fine angular resolution over the half space z > 0. Similar to [6], we illuminate both the MG and the reference metal plate from the designated an- gle of incidence and record the scattering pattern in all directions. However, in contrast to these previous exper- imental reports, we employ a beam-integration approach to characterize the coupling efficiency to the various FB modes. Specifically, we identify the scattered beam corre- sponding to each one of the propagating FB modes, and integrate the pattern over the angular range associated with it to estimate the power coupled to that mode. The proposed approach has two appealing features. First, it relies on an integrated set of measurement points, thus reducing the impact of measurement noise. Second, integrating over the entire beam ensures that all of the radiated power associated with it is properly accounted for, avoiding the need to explicitly take into consideration the differences in effective aperture sizes when comparing beams at different angles [6, 7, 23, 24] (such a calibration may be especially challenging when the measurements are not entirely in the far field). In view of the growing interest in MSs and MGs implementing controlled reflec- tion, the presented alternative approach, which may lead to improved experimental accuracy and can be applied to other scattering scenarios as well, forms a valuable contribution of this work. Finally, we quantify the losses in the fabricated MG, and investigate their origin. Surprisingly, our combined experimental and theoretical evidence implies that the observed losses are not related to dissipation in the di- electric substrate and metallic traces, but rather stems from the finite size of the characterized sample, leading to increased edge diffraction, which is interpreted as a loss mechanism in our setup. These results are impor- tant for two reasons: first, they establish the MG as a low-absorption device, as predicted by realistic simula- tions [21, 22]; second, they highlight a loss mechanism that is hardly considered in MG and MS experimental work, although all measurement setups involve finite de- vices that could be prone to the same phenomenon. Overall, the comprehensive experimental work re- ported herein establishes analytically-designed MGs as a viable alternative to MSs for the realization of pla- nar, highly-efficient, wide-angle beam manipulating de- vices, performing very well in the presence of realistic fabrication inaccuracies, conductor and dielectric losses. Together with the demonstrated potential for multifunc- tionality, these results provide further support and mo- tivation for future exploration of this emerging concept. Equally important, the presented beam-integration ap- proach for characterizing coupling efficiencies and the new light shed on the origin of apparent losses in these devices are expected to enable proper interpretation of the measured results not only for MGs, but for general scattering configurations as well. II. THEORY, DESIGN, AND EXPERIMENT We follow the analytical scheme of [22] for the design of a MG prototype at K band that fully couples an in- coming plane wave at an angle of θin = 70◦ towards the non-specular direction θout = −10◦ [Fig. 1(a)]. The MG is comprised of capacitively-loaded wires defined on a metal-backed dielectric substrate of thickness h, cor- responding to a standard single-layer PCB layout [Fig. 1(b) and (c)]. The periodicity Λ along the y axis fa- cilitates the coupling between the incident and reflected wave momenta, thus is set as Λ = λ/sin θout − sin θin, typically comparable with the free-space wavelength λ of the excitation fields. On the other hand, the period- icity l along the x axis is chosen to be much smaller, l = λ/10 (cid:28) λ, such that we can treat the closely-spaced capacitors as a uniform distributed impedance loading of the current-carrying wires [5, 21, 26]. The full details regarding the synthesis procedure can be found in [22]; for completeness, we briefly repeat here the main formulation results, as applicable to the pro- totype designed herein. For given substrate permittivity ε2 and frequency of operation f = c/λ (c is the speed 3 of light in vacuum), the procedure focuses on setting the two degrees of freedom of the MG structure, namely, the substrate thickness h and the capacitor width W [Fig. 1(c)], such that full coupling of the incident wave towards the prescribed θout would be achieved. In the first step, we assess the optimal h by imposing the two design con- straints, namely, elimination of specular reflection and power conservation, on the modal field expressions, yield- ing the condition for perfect anomalous reflection, read- ing [22] (1) ρ = cos θout cos θin − 1 + R−1(h)2 1 + R0(h)2 = 0, (cid:113) where Rm = jγm tan βm,2h−1 the mth FB mode, γm = η2k2βm,1 η1k1βm,2 jγm tan βm,2h+1 is the reflection coefficient of is the mth mode wave- p − (k1 sin θin + 2πm k2 Λ )2 impedance ratio, and βm,p = is the longitudinal wavenumber. The index p stands for the medium in which the parameters are evaluated, with p = 1 being air and p = 2 being the dielectric substrate; correspondingly, ηp and kp are the wave impedance and wave number in medium p. We note that the periodicity Λ dictates the propaga- tion angle of the mth FB mode scattered off the MG via the parameter βm,1. Following the guidelines of [5, 21, 22], it can be verified that the period defined at the beginning of this section forms a FB series in which the m = −1 mode coincides with the desirable anomalous reflection toward θout, and all modes other than the ones corresponding to specular and anomalous reflections are evanescent ((cid:61){βm,1} < 0,∀m (cid:54)= 0,−1). For this reason, demanding that the specular reflection would vanish via Eq. (1) guarantees exclusive coupling of the real power to the m = −1 mode, as no other radiation channels are available. Once the substrate thickness h minimizing ρ of Eq. (1) for the given θin and θout is found, we proceed to set- ting the MG's second degree of freedom, the capacitor width W , assessed from the load capacitance C required to establish the suitable grid currents to sustain the scat- tering phenomena. Indeed, this capacitance is resolved by invoking Ohm's law with these required currents and the total fields acting on the wire shell [22, 27]. This leads to the following expression for C (cid:21) + 1 γ−1 tan β−1,2h (cid:21) , (2) (cid:20) 1 2πlf C =− η1 1 + R−12 4Λ cos θout (cid:18) 1 ∞(cid:88) 2 1 (cid:19) (cid:20) k1(1 + Rm) γ0 tan β0,2h πw 2Λ + log m=−∞ m(cid:54)=0,−1 2αm,1 + k1η1 2π − η1 Λ − k1Λ 4π 1 m where w is the width of the wire trace [Fig. 1(c)], and we define αm,p = −jβm,p ; substituting the optimal h found earlier allows evaluation of the required load capacitance 4 FIG. 2. Deviation from the perfect anomalous reflection con- dition ρ of Eq. (1) (in dB scale), as a function of the sub- strate thickness h and the reflection angle θout, for the con- sidered MG prototype configuration (θin = 70◦). The dashed purple vertical lines indicate the solution branches where the deviation ρ is minimal. We choose the first branch as a preferable solution (see discussion in [22]). C to implement the desirable functionality. Finally, when the capacitance is realized as a printed capacitor with identical trace width and separation w = s [Fig. 1(b)], the capacitor width W can be retrieved from C using W ≈ 2.85KcorrC/εeff [mm/fF], where Kcorr is a frequency dependent correction factor, and the average dielectric constant is εeff = (ε1 + ε2) / (2ε0) [21, 22, 28]. To apply the outlined procedure to our MG prototype, we merely need to substitute the relevant parameters into Eqs. (1) and (2) and solve for the fabrication-ready de- sign specifications. We chose to use a Rogers RO3003 laminate of thickness h = 60mil = 1.524mm as the sub- strate for the prototype, featuring anisotropic permit- tivity of ε2,xx = ε2,yy = 3.39ε0, ε2,zz = 3ε0, and loss tangent of tan δ = 0.001. The laminate arrived cov- ered with 0.5oz. electrodeposited copper, corresponding to trace thickness of t = 18µm; to operate well within standard PCB fabrication constraints, we set the copper trace width and separation to w = s = 6mil = 0.1524mm [Fig. 1(c)]. We used these values in Eq. (1) and solved the equa- tion for ρ graphically, as illustrated in Fig. 2; the so- lutions correspond to valleys in the plot (purple dahsed lines), where ρ approaches 0. The graph revealed that perfect anomalous reflection with θin = 70◦, θout = −10◦, and the chosen h = 60mil, can be obtained at the working frequency f = 25.18GHz, implying periodici- ties along the y and x directions of Λ = 10.69mm≈ 0.9λ and l = 1.19mm = 0.1λ, respectively; in terms of the wavelength, the resulting device thickness corresponds to h ≈ λ/8. Invoking Eq. (2), the required capacitance can be assessed, proportional to the printed capacitor width as discussed therein. Due to the anisotropic dielectric FIG. 3. Electric Field distributions (cid:60){Ex(y, z)} for the MG design for incident angle of θin = 70◦ at f = 25.18GHz. The analytical calculation (a) is compared with the full-wave simulation (b). The dashed black vertical lines indicate the metagrating plane (z = −h), while the black circles mark a small region around the meta-atom where deviation between the analytical prediction and the full-wave simulations is ex- pected due to the finite size of the printed capacitors [21, 22]. . constant of the substrate, we did not use the analytical approximation as is to retrieve W from C, but rather con- ducted a short sweep of the capacitor width around the estimated value, yielding an optimum at W = 1.39mm. The resulting MG was defined in a commercial full- wave simulation tool (CST Microwave Studio) to ver- ify the analytical model prior to fabrication. In these simulations, realistic losses were accounted for, using the standard copper conductivity value of σcopper = 5.96·107[S/m] in addition to the dielectric losses specified above. Figure 3 compares the electric field distributions as calculated using the analytical formulation [Fig. 3(a)] and as recorded in simulations [Fig. 3(b)]. As can be ob- served therein, excellent agreement between theory and simulations is obtained, except for the expected deviation at the close vicinity of the meta-atoms [21, 22]. Overall, these MG simulations imply that the final PCB layout should exhibit extremely high efficiency, with 98.3% of the incident power being anomalously reflected towards θout, limited only by minor absorption by the copper and laminate of 1.7%. In order to verify the design and characterize the pro- totype experimentally, a 9(cid:48)(cid:48) × 12(cid:48)(cid:48) MG was fabricated by PCB Technologies Ltd., Migdal Ha'Emek, Israel; the re- sulting board is shown in Fig. 4. To evaluate the device performance, we have assembled an experimental setup in an anechoic chamber in Rafael, Advanced Defense Sys- 5 III. RESULTS AND DISCUSSION A. Anomalous reflection Our first goal is to verify that the fabricated proto- type indeed implements the desirable functionality. To this end, it is required to process the raw data as to re- trieve the portion of power coupled to the various FB modes scattered off the MG. As denoted in Section I, in order to evaluate the coupling efficiency to the individ- ual modes as a function of the frequency, we utilize an original beam-integration approach. According to this scheme, for each of the considered frequencies, we first identify the beams corresponding to the various propa- gating modes. This is achieved by looking for peaks of scattered power around the modal propagation angles, as predicted by the FB theorem, namely [22] sin θ(m) out = sin θin + , m = 0,±1,±2... (3) 2πm k1Λ where the superscript (m) denotes the mth FB mode. Figure 6 presents these angles for the MG illuminated from θin = 70◦ across the frequency range considered in the experiment, comparing the theoretical predictions of Eq. (3) to the modal peaks identified in the measured data; it is apparent that the two agree very well. As can be observed in the figure, although the modal indices can take arbitrary integer values, for the considered angle of incidence only the m = 0 and m = −1 FB modes are propagating for all the frequencies in the relevant range (in consistency with Section II), whereas the m = −2 mode becomes propagating only at high frequencies; for higher order modes, θ(m) out of Eq. (3) will not be purely real, indicating their evanescent nature. Next, we define the beam boundaries for each mode at these angles around the beam peak where the scattered power drops below −20dB of the maximum; we denote these angles by θ(m)− and θ(m) + for the mth-order mode. Subsequently, we integrate the power over the angular range associated with the beam, and divide it by the total power scattered to all the FB modes to estimate the fraction of power coupled to each of the relevant modes. In other words, considering an excitation from θin, the coupling efficiency to the mth FB mode can be assessed, as a function of frequency, via (cid:90) θ(m) (cid:90) θ(n) (cid:88) θ(m)− + n θ(n)− η(m) cpl,θin (f ) = PMG,θin (f, θ)dθ + , (4) PMG,θin (f, θ)dθ where, as mentioned above, PMG,θin (f, θ) is the recorded scattering pattern for the MG when illuminated from θin at the frequency f , and the summation in the denomi- nator is over all the propagating modes in the tested sce- nario, according to the FB theorem. It should be noted FIG. 4. (a) The fabricated MG prototype. (b) Zoom in on three lateral periods of the MG. FIG. 5. Experimental setup assembled in an anechoic cham- ber, consisting of a transmitter and reciever standard horn antennas, the MG prototype board and a rotatable arm. The transmitter antenna and the MG are fixed in position while the reciever antenna rotates and records the scattered power at each angle. tems, Ltd. (Fig. 5), with the MG placed on a stand, and a pair of standard horn antennas (Schwarzbeck, BBHA9170) for excitation and measurement. The trans- mitter antenna was aligned at θin = 70◦ relative to the MG, illuminating it from a distance (approximately 2m ≈ 168λ); the receiving horn was mounted on a plastic arm connected to a rotatable stage, recording scattered signals approximately 1.2m ≈ 100λ away from the MG. Therefore, by rotating the receiver in angular steps of ∆θ = 0.2◦ while keeping the transmitter and the MG board fixed and aligned, we could measure the power scattered towards the region −90◦ < θ < 90◦. For each angle, the reflected power was measured as a function of frequency, covering the spectral range of the horn anten- nas 15GHz − 40GHz in steps of ∆f = 30MHz. Overall, the experiment retrieves the frequency-resolved (bistatic) scattering pattern of the MG when illuminated from θin, denoted as PMG,θin (f, θ). For reference, we have repeated the experiment with the MG replaced by a very thin metallic plate (made of aluminium) with the same di- mensions in the xy plane; correspondingly, the recorded scattering patterns are denoted by PMetal,θin (f, θ). 6 effective aperture size differences are inherently taken into account [29], rendering explicit calibration unnec- essary. This can be an advantage, since such a calibra- tion usually requires certain assumptions on the far-field nature of the measurement setup, which are not always easily met. Applying the beam-integration methodology to the scattering patterns recorded for the MG upon illumina- tion from the designated angle of incidence θin = 70◦ yields the results presented in Fig. 7 (dash-dotted blue curves), where the modal coupling efficiencies for the m = 0, m = −1, and m = −2 modes are plotted as a function of the excitation frequency (the rest of the FB modes for this scenario are evanescent). First, we inspect the coupling efficiency to the anomalous reflection mode m = −1 [Fig. 7(b)]. The measured efficiency reaches a near-unity value of 96.11%, peaking at f = 26.87GHz (vertical dashed cyan line), which we consider as the opti- mal operating frequency of the fabricated prototype from now on. The performance of the MG as an anomalous reflector is further verified by the corresponding scatter- ing patterns recorded at this frequency, presented in Fig. 8. The beam boundaries for the m = −1 and m = 0 modes as identified by the automated data analysis code appear as black dashed lines around the predicted angles (see vertical dashed cyan line in Fig. 6), clearly mark- ing the dominant scattering from the MG (solid blue) and a metallic plate of the same dimensions (dash-dotted red). As expected, the specular reflection from the MG is highly suppressed with respect to the reference metallic plate, with the scattered power predominantly coupled to the anomalous reflection mode. Indeed, these findings ex- perimentally confirm that the analytically designed PCB MG prototype, featuring a single polarizable particle gov- erned by a single degree of freedom, implements a highly efficient wide-angle anomalous reflection. In addition, as predicted by previous work [5, 6, 21, 22, 30], the frequency response of Fig. 7 reveals a moderate 90% (−0.5dB) bandwidth of about 8% for the coupling effi- ciency. A closer look into Fig. 7(b) reveals that the opti- mal operating frequency found in the experiment devi- ates by 6.7% from the design frequency denoted in Sec- tion II, as can also be observed by comparison to the coupling efficiency trend of the simulated design (dotted red curve). Besides minor fabrication errors and trans- mitter alignment difficulties, which could somewhat con- tribute to such a discrepancy, we associate this frequency shift mainly with a deviation between the actual laminate permittivity and the one used for the design. Indeed, when we consider a modified substrate permittivity of ε2 = 2.7ε0 [solid green curve in Fig. 7(b)], the simu- lations produce a frequency response which is in close agreement with the measured one across most of the fre- quency range, both in terms of the optimal operating frequency and the measured bandwidth. Moreover, con- sidering the coupling efficiencies to the specular reflection mode [Fig. 7(a)] and to the higher-order (m = −2) FB FIG. 6. The angles θ where the peak scattered power is ob- served for each of the propagating modes, as a function of frequency, for excitation from θin = 70◦. The theoretical pre- dictions (solid lines) following Eq. (3) are compared with the experimentally recorded angles (dash-dotted lines) for the m = 0 mode (green), the m = −1 mode (blue), and the m = −2 mode (red). Cyan and magenta vertical dashed lines denote, respectively, the frequencies where optimal anomalous reflection (Section III A) and beam splitting (Section III B) are observed experimentally. that although various definitions for the beam bound- aries can be considered, e.g., where the scattered power drops to 25dB or 30dB below the identified beam peak, we found that these do not affect much the evaluated coupling efficiencies, as long as the definitions are applied consistently to all modes. Considering the typical evaluation procedure of the coupling efficiencies in recent experimental work involv- ing anomalous reflection MSs at microwave frequencies [6, 23], it may be worthwhile to highlight two appealing features associated with the beam-integration approach followed herein. In previous work, the coupling efficiency was commonly assessed by comparing the peak power recorded at the angle of anomalous reflection for an MG (single measurement point) with the peak specular re- flection power measured for a metallic plate of the same dimensions, considering the latter as a good measure for the total scattered power [analogous to the denomina- tor in Eq. (4)]; due to the different observation angles in which these two peaks (anomalous and specular) are received, proper calibration of the results is required, taking into account the difference in effective aperture size [7, 25]. In contrast, the beam-integration approach utilized herein relies on a large number of measurement points for the evaluation of the coupled power, corre- sponding to the angular range associated with the mth FB mode beam. Thus, this evaluation method is ex- pected to be more robust to noise effects, averaging the measured data through the integration process [Eq. (4)]. Using this scheme has an additional positive side effect, as when the overall integrated beam power is considered, 7 FIG. 7. Coupling efficiency as a function of frequency to the (a) m = 0 (b) m = −1, and (c) m = −2 FB modes, for θin = 70◦. The experimental results (dash-dotted blue line) are compared to the results obtained via full-wave simulation with the designed parameters, i.e. εr,xx = 3.39, εr,yy = 3.39, εr,zz = 3 (dotted red line), and those from full-wave simulation with the modified effective dielectric constant εr = 2.7 (green solid line). Cyan and magenta vertical dashed lines denote, respectively, the frequencies where optimal anomalous reflection (Section III A) and beam splitting (Section III B) are observed experimentally. fective laminate permittivity suitable for interpreting the measured scattering patterns. B. Additional functionalities As often is the case with experimental work, the mea- surement setup allows probing the performance of the device under test (DUT) beyond the nominal operating conditions a priori considered by design. One such op- portunity is provided by the extended frequency range measured throughout the experiment. As can be deduced from Eq. (3) and Fig. 6, for frequencies above 28.9GHz, the FB theorem indicates that the second-order mode (m = −2) enters the light cone and can be detected in the far field. Scanning this high-frequency region in Fig. 7(a) reveals that the MG features another operating frequency for which specular reflection almost completely vanishes. More specifically, at the frequency f = 31.88GHz (verti- cal dashed magenta line in Figs. 6 and 7) the MG acts as a "perfect" anomalous beam splitter, coupling the wave incident from θin = 70◦ almost equally to the m = −1 (θ(−1) out = 4◦) and m = −2 (θ(−2) out = −55◦) modes [Fig. 7(b) and (c), respectively]. This can be further observed in the corresponding scattering pattern (Fig. 9); quanti- tatively, ∼ 47% of the scattered power is coupled to each one of these FB modes, while specular reflection is effec- tively suppressed, coupling less than 6% of the scattered power. This indicates the potential of MGs to meet the demands of multiband applications, performing different diffraction engineering tasks (different beam steering) at different frequencies. Another feature single-element MGs inherently exhibit is multichannel reflection. This term, as was utilized re- cently in the context of MSs [24] and MGs [4], refers to simultaneous high-efficiency coupling between waves of FIG. 8. Experimentally recorded scattering patterns for the designated angle of incidence θin = 70◦, at the optimal oper- ation frequency f = 26.87GHz (cyan vertical lines in Figs. 6 and 7). The received power as a function of θ as obtained for the MG board (solid blue) is compared to the profile obtained for the reference metallic plate (dash-dotted red) at the opti- mal frequency, clearly showing the expected suppression of the specular reflection and the efficient coupling to the anomalous reflection m = −1 FB mode. Inset: schematic description of the main scattering phenomena. mode [Fig. 7(c)], a very good agreement between simu- lations with the modified permittivity and the measured data is observable as well [31]. Hence, considering the ex- pected difference between the manufacturer provided val- ues, measured using a microstrip configuration, and the permittivity values relevant for our scattering scenario, involving plane-wave-like excitations [32], we regard the experimentally evaluated value of ε2 = 2.7ε0 as the ef- 8 At the operating frequency f = 26.87GHz found in Section III A (vertical dashed cyan line), Fig. 10(b) in- deed demonstrates that efficient anomalous reflection is retained even for a substantial deviation from the des- ignated angle of incidence, with coupling efficiencies to the m = −1 mode approaching ∼ 90% (accompanied by a notable bandwidth of about 15%). The corresponding scattering pattern is presented in Fig. 12, where it can be clearly seen, by comparison to the scattering from the ref- erence metallic plate, that the MG greatly suppresses the specular reflection while funnelling the scattered power to the anomalous reflection beam at θ(−1) out = −16◦. Finally, we quantify in Fig. 13 the ability of the MG to perform as a continuous multichannel reflector for different angles of incidence at the operating frequency f = 26.87GHz. Full-wave simulation results (solid lines) indicate that for a broad range of angles θin ∈ (45◦, 74◦), highly-efficient anomalous reflection (above 90%) can be achieved, funnelling the power to the angular range θout ∈ (−20◦,−5◦). The measured data points for the scenarios probed in the course of this work, θin = 70◦ and θin = 50◦ (diamond markers), indicate a very good correspondence with the simulation results, corroborat- ing the concept experimentally. Reciprocity implies sim- ilar multichannel reflection to be observed also for the range θin ∈ (−74◦,−45◦). Overall, the ability of MGs for multifunctionality, whether in terms of frequency or angle of incidence, is demonstrated. C. Loss mechanisms To complete the MG prototype characterization, we seek to quantify its losses. In previous work related to reflecting MSs [6, 23, 24], this was typically achieved by comparing the anomalous reflection peak power scattered off the MS with the specular reflection peak power scat- tered off a metallic plate with the same dimensions. As- suming the receiver and transmitter are located in the far field region with respect to the DUT, and measurement noise is sufficiently low, it is expected that for an ideal (namely, lossless and with unitary coupling efficiency) MS, the ratio between these two power maxima would correspond to the ratio between the respective effective aperture sizes of the two measurements (these are differ- ent due to the different observation angles [7, 25]). Alas, even if these assumptions are valid, the mea- surement technique described above can only yield the product of the two contributions to the total anomalous reflection efficiency ηtot = ηscatηcpl, namely, the scatter- ing efficiency (ηscat, the fraction of incident power that is scattered overall, and not dissipated, for instance), and the the coupling efficiency (ηcpl, the fraction of scattered power that is coupled to the m = −1 mode, and not to spurious modes). In contrast, the beam-integration approach employed in Section III A enables decoupling these two contributions, evaluating ηcpl solely based on the MG scattering patterns; this allows verification of the FIG. 9. Experimentally recorded scattering patterns for the designated angle of incidence θin = 70◦, at the frequency f = 31.88GHz (magenta vertical lines in Figs. 6 and 7). The received power as a function of θ as obtained for the MG board (solid blue) and the reference metallic plate (dash-dotted red) are shown. It is clear that in this working frequency specular reflection is strongly suppressed, while the power splits almost equally between the m = −1 and m = −2 FB modes, indicat- ing the MG acts as a beam splitter in this frequency. Inset: schematic description of the main scattering phenomena. different angles of incidence and anomalous reflection. In [24], this concept was demonstrated by designing MSs to implement a discrete set of reflection channels, harness- ing symmetry, reciprocity, and some post-optimization. For the MGs presented herein, on the other hand, con- tinuous multichannel reflection is achieved simply due to the non-resonant working point of the meta-atoms, lead- ing to certain insensitivity to deviation from the design parameters [5, 6, 21]. In particular, when the angle of in- cidence deviates from the designated value of θin = 70◦, scattering is expected to still be predominantly to the anomalous m = −1 FB mode, now featuring a different θ(−1) out as per Eq. (3). To probe this hypothesis, we have characterized the MG response for an excitation arriving from θin = 50◦, namely, 20◦ away from the nominal one. The corre- sponding coupling efficiencies to the anomalous reflec- tion (m = −1) FB mode appear in Fig. 10(b), show- ing, once more, that good agreement between simulation and measurements is achieved when the modified (effec- tive) permittivity extracted in Section III A is used (green solid curve). For the same illumination scenario, Fig. 11 presents the predicted [Eq. (3)] and measured scattering angles of the various modes, which, again, are in a very good agreement. This figure also highlights one possible reason for the discrepancies between simulation and mea- surement observed for f > 32GHz in Fig. 10(a) and (c), attributed to partial blockage of the transmitting horn under retroreflection θ(−2) out ≈ −θin = −50◦ (Fig. 11), causing underestimation of the coupling to the m = −2 mode. 9 FIG. 10. Coupling efficiency as a function of frequency to the (a) m = 0 (b) m = −1, and (c) m = −2 FB modes, for θin = 50◦. The experimental results (dash-dotted blue line) are compared to the results obtained via full-wave simulation with the designed parameters, i.e. εr,xx = 3.39, εr,yy = 3.39, εr,zz = 3 (dotted red line), and those from full-wave simulation with the modified effective dielectric constant εr = 2.7 (green solid line). The cyan vertical dashed line denotes the optimal operating frequency observed experimentally for the designated angle of incidence (Section III A), f = 26.87GHz. FIG. 11. The angles θ where the peak scattered power is observed for each of the propagating modes, as a function of frequency, for excitation from θin = 50◦. The theoretical predictions (solid lines) following Eq. (3) are compared with the experimentally recorded angles (dash-dotted lines) for the m = 0 mode (green), the m = −1 mode (blue), and the m = −2 mode (red). The cyan vertical dashed line denotes the optimal operating frequency observed experimentally for the designated angle of incidence (Section III A), f = 26.87GHz. diffraction control properties of the MG, independent of the losses, if exist. At the same time, however, the MG measurements alone do not allow quantification of the scattering efficiency. Consequently, we utilize the scattering patterns mea- sured for the reference metallic plate (dash-dotted red lines in Figs. 8, 9, and 12) to estimate ηscat for the various configurations, and investigate through it the possible loss mechanisms. In particular, we assume the metallic FIG. 12. Experimentally recorded scattering patterns for θin = 50◦, at the optimal operation frequency f = 26.87GHz resolved in Section III A for the designated angle of incidence (cyan vertical lines in Figs. 6 and 7). The received power as a function of θ as obtained for the MG board (solid blue) is compared to the profile obtained for the reference metallic plate (dash-dotted red) at this frequency, indicating that the MG manages to suppress well the specular reflection even for such a considerable deviation from the designated angle of in- cidence. Inset: schematic description of the main scattering phenomena. plate possesses negligible losses, such that the overall (in- tegrated) power scattered off it serves as a good reference to the overall power incident upon the MG (recall that the MG and the metal plate have the same dimensions, and are illuminated from the same angle of incidence). 10 FIG. 13. The MG as a continuous multichannel reflector for different input angles at the operating frequency f = 26.87GHz. The efficiency towards mode m = −1 given by full-wave simulation is presented in blue (corresponding to the left y-axis) while the output angle θout from the analyti- cal calculation [Eq. (3)] is presented in red (corresponding to the right y-axis). The experimentally measured data points corresponding to the incident angles of 50◦ and 70◦ are de- noted with X markers. (cid:82) π (cid:82) π ηscat,θin (f ) = Hence, the scattering efficiency can be estimated via 0 PMG,θin (f, θ)dθ 0 PMetal,θin (f, θ)dθ = 1 − Lossθin (f ) (5) where Lossθin (f ) is defined as the fraction of incident power that is lost in the sense that it is not scattered off the MG, as per the recorded patterns at f , thus cannot contribute to the desirable anomalous reflection. The loss as evaluated by applying Eq. (5) to the mea- sured scattering patterns for θin = 70◦ and θin = 50◦ is presented in Figs. 14(a) and 14(b), respectively. At the operating frequency f = 26.87GHz (dashed vertical lines), we observe that Loss70◦ ∼ 15% and Loss50◦ ∼ 30%. In the full-wave simulations presented in Section II for the prototype design, however, we recieved losses in the order of ∼ 2%. These are contributed by dielec- tric losses in the substrate of the MG and conduction losses in the copper, realistically accounted for in these simulations. In order to examine the origin of the increased loss values estimated from the experimental data, we show in Fig. 15 the results of full-wave simulations for en- hanced losses, in the dielectric as well as in the conduc- tors. For both angles of incidence considered in the ex- periment, θin = 70◦ [Fig. 15(a)] and θin = 50◦ [Fig. 15(b)], we present the predicted Lossθin(f ) when the loss tangent is gradually increased by an order of magnitude from the nominal tan δ = 0.001 to tan δ = 0.01, and the wire conductivity is reduced by a factor of 5 with re- spect to the typical copper conductivity. In all cases con- sidered, it can be observed that even for unrealistically FIG. 14. The experimentally evaluated loss of the MG ac- cording to Eq. (5), for (a) θin = 70◦ and for (b) θin = 50◦. The cyan dashed vertical line denotes the optimal operating frequency defined in Section III A. large material loss parameters (e.g., see green curves with tan δ = 0.01 and σ = 0.2σcopper), the dissipated power as predicted by full-wave simulations does not reach the values extracted from the experiments, nor do they ex- hibit similar trends (Fig. 14). Thus, we infer that even if the dissipation in the fabricated MG board is somewhat increased with respect to the manufacturer provided pa- rameters, this alone cannot account fully for the large values of power loss deduced from the experimental data (Fig. 14); it is evident that an additional mechanism substantially contributes to these observations. We propose here that the alleged losses presented in Fig. 14 do not correspond to dissipation in the MG, as one may initially suspect, but are mainly a result of edge diffraction. It is well known that finite surfaces, such as the MG and metal plate in our experiment, suffer from edge diffraction effects, stemming from the inevitable dis- tortion of the fields towards the edges and corners (with respect to the scattering solution of the ideal infinite peri- odic surface) [33]. Following the common physical optics 11 fects are actually expected to be more pronounced along this axis. Following Eq. (5), however, the estimation of losses in the MG is obtained via comparison to the overall scat- tering by the reference metal plate. Therefore, if the metallic plate and the MG feature similar edge diffrac- tion characteristics, proportional to the incident power, then the division in Eq. (5) should calibrate these effects out. Nonetheless, as we discuss in the following, the edge diffraction from a thin metallic sheet and the MG are fundamentally different. Scattering from a metallic plate is local in essence: the incident power is immediately be- ing reflected (or diffracted) at the point of incidence, and the scattered power is generally proportional to the lo- cal field impinging at this point [34, 35]. MGs, on the other hand, rely on a metal-backed dielectric configura- tion, which partially guides the incident power in the sub- strate while reradiating it as reflected waves away from the point of incidence [15, 23, 24, 36]. While the latter nonlocal phenomenon does not affect the power balance for an infinite MG (the incident, guided, and reflected modes exchange power periodically to form a uniform response), when finite MGs are considered, the guided power might abruptly meet the edge of the MG before full reradiation has been achieved. If this guided power is substantial, such an event could cause increased edge diffraction with respect to the locally reflecting metallic plate, subsequently interpreted by Eq. (5) as loss. Before providing further support to this hypothesis, two comments are in place. First, we note that the edge diffraction phenomenon as described in the previ- ous paragraph is completely consistent with the full-wave simulation results shown so far, indicating that essen- tially all of the incident power is coupled to anomalous reflection considering typical dielectric and conductor pa- rameters (Fig. 15). Indeed, when an infinite periodic MG is considered, no such "scattering losses" are expected (there are no edges to cause edge diffraction). Second, while power guided in the MG substrate is the apparent reason to the "losses" estimated based on the experimen- tal data (Fig. 14), this phenomenon is far more general, expected to reveal significant edge diffraction differences between any finite-size metal-backed dielectrics and thin metallic plates of the same dimensions. In fact, in order to reduce the complexity of analysis and simulations, we use this simpler configuration to demonstrate said phys- ical effects. Correspondingly, we calculate analytically the power guided in the substrate (relative to the incident power impinging the surface in a single period) for the scenario of Fig. 1(a), but excluding the conducting wires (see appendix). More specifically, the reltaive y-propagating real power in a dielectric slab of thickness h = 60mil when excited by a plane wave incoming from either θin = 70◦ or θin = 50◦ is evaluated as a function of the excitation frequency and plotted in Figs. 16(a) and 16(b), respec- tively; due to the uncertainty in the substrate permittiv- ity, the results for the two limiting values of ε2 = 2.7ε0 FIG. 15. MG loss as evaluated from full-wave simulations for various values of dielectric loss and copper conductivity, for angles of incidence (a) θin = 70◦ and (b) θin = 50◦. In- creasing substrate loss tangent values are considered, starting from the manufacturer provided tan δ = 0.001 (blue), through tan δ = 0.005 (red), to tan δ = 0.01 (green); in all of these simulations the nominal copper conductivity was used for the wires σ = σcopper. To demonstrate that even highly unrealis- tic loss parameters cannot fully account for the losses deduced experimentally (Fig. 14), the magenta curve considers a re- duced copper conductivity by a factor of 5, σ = 0.2σcopper, and loss tangent larger by a factor of 10, tan δ = 0.01, with respect to the nominal. The cyan dashed vertical line denotes the optimal operating frequency defined in Section III A. approximation, for instance, a plane wave propagating in the (cid:99)yz plane impinging upon the DUT, will not scatter only to the(cid:99)yz plane, as in the case of an infinite surface [Fig. 1(a)], but would also generate scattered waves with non-vanishing ±x wavevector component. As the trans- mitting and receiving horn antennas in our experimental setup are horizontally aligned with the center of the DUT (Fig. 5), namely, situated at the plane x = 0, the edge diffracted power radiating towards x > 0 and x < 0 will not be recorded, and might be interpreted as losses. In fact, due to the fact that the MG and metal plate are shorter along the x dimension, these edge diffraction ef- 12 FIG. 17. The loss associated with finite size metal-backed dielectric substrate when illuminated from θin = 70◦, eval- uated from full-wave simulations for different surface ar- eas and aspect ratios of the slab: 20mm × 100mm (dash- dotted cyan line), 25mm×100mm (dash-dotted magenta line), 50mm× 100mm (solid blue line), 75mm× 100mm (solid green line), 100mm× 100mm (solid black line), and 60mm× 240mm (solid red line). Note that while Ly is fixed on 100mm for most of these configurations, Lx is modified to inspect increasing aspect ratios (denoted in brackets in the plot legend). Fi- nally, the 25mm × 100mm (dash-dotted magenta line) and the 60mm × 240mm (solid red line) curves share the same aspect ratio [1 : 4] but different surface area, indicating that both surface area and aspect ratio are of importance in de- termining the edge scattering losses as defined herein. quency response plotted in Fig. 14, provides supporting evidence to the hypothesis that increased edge diffrac- tion is the main mechanism contributing to the reduced overall scattered power to the x = 0 plane as recorded for the MG, relative to the reference metal plate. In addition to the analysis of the guided power vari- ation with frequency and angle of incidence for the in- finite structure, we have also performed a parametric study examining the scattering off a finite size surface, this time using full-wave simulations (Fig. 17). To en- able simulations of large samples approaching the size of the fabricated prototype, we have again considered the simplified metal-backed substrate formation, excluding the conducting wires from the simulated configurations. Denoting the finite dimensions of the structures along the x and y axes as Lx and Ly, respectively, we have thus simulated several slabs with different Lx : Ly ra- tios and various overall surface area A = LxLy. Each of these slabs was illuminated by a plane wave incom- ing from θin = 70◦, and the scattering pattern along the x = 0 plane was extracted from the simulated 3D scat- tering pattern to emulate the experimental conditions. Subsequently, the numerical "experiment" was repeated with a corresponding finite-size metallic plate, and the expression in Eq. (5) was used to calculate the loss, in consistency with the methodology used to generate Fig. 14. FIG. 16. The power guided in the dielectric substrate rel- ative to the incident power impinging in a single period, as calculated analytically (see appendix) for (a) θin = 70◦ and (b) θin = 50◦. In view of the uncertainty in the substrate permittivities, the two limiting cases of ε2 = 2.7ε0 (blue solid line) and ε2 = 3.39ε0 (dash-dotted red line) considered in this paper, are presented. (solid blue) and ε2 = 3.39ε0 (dash-dotted red) considered herein are presented. As can be observed, the variation of the losses estimated based on the measurements with frequency and angle of incidence (Fig. 14) qualitatively correspond to the trend of the relative power guided in the substrate, especially for ε2 = 3.39ε0. In particular, in both Fig. 14 and Fig. 16 the loss peaks corresponding to θin = 70◦ are higher and narrower than those corre- sponding to θin = 50◦, and appear at higher frequencies [37]. Although the agreement is only qualitative (recall that we use a very simplified model here), it clearly ties be- tween the alleged "losses" of Fig. 14 and the guided power causing increased edge scattering for the MG with respect to the metal plate; similar correspondence is ob- served when considering the power guided in the sub- strate of the full MG structure, i.e., including the loaded wires (not shown). Combining these observations with the results presented in Fig. 15, indicating that dielectric and conductor losses cannot on their own explain the fre- This parametric study leads to several important con- clusions. First, one can observe that the edge diffraction losses can indeed surmount to significant fractions of the incident power (above 20% for some configurations and frequencies), and they become more severe as the aspect ratio Lx : Ly and the overall area A become smaller. In other words, the off-plane scattering becomes more dom- inant for the MG when the edges along the x axis are closer to the center of the surface, thus play a more cru- cial role in the scattering phenomena. However, this also points out a possible way to mitigate this effect, by uti- lizing square MGs with large areas. Second, it can be observed that for the aspect ratio relevant to the proto- type characterized herein, Lx : Ly = 3 : 4 (green solid line), the edge scattering losses are predicted to be quite moderate with respect to the ones found experimentally (Fig. 14). Therefore, we conclude that the transmitting and receiving horns were not perfectly aligned with re- spect to the MG, and non-uniform illumination of the surface has formed in practice a different effective as- pect ratio for the illuminated ares, featuring increased asymmetry leading to increased edge diffraction. Third, although the edge scattering losses as interpreted follow- ing Eq. (5) can be very significant for certain aspect ratios and surface areas, when the overall integrated 3D power scattered from the dielectric slab and reference metallic plate are compared, simulation results indicate that the overall dissipation in the substrate is minor for the finite structure (not shown), in consistency with the results obtained for the infinite structure (Fig. 15). These combined observations imply that the fabricated MG prototype indeed succeeds in implementing a highly- efficient anomalous reflection, not only in the sense of spurious scattering (Section III A), but also in the sense of power dissipation. Although it is hard to provide a quantitative limit to the amount of power absorbed in the MG, the investigation presented in this section clearly demonstrates that it is expected to be mild. A more reasonable mechanism to explain the observations of reduced overall scattered power with respect to the reference metallic plate (Fig. 14) would be edge diffrac- tion effects due to substrate-guided power. Nonethe- less, our analysis shows that this effect does not dimin- ish from the MG performance as anomalous reflector: it merely provides some x directed momentum to the anomalous reflection mode, without affecting the overall (3D) power balance. Furthermore, simulation results im- ply that these effects are expected to become less and less dominant as the prototype size increases and the Lx : Ly aspect ratio approaches unity, thus can be mitigated in practical applications. 13 ical design, fabrication, and testing of a MG prototype was described in detail, verifying its performance as pre- dicted by the analytical model. In addition to explor- ing the angular and frequency dependence of the proto- type, pointing out its potential for implementing multi- functional and multichannel planar reflectors, our report highlights two experimental aspects with relevance to the broad field of reflecting MSs and MGs. First, we have utilized an original beam-integration approach to evalu- ate the coupling efficiency to the various propagating FB modes. This assessment methodology, which was found to agree very well with full-wave simulations, is expected to be more resilient to measurement noise, and does not require explicit calibration with respect to changes in the effective aperture for different observation angles, non- trivial at times. Second, we shed light on the origin of alleged prototype losses estimated based on the compar- ison to a reference metal plate. In contrast to the com- mon notions, we showed that it is not plausible that the reduced scattering stems from dissipation in the dielec- tric substrate and metallic traces; rather, we presented evidence to support a different mechanism to generate similar reductions, namely, edge diffraction due to power guided in the MG substrate. Such a phenomenon, which can be observed in every finite-size devices supporting guided modes, could be relevant to experimental charac- terization of planar scattering formations in general. This practical demonstration of efficient analytically- designed MGs forms a crucial step in establishing the via- bility of this emerging concept for implementing realistic diffraction engineering devices. It indicates that as little as a single degree of freedom in the polarizable element (the printed capacitor width) is sufficient for achieving near-unity coupling of the incident power to a desirable mode, if the proper analytical model is harnessed to this end. These results, together with the more basic observa- tions regarding the experimental characterization of such devices, are expected to contribute to the development of future MG-based components for advanced beam manip- ulation. Appendix: Power guided in the dielectric substrate For completeness, we present here the derivation of the analytical expressions for the power guided in a metal- backed dielectric substrate, relative to the impinging power over a single period, used in Section III C to in- vestigate the expected edge diffraction effects in the cor- responding MG structure. For the configuration under consideration, namely that of Fig. 1(a) without the con- ducting wires, the electric field in the dielectric substrate (−h < z < 0) is given by [22] IV. CONCLUSION Ex,2(y, z) = To conclude, we have presented a complete demonstra- tion and thorough experimental investigation of a single- element PCB MG for anomalous reflection. The analyt- −Ein 2jγ0 1 + jγ0 tan β0,2 sin(β0,2z) cos(β0,2h) ejβ0,1he−jkt,0y (A.1) the amplitude of the incident field, where Ein is Ex,1 (y, z) = Eine−jβ0,1zejkt,0y, kt,m = k1 sin θin + 2πm Λ , and the parameters γm, βm,p are defined in Section II (note that for this scenario, where the periodic wire ar- ray is absent, only the zeroth order mode contributes to the scattered fields). Consequently, the z component of the magnetic field is given by Hz,2(y, z) = Ein 2jγ0 − kt,0 k2η2 ejβ0,1he−jkt,0y (A.2) Thus, the overall real power travelling in the dielectric in the y direction can be calculated using sin(β0,2z) cos(β0,2h) 1 + jγ0 tan β0,2 1 + jγ0 tan β0,22 cos2(β0,2h) (A.3) · kt,0 k2η2 h + 1 4β0,2 2 sin(−2β0,2h) , (cid:90) 0 −h (cid:60)(cid:8)Ex,2H∗ z,2 γ02 (cid:9)dz = (cid:21) 1 Py,2(y, z) = 1 2 2Ein2 (cid:20) 1 14 given by Pz,1(y, z) = (cid:90) Λ/2 −Λ/2 1 2 (cid:60)(cid:8)Ex,1H∗ y,1 (cid:9)dy = β0,1 2k1η1 ΛEin2 (A.4) , which is also independent of y and z. Overall, the rela- tive power guided in the dielectric substrate is given by Py,2/Pz,1, using which the plots presented in Fig. 16 were generated. ACKNOWLEDGMENTS This research was supported by the Israel Science Foundation (Grant No. 1540/18). The authors also wish to thank Rogers Corporation for providing the laminates used in this study. which is independent of y and z. We normalize this quan- tity to the power impinging the MG in a single period Λ, [1] M. Khorasaninejad and F. Capasso, "Broadband mul- tifunctional efficient meta-gratings based on dielectric waveguide phase shifters," Nano Lett. 15, 6709 -- 6715 (2015). [2] D. Sell, J. Yang, S. Doshay, R. Yang, and J. A. Fan, "Large-angle, multifunctional metagratings based on freeform multimode geometries," Nano Lett. 17, 3752 -- 3757 (2017). [3] Z. Fan, M. R. Shcherbakov, M. Allen, J. Allen, and G. Shvets, "Perfect diffraction with bianisotropic meta- gratings," arXiv preprint arXiv:1802.01269 (2018). [4] Y. Ra'di and A. Al`u, "Reconfigurable metagratings," ACS Photonics 5, 1779 -- 1785 (2018). [5] Y. Ra'di, D. L. Sounas, and A. Al`u, "Metagratings: Be- yond the limits of graded metasurfaces for wave front control," Phys. Rev. Lett. 119, 067404 (2017). [6] A. M. H. Wong and G. V. Eleftheriades, "Perfect anoma- lous reflection with a bipartite Huygens' metasurface," Phys. Rev. X 8, 011036 (2018). [7] A. M. H. Wong, P. Christian, and G. V. Eleftheriades, "Binary Huygens' metasurfaces: Experimental demon- stration of simple and efficient near-grazing retroreflec- tors for TE and TM polarizations," IEEE Trans. Anten- nas Propag. 66, 2892 -- 2903 (2018). [8] N. Yu, P. Genevet, M. A. Kats, F. Aieta, J.-P. Tetienne, F. Capasso, and Z. Gaburro, "Light propagation with phase discontinuities: generalized laws of reflection and refraction," Science 334, 333 -- 337 (2011). [9] N. M. Estakhri and A. Al`u, "Recent progress in gradient metasurfaces," J. Opt. Soc. Am. B 33, A21 -- A30 (2016). and C. L. Holloway, "Averaged transition conditions for elec- [10] E. F. Kuester, M. A. Mohamed, M. Piket-May, tromagnetic fields at a metafilm," IEEE Trans. Antennas Propag. 51, 2641 -- 2651 (2003). [11] C. Pfeiffer and A. Grbic, "Metamaterial Huygens sur- faces: tailoring wave fronts with reflectionless sheets," Phys. Rev. Lett. 110, 197401 (2013). [12] C. Pfeiffer and A. Grbic, "Bianisotropic metasurfaces for optimal polarization control: Analysis and synthesis," Phys. Rev. Appl. 2, 044011 (2014). [13] A. Epstein and G. V. Eleftheriades, "Huygens metasur- faces via the equivalence principle: design and applica- tions," J. Opt. Soc. Am. B 33, A31 -- A50 (2016). [14] A. Epstein and G. V. Eleftheriades, "Arbitrary power- conserving field transformations with passive lossless omega-type bianisotropic metasurfaces," IEEE Trans. Antennas Propag. 64, 3880 -- 3895 (2016). [15] A. Epstein and G. V. Eleftheriades, "Synthesis of passive lossless metasurfaces using auxiliary fields for reflection- less beam splitting and perfect reflection," Phys. Rev. Lett. 117, 256103 (2016). [16] V. S. Asadchy, M. Albooyeh, S. N. Tcvetkova, A. D´ıaz- Rubio, Y. Ra'di, and S. A. Tretyakov, "Perfect control of reflection and refraction using spatially dispersive meta- surfaces," Phys. Rev. B 94, 075142 (2016). [17] G. Lavigne, K. Achouri, V. S. Asadchy, S. A. Tretyakov, and C. Caloz, "Susceptibility derivation and experimen- tal demonstration of refracting metasurfaces without spurious diffraction," IEEE Trans. Antennas Propag. 66, 1321 -- 1330 (2018). [18] M. Chen, E. Abdo-S´anchez, A. Epstein, and G. V. Eleft- heriades, "Theory, design, and experimental verification of a reflectionless bianisotropic Huygens' metasurface for wide-angle refraction," Phys. Rev. B 97, 125433 (2018). [19] P. St. J. Russell, "Optics of floquet-bloch waves in dielec- tric gratings," Appl. Phys. B: Lasers Opt. 39, 231 -- 246 (1986). [20] H. Chalabi, Y. Ra'di, D. L. Sounas, and A. Al`u, "Effi- cient anomalous reflection through near-field interactions in metasurfaces," Phys. Rev. B 96, 075432 (2017). [21] A. Epstein and O. Rabinovich, "Unveiling the properties of metagratings via a detailed analytical model for syn- thesis and analysis," Phys. Rev. Appl. 8, 054037 (2017). [22] O. Rabinovich and A. Epstein, "Analytical design of printed-circuit-board (PCB) metagratings for perfect anomalous reflection," IEEE Trans. Antennas Propag. 66, 4086 -- 4095 (2018). [23] A. D´ıaz-Rubio, V. S. Asadchy, A. Elsakka, and S. A. Tretyakov, "From the generalized reflection law to the realization of perfect anomalous reflectors," Science Adv. 3, e1602714 (2017). [24] V. S. Asadchy, A. D´ıaz-Rubio, S. N. Tcvetkova, D.-H. Kwon, A. Elsakka, M. Albooyeh, and S. A. Tretyakov, "Flat engineered multichannel reflectors," Phys. Rev. X 7, 031046 (2017). [25] D. F. Sievenpiper, "Forward and backward leaky wave radiation with large effective aperture from an electron- ically tunable textured surface," IEEE Trans. Antennas Propag. 53, 236 -- 247 (2005). [26] P. M. T. Ikonen, E. Saenz, R. Gonzalo, and S. A. Tretyakov, "Modeling and analysis of composite antenna superstrates consisting on grids of loaded wires," IEEE Trans. Antennas Propag. 55, 2692 -- 2700 (2007). [27] Sergei Tretyakov, Analytical modeling in applied electro- magnetics (Artech House, 2003). [28] K. C. Gupta, R. Garg, I. Bahl, and P. Bhartia, Mi- crostrip Lines and Slotlines (Artech House, 1996). [29] (), As the observation angle deviates from broadside to- wards grazing angles, the peak power decreases following a cosine law for a uniformly illuminated aperture (with linear phase); however, the beamwidth in this case in- creases following the same law, such that the overall in- tegrated power remains the same regardless of the obser- vation angle. 15 [30] V. Popov, F. Boust, and S. N. Burokur, "Control- ling diffraction patterns with metagratings," Phys. Rev. Appl. 10, 011002 (2018). [31] (), The slightly larger deviations around f ≈ 30GHz stem from a partial blockage affecting the 2nd order FB mode measurement, coinciding in these frequencies with out ≈ −70◦, as per Fig. 6. the retro-reflection scenario θ(−2) [32] J. Coonrod, "Understanding the variables of dielec- tric constant for pcb materials used at microwave fre- quencies," in 2011 41st European Microwave Conference (2011) pp. 938 -- 944. [33] Pyotr Ya Ufimtsev, Fundamentals of the physical theory of diffraction (John Wiley & Sons, 2014). [34] Joseph B Keller, "Geometrical theory of diffraction," JOSA 52, 116 -- 130 (1962). [35] Max Born and Emil Wolf, Principles of optics: electro- magnetic theory of propagation, interference and diffrac- tion of light (Elsevier, 2013). [36] D.-H. Kwon, "Lossless scalar metasurfaces for anomalous reflection based on efficient surface field optimization," IEEE Antennas Wireless Propag. Lett. 17, 1149 -- 1152 (2018). [37] (**), It should be noted that the guided power in an infinite periodic structure is not limited by the incident power impinging it in a single period; in fact, in certain circumstances it can exceed it. The reason is that in an inifinite system, the guided power is provided by the pe- riodic boundary conditions, without drawing power from the incident fields. As in the lossless limit the incident power is reflected in its entirety from the surface, and the guided power enters the period from one side and leaves it from the other without dissipation, real power is consistently conserved over a period. In a finite surface excited by a finite beam, however, the waveguided mode evolves from the beam tail, drawing power at the first periods, then propagates without exchanging power with the incident fields (as in the infinite periodic case), and fi- nally (ideally) returns the power to the scattered fields at the other beam tail. These phenomena are demonstrated, for instance, in [15] and the associated Supplementary Material.
1901.01355	1	1901	2019-01-05T02:46:01	Investigating the Effect of Magnetic Dipole-Dipole Interaction on Magnetic Particle Spectroscopy (MPS): Implications for Magnetic Nanoparticle-based Bioassays and Magnetic Particle Imaging (MPI)	[ "physics.app-ph" ]	Superparamagnetic iron oxide nanoparticles (SPIONs), with comparable size to biomolecules (such as proteins, nucleic acids, etc.) and unique magnetic properties, good biocompatibility, low toxicity, potent catalytic behavior, are promising candidates for many biomedical applications. There is one property present in most SPION systems, yet it has not been fully exploited, which is the dipole-dipole interaction (also called dipolar interaction) between the SPIONs. It is known that the magnetic dynamics of an ensemble of SPIONs are substantially influenced by the dipolar interactions. However, the exact way it affects the performance of magnetic particle-based bioassays and magnetic particle imaging (MPI) is still an open question. The purpose of this paper is to give a partial answer to this question. This is accomplished by numerical simulations on the dipolar interactions between two nearby SPIONs and experimental measurements on an ensemble of SPIONs using our lab-based magnetic particle spectroscopy (MPS) system. Our results show that even moderate changes in the SPION concentration may have substantial effects on the magnetic dynamics of the SPION system and the harmonic signal magnitudes can be increased or decreased by 60%, depending on the values of MPS system parameters.	physics.app-ph	physics	Investigating the Effect of Magnetic Dipole-Dipole Interaction on Magnetic Particle Spectroscopy (MPS): Implications for Magnetic Nanoparticle- based Bioassays and Magnetic Particle Imaging (MPI) Kai Wu†, Diqing Su‡, Renata Saha†, Jinming Liu†, and Jian-Ping Wang†,* †Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, Minnesota 55455, USA ‡Department of Chemical Engineering and Material Science, University of Minnesota, Minneapolis, Minnesota 55455, USA Corresponding author E-mail: [email protected] (Dated: January 4, 2019) Abstract Superparamagnetic iron oxide nanoparticles (SPIONs), with comparable size to biomolecules (such as proteins, nucleic acids, etc.) and unique magnetic properties, good biocompatibility, low toxicity, potent catalytic behavior, are promising candidates for many biomedical applications. There is one property present in most SPION systems, yet it has not been fully exploited, which is the dipole-dipole interaction (also called dipolar interaction) between the SPIONs. It is known that the magnetic dynamics of an ensemble of SPIONs are substantially influenced by the dipolar interactions. However, the exact way it affects the performance of magnetic particle-based bioassays and magnetic particle imaging (MPI) is still an open question. The purpose of this paper is to give a partial answer to this question. This is accomplished by numerical simulations on the dipolar interactions between two nearby SPIONs and experimental measurements on an ensemble of SPIONs using our lab-based magnetic particle spectroscopy (MPS) system. Our results show that even moderate changes in the SPION concentration may have substantial effects on the magnetic dynamics of the SPION system and the harmonic signal magnitudes can be increased or decreased by 60%, depending on the values of MPS system parameters. Keywords: Superparamagnetic nanoparticle, dipole-dipole interaction, dipolar field, magnetic particle spectroscopy, magnetic particle imaging, biosensors. 1 1. Introduction In the past decade, superparamagnetic iron oxide nanoparticles (SPIONs) and their liquid dispersions are the subject of intense research due to their unique magnetic and physical properties and their potential applications in several fields such as hyperthermia therapy1-4, catalysis1, 5-8, drug/gene delivery9-14, magnetic bioassays15-18, magnetic particle imaging (MPI) and magnetic resonance imaging (MRI)1, 19-24. Besides their unique magnetic properties, comparable size to biomolecules, low toxicity, and good biocompatibility, the recent advances in the area of chemical engineering and material science have enabled the facile synthesis and surface functionalization of SPIONs, placing them as one of the most popular nanocomposites in many biological and biomedical applications. SPIONs show zero magnetization in the absence of magnetic fields, while an external magnetic field can magnetize the SPIONs, like a paramagnet but with larger magnetic susceptibility. Under an external magnetic field, these magnetized SPIONs exert dipolar fields (also called stray fields), which allows the detection and measurement of the SPIONs in magnetometers and susceptometers. On the other hand, the magnetic dynamics of an ensemble of SPIONs are known to be substantially influenced by the dipolar interactions25-28. This is due to the fact that, for a highly concentrated SPION suspension under an external magnetic field, the effective magnetic field on each SPION is reduced because of the dipolar fields. In this work, we report a magnetic particle spectroscopy (MPS)-based method to investigate how magnetic dipolar interactions modify the responses of an ensemble of SPIONs in the presence of the external magnetic fields. At first, this investigation is carried out via numerical simulations based on the Object Oriented MicroMagnetic Framework (OOMMF) micromagnetic software29, 30, followed by experimental measurements on different SPIONs suspensions with nanoparticle concentrations varying from 0.1133 to 3.4 nmole/mL (volume fractions varying from 0.001 to 0.029). It is seen that even moderate changes in the SPION concentration may have substantial effects on the magnetic dynamics of the system. The magnitude of harmonics can be increased or decreased by 60%, depending on the parameters of the MPS system. This work could provide insights on the design of magnetic bioassays, MPS and MPI systems where the magnitude of signals relies on the magnetic responses of SPIONs. Furthermore, it is shown that the clustering of SPIONs gives rise to a weaker magnetic response due to the presence of strong dipolar interactions. 2. Numerical Simulations 2.1 OOMMF Simulation Design Object Oriented Micormagnetic Framework (OOMMF) was employed to simulate the demagnetization fields (dipolar fields) from two nearby SPIONs separated by distances of 20 nm, 80 nm, 120 nm, 200 nm and 300 nm, respectively. The building block of the simulation is the Landau-Lifshitz-Gilbert equation: 𝑑𝑀𝑑𝑡=−𝛾𝑀×𝐻+++𝛼𝑀./𝑀×𝑑𝑀𝑑𝑡0 2 Where the saturation magnetization 𝑀. is 3.5 × 105 A/m (0.35 T), the Gilbert damping constant 𝛼 is 0.1, and the gyromagnetic ratio 𝛾 is -2.21 × 105 mA-1s-1. Other material properties used in this simulation were the exchange stiffness constant A (2.64 × 10-11 J/m, 2.64 × 10-6 erg/cm) and the crystalline anisotropy constant (1.25 × 104 J/m3, 1.25 × 105 ergs/cm3). The cell size for the simulation was 2 nm. 2.2 Simulation Results As shown in Figure 1 (a), when an external magnetic field is applied along the +x direction, the magnetizations of both SPIONs lie along the field direction. The demagnetization field at the center of the SPION is around 2.2 × 105 A/m. As the distance between the SPIONs increases, the overlap area between the demagnetization fields decreases (see Figures 1(b) - (f)). It was observed that the demagnetization fields from the SPIONs overlap at dipole-dipole distances below 120 nm, which indicates the strong dipolar interactions. When the spacing between SPIONs is larger than 200 nm, the demagnetization field from one SPION has negligible influence on the other, indicating much weaker or none dipolar interactions. Figure 1. OOMMF simulation results on the dipolar interactions between two nearby SPIONs. (a) Two SPIONS with magnetizations lying along the external magnetic field direction (+x direction). (b)-(f) Demagnetization fields from two SPIONs that are separated by distances of 20 nm (b), 80 nm (c), 120 nm (d), 200 nm (e), and 300 nm (f). The demagnetization field unit is A/m in this figure. 3. Materials and Methods 3.1 Materials In this work, we used superparamagnetic iron oxide nanoparticles (SPIONs) with an average magnetic core diameter of 30 nm (purchased from Ocean NanoTech, catalog#SHB-30). These single-core SPIONs are lipid 3 coated iron oxide (g-Fe2O3/Fe3O4) with biotin. The hydrodynamic size is about 20 nm larger than the magnetic core size measured by transmission electron microscopy (TEM, FEI Tecnai T12, 120 kV) and dynamic light scattering (DLS), see Supporting Information S1. The SHB-30 can be reconstituted in phosphate buffered saline (PBS). 3.2 Sample Preparation SPION suspensions having nanoparticle volume fractions (VFs, f) ranging from 0.001 to over 0.029 are experimentally studied in this paper. The SHB-30 specimen originally bought from Ocean NanoTech has a SPION concentration of 0.068 nmole/mL (volume fraction f=0.000579), which is below the volume fraction range of interest in this work. At first, a 250 µL SHB-30 suspension was ultra-centrifuged at 11,000 RPM, acceleration 11,200 g for 45 min (PowerSpinTM BX Centrifuge). Then a 170 µL supernatant was removed, followed by 30-mins of water bath ultra-sonication (Branson 1510 Ultrasonic Cleaner) to make SPIONs evenly dispersed, resulting in the sample i as shown in Figure 2. Sample i contains 80 µL SPION suspension with a nanoparticle concentration of 0.2125 nmole/mL, volume fraction f=0.0018 and an averaged inter-particle distance of 198 nm. MPS measurements were carried out on sample i. To get sample ii from sample i, ultra- centrifugation, supernatant removal, ultra-sonication processes were repeated (the condensation process (a) labeled in Figure 2). Sample ii was a 20 µL SPION suspension with a nanoparticle concentration of 0.85 nmole/mL, volume fraction f=0.0072, and an averaged inter-particle distance of 125 nm. Then MPS measurements were carried out on sample ii. To get sample iii from sample ii, a dilution process (b) as labeled in Figure 2 was performed, which includes addition of a certain volume of PBS buffer to SPION suspension to dilute sample to a lower nanoparticle concentration (as well as a lower volume fraction f and a larger averaged inter- particle distance), followed by 30-mins of water bath ultra-sonication. Samples iv - vii were prepared in the similar manner and each sample was measured by our lab-based MPS system. Details on the volume fractions and inter- particle distances of specimens i - vii can be found in Supporting Information S2. 4 Figure 2. Samples: (i) 80 µL SPION suspension with volume fraction f=0.0018; (ii) 20 µL SPION suspension with volume fraction f=0.0072; (iii) 100 µL SPION suspension with volume fraction f=0.0014; (iv) 5 µL SPION suspension with volume fraction f=0.029; (v) Dried SPION powder with volume fraction f>0.029; (vi) 120 µL SPION suspension with volume fraction f=0.0012; (vii) 150 µL SPION suspension with volume fraction f=0.001. Process (a) represents the condensation of SPION suspension: at first, the suspension is ultra-centrifuged at 11,000 RPM, acceleration 11,200 g for 45 min, then removal of a certain volume of supernatant, followed by 30-mins of water bath ultra-sonication to make SPIONs evenly dispersed. Process (b) represents the dilution of SPION suspension: initially, the suspension is added with a certain volume of PBS buffer to dilute sample to a lower nanoparticle concentration (as well as a lower volume fraction f and a larger averaged inter-particle distance), followed by 30-mins of water bath ultra-sonication. 3.3 MPS Experimental Setups In this work, a lab-based magnetic particle spectrometer (MPS) was used to monitor the magnetic responses of SPIONs with different particle concentrations c (varying from 0.1133 nmole/mL to 3.4 nmole/mL), volume fractions f (varying from 0.001 to over 0.029), and inter-particle distances d (varying from below 78 nm to 245 nm). The MPS system setups have been reported by our previous works31-36. This system consists of: a PC with 5 LabVIEW program to control the digital acquisition card (DAQ), carry out analog to digital convertor (ADC) and discrete Fourier Transform (DFT) on the analog signals that were sent back from the pick-up coils, two instrument amplifiers (IAs) receive commands from LabVIEW and send sinusoidal waves to drive coils, two sets of drive coils to generate high and low frequency alternating magnetic fields, one pair of differentially wound pick-up coils to collect the magnetic responses from SPIONs (Faraday's law of Induction), and a plastic vial to hold SPION sample. The schematic drawings of MPS system can be found in Supporting Information S3. 3.4 Experimental Methods Two sinusoidal magnetic fields, one with high frequency 𝑓2 (in this work we vary 𝑓2 from 400 Hz to 20 kHz) but low amplitude 𝐴2=17 𝑂𝑒, the other with low frequency 𝑓9=10 𝐻𝑧 but high amplitude 𝐴9=170 𝑂𝑒 are can be found in Supporting Information S3). The 3rd and the 5th harmonics at combinational frequencies 𝑓2±2𝑓9 and 𝑓2±4𝑓9 were used as indicators of the physical properties of SPION suspensions. For each sample, we carried out MPS measurements at different high frequencies 𝑓2 (400 Hz to 20 kHz). At each frequency 𝑓2, three applied to different SPION suspensions (samples i - vii)35, 37-41. Under the time-varying magnetic fields, the nonlinear magnetic responses of SPIONs induce electromotive force (EMF) in the pick-up coils (Faraday's Law of Induction), which is then sent back to DAQ and PC for harmonic signal extraction (details on the signal chain consequent measurements were performed. In each measurement, the background noise floor was monitored for 10 s (10 data points). Then, the SPION sample in vial was inserted into the MPS system and the total signal was collected for another 10 s (10 data points). The voltage signals due to the nonlinear magnetic responses of SPIONs can be extracted by subtracting the background noise from the total signal according to the phasor theory34-36 (details on the phasor theory can be found in Supporting Information S4). 4. MPS Measurement Results and Discussions The magnetic responses of SPIONs are largely dependent on their magnetic core size, anisotropy parameter, hydrodynamic size36, 42, 43, viscosity32, 44 and temperature45, 46 of suspension, as well as the effective magnetic field2, 35 sensed by each SPION. In this work, we used the same batch of SPIONs, which could effectively minimize the variances in the magnetic core size, anisotropy parameter and hydrodynamic size from each measurement (see Supporting Information S5). Furthermore, the temperature and viscosity of these SPION suspensions were set identical so that the only factor that affects the magnetic responses would be the effective magnetic field sensed by each SPION (see Equations (6) & (13) from Supporting Information S6). The mathematical models of the magnetic responses of SPIONs and the induced voltage model from pick-up coils can be found in Supporting Information S6, and the mathematical models of the harmonic signals can be found in Supporting Information S7. 6 Figure 3. Measured (a) 3rd and (b) 5th harmonic amplitudes from samples i - vii as 𝑓2 varies from 400 Hz to 20 kHz. Error bar represents the standard deviation. These seven SPION suspension samples come from the same batch and same amount of SPIONs but with different concentrations. Their magnetic responses can be divided into three groups: group 1, no dipolar interactions in samples i, iii, vi, and vii; group 2, moderate dipolar interactions in samples ii and iv; group 3, strong dipolar interactions in sample v. frequencies 𝑓2 . By increasing SPION concentrations from 0.1133 nmole/mL (sample vii) to 3.4 nmole/mL As is shown in Figure 3, the averaged 3rd and 5th harmonic amplitudes are summarized as a function of the high (sample iv), the inter-particle distances decrease from 245 nm (sample vii) to 78 nm (sample iv). The dipolar magnetic fields (dipole-dipole interactions) generated by the nearby SPIONs increase with decreasing inter- particle distances, which, as a result, reduces the effective magnetic field on each SPION. A higher concentration of SPIONs in suspension leads to stronger dipolar interactions and lower harmonic amplitudes. In Figure 3, the magnetic responses of these SPION samples can be divided into 3 groups: 1) no dipolar interactions in samples i, iii, vi, and vii; 2) moderate dipolar interactions in samples ii and iv; 3) strong dipolar interactions in sample v. In group 1 (no dipolar interactions), the magnetic responses are identical from each SPION suspension even with the dilution of nanoparticles. In group 2 (with moderate dipolar interactions), the magnetic responses from sample ii (volume fraction f=0.0072, averaged inter-particle distance d=125 nm) and sample iv (volume fraction f=0.029, averaged inter-particle distance d=78 nm) are identical in the low 𝑓2 range (below 7 kHz), and in the high 𝑓2 The harmonic signals from samples i to vii at four 𝑓2 frequencies: 400 Hz, 3 kHz, 5 kHz, and 20 kHz are range, sample iv shows smaller harmonic amplitudes due to the stronger dipolar interactions than sample ii. In group 3 (strong dipolar interactions), the densely stacked SPIONs dried powder leads to a stronger dipolar interaction than any other samples in this work. Consequently, weaker magnetic responses are detected in MPS system. summarized in Figure 4. The group 1 samples (no dipolar interactions) are at similar amplitude levels at all 7 frequencies. The group 2 and 3 samples (with moderate dipolar interactions and strong dipolar interactions) are at similar amplitude level in low frequency region (see Figure 4(a)). However, significant differences in the magnetic responses (i.e., the harmonic amplitudes) are found as we increase 𝑓2 to 20 kHz (see Figure 4(c)). The effect of dipolar interaction reaches its maximum at 𝑓2=3 𝑘𝐻𝑧 when a harmonic magnitude difference of up to 60% is measured from the same batch of SPIONs with different volume fractions (different nanoparticle concentrations and inter-particle distances). Figure 4. Measured 3rd harmonic amplitudes from samples i-vii at driving field frequencies of (a) 𝑓2=400 𝐻𝑧, (b) 𝑓2=3 𝑘𝐻𝑧, (c) 𝑓2=5 𝑘𝐻𝑧, and (d) 𝑓2=20 𝑘𝐻𝑧. Error bar represents the standard deviation. 8 5. Conclusions In this paper, we reported a MPS-based method to experimentally investigate the effect of dipolar interactions on the magnetic dynamics of an ensemble of SPIONs. We measured the magnetic responses of seven SPION suspension samples that come from the same batch of nanoparticles with varying volume fractions f from 0.001 to above 0.029 (inter-particle distances drop from 245 nm to below 78 nm, nanoparticle concentrations increase from 0.1133 to 3.4 nmole/mL). Our results show that even moderate changes in the SPION concentration could cause substantial effects on the magnetic dynamics of the SPION system, being capable of increasing or decreasing the harmonic signal magnitudes by 60%, depending on the values of MPS system parameters. Furthermore, our simulation results agree well with the MPS experimental results, where the strong dipolar interactions were observed at dipole-dipole distances below 120 nm whereas the dipolar field from one SPION has negligible influence on the other when the spacings between two SPIONs are larger than 200 nm, indicating much weaker or none dipolar interactions. Our OOMMF simulation results prove the feasibility of using the MPS system as a promising candidate for characterizing the dipolar interactions in any SPION systems. These results give new insights on the design of magnetometers such as magnetic biosensors and MPI devices where the cumulative magnetic responses of an ensemble of SPIONs are of great importance in exerting magnetic signals for diagnosing and image reconstruction purposes. For example, in magnetometer-based bioassays (i.e., Superconducting Quantum Interference Devices (SQUID), Giant Magnetoresistive (GMR), etc.) and relaxometry-based bioassays (Magnetic Resonance Imaging (MRI), Nuclear Magnetic Resonance (NMR), etc.)47- 50, the assay results could largely be biased due to the fact that the magnetic characteristics of the SPION clusters differ from the non-interacting SPION system. In MPI where this non-invasive tomographic technique relies on the direct detection of SPION tracers' magnetic signals, however, the clustering of SPIONs at target tissues could lead to a loss of up to 60% in magnetic signals. On the other hand, the method reported in this work provides a possible way of using lab-based MPS for quantifying the degree of dipolar interactions in SPION systems as well as distinguishing the interacting and non-interacting SPION systems. Acknowledgements Portions of this work were conducted in the Minnesota Nano Center, which is supported by the National Science Foundation through the National Nano Coordinated Infrastructure Network (NNCI) under Award Number ECCS- 1542202. Portions of this work were carried out in the Characterization Facility, University of Minnesota, a member of the NSF-funded Materials Research Facilities Network (www.mrfn.org) via the MRSEC program. Supporting Information Supporting Information S1. TEM and DLS characterization of SPIONs from SHB-30 specimen Supporting Information S2. SPION volume fraction and the averaged inter-particle distances Supporting Information S3. The lab-based MPS system and the signal chain 9 Wu, K.; Wang, J.-P., Magnetic hyperthermia performance of magnetite nanoparticle assemblies under Biehl, P.; von der Luhe, M.; Dutz, S.; Schacher, F. H., Synthesis, Characterization, and Applications of Supporting Information S4. The phasor theory Supporting Information S5. Using same batch of SPIONs for MPS measurements Supporting Information S6. Mathematical models of the magnetic responses of SPIONs and the induced voltage from pick-up coils Supporting Information S7. Mathematical models of the harmonic signals from SPIONs AUTHOR INFORMATION Corresponding Author E-mail: [email protected] (J.-P. W) Notes The authors declare no competing financial interest. References 1. Magnetic Nanoparticles Featuring Polyzwitterionic Coatings. Polymers 2018, 10 (1), 28. 2. different driving fields. AIP Advances 2017, 7 (5), 056327. 3. Coral, D. F.; Mendoza Zélis, P.; Marciello, M.; Morales, M. D. P.; Craievich, A.; Sanchez, F. H.; Fernández van Raap, M. B., On the effect of nanoclustering and dipolar interactions in heat generation for magnetic hyperthermia. Langmuir 2016. 4. Coral, D. F.; Mendoza Zélis, P.; Marciello, M.; Morales, M. a. d. P.; Craievich, A.; Sánchez, F. H.; Fernández van Raap, M. B., Effect of nanoclustering and dipolar interactions in heat generation for magnetic hyperthermia. Langmuir 2016, 32 (5), 1201-1213. Hu, Y.; Ma, W.; Tao, M.; Zhang, X.; Wang, X.; Wang, X.; Chen, L., Decorated‐magnetic‐nanoparticle‐ 5. supported bromine as a recyclable catalyst for the oxidation of sulfides. Journal of Applied Polymer Science 2018, 135 (13). 6. nanoparticle separation and catalyst recycling. Environmental Science: Nano 2018. Moghaddam, F. M.; Saberi, V.; Kalhor, S.; Veisi, N., Palladium (II) Immobilized Onto the Glucose 7. Functionalized Magnetic Nanoparticle as a New and Efficient Catalyst for the One‐pot Synthesis of Benzoxazoles. Applied Organometallic Chemistry 2018. 8. Zhou, Q.; Wan, Z.; Yuan, X.; Luo, J., A new magnetic nanoparticle‐supported Schiff base complex of manganese: an efficient and recyclable catalyst for selective oxidation of alcohols. Applied Organometallic Chemistry 2016, 30 (4), 215-220. Leshuk, T.; Holmes, A.; Ranatunga, D.; Chen, P. Z.; Jiang, Y.; Gu, F., Magnetic flocculation for 10 9. Yao, X.; Niu, X.; Ma, K.; Huang, P.; Grothe, J.; Kaskel, S.; Zhu, Y., Graphene Quantum Dots‐Capped Magnetic Mesoporous Silica Nanoparticles as a Multifunctional Platform for Controlled Drug Delivery, Magnetic Hyperthermia, and Photothermal Therapy. Small 2017, 13 (2). 10. Chowdhuri, A. R.; Bhattacharya, D.; Sahu, S. K., Magnetic nanoscale metal organic frameworks for potential targeted anticancer drug delivery, imaging and as an MRI contrast agent. Dalton Transactions 2016, 45 (7), 2963-2973. 11. Hervault, A.; Dunn, A. E.; Lim, M.; Boyer, C.; Mott, D.; Maenosono, S.; Thanh, N. T., Doxorubicin loaded dual pH-and thermo-responsive magnetic nanocarrier for combined magnetic hyperthermia and targeted controlled drug delivery applications. Nanoscale 2016, 8 (24), 12152-12161. 12. Wang, G.; Ma, Y.; Wei, Z.; Qi, M., Development of multifunctional cobalt ferrite/graphene oxide nanocomposites for magnetic resonance imaging and controlled drug delivery. Chemical Engineering Journal 2016, 289, 150-160. 13. Huang, J.; Li, Y.; Orza, A.; Lu, Q.; Guo, P.; Wang, L.; Yang, L.; Mao, H., Magnetic Nanoparticle Facilitated Drug Delivery for Cancer Therapy with Targeted and Image‐Guided Approaches. Advanced functional materials 2016, 26 (22), 3818-3836. 14. Estelrich, J.; Escribano, E.; Queralt, J.; Busquets, M. A., Iron oxide nanoparticles for magnetically-guided and magnetically-responsive drug delivery. International journal of molecular sciences 2015, 16 (4), 8070-8101. 15. Krishna, V. D.; Wu, K.; Perez, A. M.; Wang, J. P., Giant Magnetoresistance-based Biosensor for Detection of Influenza A Virus. Frontiers in Microbiology 2016, 7, 8. 16. Wang, Y.; Wang, W.; Yu, L. N.; Tu, L.; Feng, Y. L.; Klein, T.; Wang, J. P., Giant magnetoresistive- based biosensing probe station system for multiplex protein assays. Biosensors & Bioelectronics 2015, 70, 61-68. 17. Rizzi, G.; Lee, J. R.; Dahl, C.; Guldberg, P.; Dufva, M.; Wang, S. X.; Hansen, M. F., Simultaneous Profiling of DNA Mutation and Methylation by Melting Analysis Using Magnetoresistive Biosensor Array. Acs Nano 2017, 11 (9), 8864-8870. 18. GMR biosensor platform with smartphone interface. Biosensors and Bioelectronics 2016, 85, 1-7. 19. Wells, J.; Paysen, H.; Kosch, O.; Trahms, L.; Wiekhorst, F., Temperature dependence in magnetic particle imaging. AIP Advances 2018, 8 (5), 056703. 20. In Iron Oxide Nanoparticles for Biomedical Applications, Elsevier: 2018; pp 115-133. 21. Ludewig, P.; Gdaniec, N.; Sedlacik, J.; Forkert, N. D.; Szwargulski, P.; Graeser, M.; Adam, G.; Kaul, M. G.; Krishnan, K. M.; Ferguson, R. M. J. A. n., Magnetic particle imaging for real-time perfusion imaging in acute stroke. 2017, 11 (10), 10480-10488. Choi, J.; Gani, A. W.; Bechstein, D. J.; Lee, J.-R.; Utz, P. J.; Wang, S. X., Portable, one-step, and rapid Zanganeh, S.; Aieneravaie, M.; Erfanzadeh, M.; Ho, J.; Spitler, R., Magnetic Particle Imaging (MPI). 11 Kechrakos, D.; Trohidou, K., Competition between dipolar and exchange interparticle interactions in Landi, G. T., Role of dipolar interaction in magnetic hyperthermia. Physical Review B 2014, 89 (1), García-Otero, J.; Porto, M.; Rivas, J.; Bunde, A., Influence of dipolar interaction on magnetic properties Shen, Z.; Wu, A.; Chen, X., Iron oxide nanoparticle based contrast agents for magnetic resonance imaging. 22. Qiao, Y.; Gumin, J.; MacLellan, C. J.; Gao, F.; Bouchard, R.; Lang, F. F.; Stafford, R. J.; Melancon, M. P., Magnetic resonance and photoacoustic imaging of brain tumor mediated by mesenchymal stem cell labeled with multifunctional nanoparticle introduced via carotid artery injection. Nanotechnology 2018, 29 (16), 165101. 23. Ravanshad, R.; Zadeh, A. K.; Amani, A. M.; Mousavi, S. M.; Hashemi, S. A.; Dashtaki, A. S.; Mirzaei, E.; Zare, B., Application of nanoparticles in cancer detection by Raman scattering based techniques. Nano Reviews & Experiments 2017, 9. 24. Molecular pharmaceutics 2016, 14 (5), 1352-1364. 25. Branquinho, L. C.; Carrião, M. S.; Costa, A. S.; Zufelato, N.; Sousa, M. H.; Miotto, R.; Ivkov, R.; Bakuzis, A. F., Effect of magnetic dipolar interactions on nanoparticle heating efficiency: Implications for cancer hyperthermia. Scientific reports 2013, 3, 2887. 26. magnetic nanoparticle films. Journal of magnetism and magnetic materials 2003, 262 (1), 107-110. 27. 014403. 28. of ultrafine ferromagnetic particles. Physical review letters 2000, 84 (1), 167. 29. 30. of Standards and Technology, Gaithersburg, MD). NIST J. Res. 114, 57-67. 31. Wu, K.; Yu, L.; Zheng, X.; Wang, Y.; Feng, Y.; Tu, L.; Wang, J.-P. In Viscosity effect on the brownian relaxation based detection for immunoassay applications, 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE: 2014; pp 2769-2772. 32. Wu, K.; Liu, J.; Wang, Y.; Ye, C.; Feng, Y.; Wang, J.-P., Superparamagnetic nanoparticle-based viscosity test. Applied Physics Letters 2015, 107 (5), 053701. 33. Wu, K.; Batra, A.; Jain, S.; Wang, J.-P., Magnetization Response Spectroscopy of Superparamagnetic Nanoparticles Under Mixing Frequency Fields. IEEE Transactions on magnetics 2016, 52 (7). 34. method. Journal of Physics D: Applied Physics 2014, 47 (15), 155001. 35. Wu, K.; Tu, L.; Su, D.; Wang, J.-P., Magnetic dynamics of ferrofluids: mathematical models and experimental investigations. Journal of Physics D: Applied Physics 2017, 50 (8), 085005. 36. Wu, K.; Schliep, K.; Zhang, X.; Liu, J.; Ma, B.; Wang, J. P., Characterizing Physical Properties of Superparamagnetic Nanoparticles in Liquid Phase Using Brownian Relaxation. Small 2017, 13 (22). Donahue, M. J. OOMMF user's guide, version 1.0; 1999. Donahue, M.; Porter, D.; Lau, J.; McMichael, R., in Interagency Report NISTIR 6376 (National Institute Tu, L.; Wu, K.; Klein, T.; Wang, J.-P., Magnetic nanoparticles colourization by a mixing-frequency 12 Nikitin, P. I.; Vetoshko, P. M.; Ksenevich, T. I., New type of biosensor based on magnetic nanoparticle Tu, L.; Jing, Y.; Li, Y.; Wang, J.-P., Real-time measurement of Brownian relaxation of magnetic Rauwerdink, A. M.; Weaver, J. B., Measurement of molecular binding using the Brownian motion of 37. Krause, H.-J.; Wolters, N.; Zhang, Y.; Offenhäusser, A.; Miethe, P.; Meyer, M. H.; Hartmann, M.; Keusgen, M., Magnetic particle detection by frequency mixing for immunoassay applications. Journal of magnetism and magnetic materials 2007, 311 (1), 436-444. 38. detection. Journal of Magnetism and Magnetic Materials 2007, 311 (1), 445-449. 39. nanoparticles by a mixing-frequency method. Applied Physics Letters 2011, 98 (21), 213702. 40. Tu, L.; Feng, Y.; Klein, T.; Wang, W.; Wang, J.-P., Measurement of Brownian Relaxation of Magnetic Nanoparticle by a Multi-Tone Mixing-Frequency Method. Magnetics, IEEE Transactions on 2012, 48 (11), 3513- 3516. 41. Tu, L.; Klein, T.; Wang, W.; Feng, Y.; Wang, Y.; Wang, J.-P., Measurement of Brownian and Néel relaxation of magnetic nanoparticles by a mixing-frequency method. Magnetics, IEEE Transactions on 2013, 49 (1), 227-230. 42. Zhang, X.; Reeves, D. B.; Perreard, I. M.; Kett, W. C.; Griswold, K. E.; Gimi, B.; Weaver, J. B., Molecular sensing with magnetic nanoparticles using magnetic spectroscopy of nanoparticle Brownian motion. Biosensors and Bioelectronics 2013, 50, 441-446. 43. magnetic nanoparticle probes. Applied Physics Letters 2010, 96 (3), 033702. 44. Weaver, J. B.; Harding, M.; Rauwerdink, A. M.; Hansen, E. W. In The effect of viscosity on the phase of the nanoparticle magnetization induced by a harmonic applied field, SPIE Medical Imaging, International Society for Optics and Photonics: 2010; pp 762627-762627-8. 45. and dc magnetic fields. Physics in medicine and biology 2009, 54 (19), L51. 46. Weaver, J. B.; Rauwerdink, A. M.; Hansen, E. W., Magnetic nanoparticle temperature estimation. Medical physics 2009, 36 (5), 1822-1829. 47. Eberbeck, D.; Wiekhorst, F.; Steinhoff, U.; Trahms, L., Aggregation behaviour of magnetic nanoparticle suspensions investigated by magnetorelaxometry. Journal of Physics: Condensed Matter 2006, 18 (38), S2829. 48. Eberbeck, D.; Trahms, L., Experimental investigation of dipolar interaction in suspensions of magnetic nanoparticles. Journal of Magnetism and Magnetic Materials 2011, 323 (10), 1228-1232. 49. of arrays of interacting Co nanocrystals. Journal of magnetism and magnetic materials 2002, 240 (1-3), 40-43. 50. Yang, Z.; Shi, F.; Wang, P.; Raatz, N.; Li, R.; Qin, X.; Meijer, J.; Duan, C.; Ju, C.; Kong, X., Detection of magnetic dipolar coupling of water molecules at the nanoscale using quantum magnetometry. Physical Review B 2018, 97 (20), 205438. Spasova, M.; Wiedwald, U.; Ramchal, R.; Farle, M.; Hilgendorff, M.; Giersig, M., Magnetic properties Rauwerdink, A. M.; Hansen, E. W.; Weaver, J. B., Nanoparticle temperature estimation in combined ac 13 Supporting Information Investigating the Effect of Magnetic Dipole-Dipole Interaction on Magnetic Particle Spectroscopy (MPS): Implications for Magnetic Nanoparticle- based Bioassays and Magnetic Particle Imaging (MPI) Kai Wu†, Diqing Su‡, Renata Saha†, Jinming Liu†, and Jian-Ping Wang†, †Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, Minnesota 55455, USA ‡Department of Chemical Engineering and Material Science, University of Minnesota, Minneapolis, Minnesota 55455, USA Corresponding author E-mail: [email protected] 1 Supporting Information S1. TEM and DLS characterization of SPIONs from SHB-30 specimen. Figure S1. (a) TEM image of the SPIONs from SHB-30 specimen. (b) Statistical hydrodynamic size distribution collected using DLS. Solid black line is cumulative distribution curve and dash blue line is the lognormal curve fitting. SPIONs show an average hydrodynamic size of 50.1 nm. 2 Supporting Information S2. SPION volume fraction and the averaged inter-particle distances. Table S1. Nanoparticle volume fractions and the averaged inter-particle distances Volume SPION Concentration c Volume Fraction Sample Label i ii iii iv v vi vii 80 µL 20 µL 100 µL 5 µL Dried powder 120 µL 150 µL (nmole/mL) 0.2125 0.85 0.17 3.4 NA (f)** 0.0018 0.0072 0.0014 0.029 > 0.029 Averaged Inter- particle Distance d (nm) 198 125 214 78 < 78 0.1417 0.1133 0.0012 0.001 227 245 Sample labels correspond to the marks in Figure 2 from manuscript. Volume fraction (VF) is calculated by 𝜙=∑ $%%$& , where 𝑉( is the magnetic core volume of SPION 𝑘 and 𝑉 is the volume of suspension. 3 Supporting Information S3. The lab-based MPS system and the signal chain. As shown in Figure S2, the MPS system consists of a PC with LabVIEW program to control the generation of two sinusoidal waves that are sent to DAQ and instrumental amplifiers (IAs), analog to digital convertor (ADC) and discrete Fourier Transform (DFT) to analyze the collected signal from pick-up coils. Two sinusoidal signals are sent to drive coils to generate two sinusoidal magnetic fields, one with high frequency but low amplitude 𝐴,sin(2𝜋𝑓,𝑡+𝜑,) and the other with low frequency but high amplitude 𝐴8sin(2𝜋𝑓8𝑡+𝜑8), where 𝐴, and 𝐴8 are the amplitudes of high and low frequency magnetic fields, 𝑓, and 𝑓8 are the frequencies of high and low frequency magnetic fields, 𝜑, and 𝜑8 are the phases of high and low frequency magnetic fields, and 𝑡 is time parameter. One pair of differentially wound pick-up coils is put in the center of two outer coils, and the plastic vial with SPION suspension inside will be inserted into the pick-up coils. The specially designed plastic vial can be filled up to 300 µL liquid. As the time-varying sinusoidal magnetic fields applied to the SPIONs, the nonlinear magnetic responses of SPIONs generate time-varying magnetic fields that can be sensed by the pick-up coils according to the Faraday's law of Induction. The real-time voltage signal from pick-up coils is sent back to DAQ and LabVIEW after a bandpass filter (BPF). The harmonic signals at combinational frequencies 𝑓,±2𝑓8 (the 3rd harmonics) and 𝑓,±4𝑓8 (the 5th harmonics) are extracted after DFT for analysis. Figure S2. Schematic view of MPS system. 4 Supporting Information S4. The phasor theory. The voltage caused by SPIONs at a certain frequency is represented by a phasor: 𝐴∙𝑒=(>?@A) (or expressed as 𝐴∠𝜑), where 𝜔 is the angular frequency of driving field, 𝜑 is the phase angle, and 𝑗=√−1. background noise can be expressed as 𝐴HIJK𝑒=ALMN&O . Secondly, a plastic vial containing SPION sample is inserted into the MPS system and the total signal is collected. The total signal is expressed as 𝐴PQP𝑒=ARSR. This total signal is the sum of two phasors: the background noise and the SPIONs (namely, 𝐴THU𝑒=AVLW). In our experimental setup, two sinusoidal magnetic fields are applied by passing alternating currents (ACs) to the driving coils. Firstly, the background noise is collected with the external magnetic fields on. The So, 𝐴HIJK𝑒=ALMN&O+𝐴THU𝑒=AVLW=𝐴PQP𝑒=ARSR, (1) X𝐴HIJK×𝑐𝑜𝑠𝜑HIJK+𝐴THU×𝑐𝑜𝑠𝜑THU=𝐴PQP×𝑐𝑜𝑠𝜑PQP 𝐴HIJK×𝑠𝑖𝑛𝜑HIJK+𝐴THU×𝑠𝑖𝑛𝜑THU=𝐴PQP×𝑠𝑖𝑛𝜑PQP (2) which reduces to an equation set: By solving the equation set above, we can get the harmonic amplitude 𝐴THU and phase 𝜑THU of SPIONs at different frequencies. 5 Supporting Information S5. Using same batch of SPIONs for MPS measurements. To minimize the variances in the magnetic core size, anisotropy parameter, and hydrodynamic size from different batches of SPIONs, the same batch of SPIONs are used for all the MPS measurements by repeating the dilution and condensation processes. The cycling steps can be found in Figure 2 from manuscript. We started with sample i (80 µL in volume, with nanoparticle concentration c=0.2125 nmole/mL, volume fraction f=0.0018, and averaged inter-particle distance d=198 nm), and sample iii is achieved after one condensation process and one dilution process on sample i. Sample iii (100 µL in volume, with nanoparticle concentration c=0.17 nmole/mL, volume fraction f=0.0014, and averaged inter-particle distance d=214 nm) consists of the same batch of SPIONs from sample i and shows the similar magnetic responses with no dipolar interactions. After several condensation and dilutions process, we can get the similar magnetic responses from SPIONs under the conditions of no dipolar interactions (samples i, iii, vi, and vii) or with dipolar interactions (samples ii, iv, and v) as shown in Figure S3. Figure S3. The 3rd harmonic amplitudes from SPION samples i -- vii measured at 𝑓,=400 𝐻𝑧. 6 Supporting Information S6. Mathematical models of the magnetic responses of SPIONs and the induced voltage from pick-up coils. In the presence of sinusoidal magnetic fields, SPIONs are magnetized and their magnetic moments tend to align with the external magnetic fields. For a SPION suspension of monodispersed, noninteracting SPIONs, the magnetic response can be expressed as the Langevin function: 𝑀d(𝑡)=𝜇𝑐𝐿(𝜉), (3) 𝐿(𝜉)=coth𝜉−mn (4) 𝜉=o,(?) (pP (5) where, expressed as: On the other hand, for the densely distributed SPIONs with interparticle interactions, the effective magnetic Assume SPIONs are spherical without interparticle interactions (i.e., dipolar interactions). The magnetic The SPIONs are characterized by magnetic core diameter 𝐷, saturation magnetization 𝑀* and concentration 𝑐. moment of each nanoparticle is 𝜇=𝑀𝑉r=𝑀𝜋𝐷s 6⁄ , 𝑉r is volume of the magnetic core, 𝜉 is the ratio of magnetic energy over thermal energy, 𝑘v is Boltzmann constant, and 𝑇 is the absolute temperature in Kelvin. 𝐻(𝑡)=𝐴,𝑠𝑖𝑛(2𝜋𝑓,𝑡+𝜑,)+𝐴8𝑠𝑖𝑛(2𝜋𝑓8𝑡+𝜑8) are the external magnetic fields. field applied to each SPION is the sum of externally applied magnetic fields 𝐻xx⃗(𝑡) and the cumulative dipolar magnetic fields, 𝐻xx⃗(𝑡)z{JI}K, also called stray field or demagnetizing field, that exerted by nearby SPIONs, 𝐻xx⃗(𝑡)zK~~J =𝐻xx⃗(𝑡)+𝐻xx⃗(𝑡)z{JI}K=[𝐴,𝑠𝑖𝑛(2𝜋𝑓,𝑡+𝜑,)+𝐴8𝑠𝑖𝑛(2𝜋𝑓8𝑡+𝜑8)]𝑧+∑𝐻xx⃗(𝑡)z{JI} = 𝐻xx⃗(𝑡)z{JI} (cid:130)𝑟⃗J,=(cid:132)= m(cid:133)(cid:134)(cid:135)s(cid:136)⃗N,(cid:137)∙(cid:130)oxx⃗(cid:137)∙(cid:136)⃗N,(cid:137)(cid:132) z(cid:136)⃗N,(cid:137)z(cid:138) − oxx⃗(cid:137) z(cid:136)⃗N,(cid:137)z(cid:139)(cid:140) (7) = is the effective magnetic field on the 𝑖th SPION, ∑𝐻xx⃗(𝑡)z{JI} Where 𝐻(𝑡)K~~J = = field generated by the nearby SPIONs, 𝐻xx⃗(𝑡)z{JI} (cid:130)𝑟⃗J,=(cid:132) is the dipolar field of the 𝑗th SPION on the 𝑖th SPION, 𝑟⃗J,= = is the length of the vector connecting SPIONs 𝑖 and 𝑗, and 𝜇⃗= is the magnetic moment vector of the 𝑗th SPION, 𝑧 The energy of SPION 𝑖 consists of the anisotropy energy, Zeeman energy, and dipolar interaction energy 1. By defining 𝜇J as the unit vector along the direction of magnetic moment 𝜇⃗J of SPION 𝑖, 𝑛(cid:142)J as the unit vector along the easy axis, 𝑧 as the unit vector along the externally applied magnetic field, 𝑟⃗J,= as vector connecting magnetic moments 𝜇J and 𝜇= 2, 𝐻xx⃗(𝑡)zK~~J sensed by particle 𝑖. as the sum of external magnetic field and dipolar fields from nearby particles = (6) is the cumulative dipolar magnetic as the unit vector along the externally applied magnetic field. 7 Anisotropy energy 𝐸(cid:144)(J) is: 𝐸(cid:144)(J)=−𝐾K~~𝑉r(𝜇J∙𝑛(cid:142)J)(cid:146) (8) Zeeman energy 𝐸(cid:147)(J) is: 𝐸(cid:147)(J)=−𝜇⃗J∙𝜇(cid:148)𝐻xx⃗(𝑡) (9) Dipolar interaction energy 𝐸d(J,=) is: 𝐸d(J,=)=−𝜇⃗J∙𝜇(cid:148)𝐻xx⃗(𝑡)z{JI} (cid:130)𝑟⃗J,=(cid:132)=o(cid:149)(cid:133)(cid:134)(cid:135)oxx⃗N ∙ oxx⃗(cid:137) z(cid:136)⃗N,(cid:137)z(cid:139)−s(oxx⃗N ∙ (cid:136)⃗N,(cid:137))(oxx⃗(cid:137) ∙ (cid:136)⃗N,(cid:137)) (cid:140) (10) = z(cid:136)⃗N,(cid:137)z(cid:138) By taking dipolar interaction into consideration, the total energy of particle 𝑖 is: 𝐸(J)=𝐸(cid:144)(J)+𝐸(cid:147)(J)+𝐸d(J,=) (11) 𝐸(J)=−𝐾K~~𝑉r(𝜇J∙𝑛(cid:142)J)(cid:146)−𝜇⃗J∙𝜇(cid:148)𝐻xx⃗(𝑡)zK~~J 𝜇(cid:148)𝐻xx⃗(𝑡)zK~~J =𝜇(cid:148)𝐻xx⃗(𝑡)－ o(cid:149)(cid:133)(cid:134)∑ (cid:135) oxx⃗(cid:137) s(cid:136)⃗N,(cid:137)(oxx⃗(cid:137) ∙ (cid:136)⃗N,(cid:137)) =(cid:150)J z(cid:136)⃗N,(cid:137)z(cid:139)－ z(cid:136)⃗N,(cid:137)z(cid:138) Rewrite above equation by adding up equivalent dipolar field and external magnetic field2: where (12) (cid:140) (13) Hence, the dipole-dipole interaction and external magnetic field have changed the initially reported magnetic anisotropy barrier, as shown in Figure S4. In this work, we have monitored the magnetic responses of SPION samples i - vii, where the MPS results from this work indicate that: samples ii, iv, and v are in status (c) from Figure S4 (with dipolar interactions), samples i, iii, vi, and vii are in status (b) from Figure S4 (without dipolar interactions). Figure S4. Schematic view of energy barriers (a) without external magnetic fields and dipolar interactions; (b) with external magnetic fields only; (c) with both external magnetic fields and dipolar interactions. 8 nonlinear magnetic responses of SPIONs: Supporting Information S7. Mathematical models of the harmonic signals from SPIONs. Taylor expansion of the 𝑀d(𝑡) shows the major mixing frequency components (harmonic signals) due to the 𝑀xx⃑d(𝑡)𝜇𝑐 =𝐿(cid:135)𝜇𝐻xx⃑(𝑡) 𝑘v𝑇(cid:140) =13(cid:153) 𝜇𝑘v𝑇(cid:154)𝐻xx⃑(𝑡)−145(cid:153) 𝜇𝑘v𝑇(cid:154)s𝐻xx⃑(𝑡)s+ 2945(cid:153) 𝜇𝑘v𝑇(cid:154)(cid:157)𝐻xx⃑(𝑡)(cid:157)+⋯ =⋯+(cid:159)−160𝐴,𝐴8(cid:146)(cid:153) 𝜇𝑘v𝑇(cid:154)s+𝑜(cid:135)(cid:153) 𝜇𝑘v𝑇(cid:154)s(cid:140)(cid:160)×𝑐𝑜𝑠[2𝜋(𝑓,±2𝑓8)𝑡]𝑧 +(cid:159) 11512𝐴,𝐴8(cid:133)(cid:153) 𝜇𝑘v𝑇(cid:154)(cid:157)+𝑜(cid:135)(cid:153) 𝜇𝑘v𝑇(cid:154)(cid:157)(cid:140)(cid:160)×𝑐𝑜𝑠[2𝜋(𝑓,±4𝑓8)𝑡]𝑧 +⋯ For an assembly of SPIONs with diameter 𝐷, suspension volume of 𝑉, and nanoparticle concentration of 𝑐, confined in a cylindrical vial inside a pair of pick-up coils (see Figure S2), the overall magnetic moment from the SPION sample consisting the 3rd and the 5th harmonic components is expressed as (in A/m): 𝑀xx⃑d(𝑡)zs(cid:136){≈𝜇𝑐𝑉¢−m£(cid:148)𝐴,𝐴8(cid:146)⁄ o(pP¥s+𝑜(cid:153)⁄ o(pP¥s(cid:154)ƒ×𝑐𝑜𝑠[2𝜋(𝑓,+2𝑓8)𝑡]𝑧 (15) 𝑀xx⃑d(𝑡)z(cid:157)?§≈𝜇𝑐𝑉¢ mm(cid:157)m(cid:146)𝐴,𝐴8(cid:133)⁄ o(pP¥(cid:157)+𝑜(cid:153)⁄ o(pP¥(cid:157)(cid:154)ƒ×𝑐𝑜𝑠[2𝜋(𝑓,+4𝑓8)𝑡]𝑧 (16) (14) According to Faraday's law of Induction, the induced time-varying 3rd and 5th harmonic voltages sensed by the pick-up coils are (in Volt): 𝑢(𝑡)s(cid:136){=−𝑁𝑆{⁄o(cid:149)Txx⃑«(?)z(cid:139)‹›¥ {? 𝑢(𝑡)(cid:157)?§=−𝑁𝑆{⁄o(cid:149)Txx⃑«(?)z(cid:138)fifl¥ where 𝑁 is the number of winding turns in pick-up coils, 𝑆 is the area contained by the pick-up coils 3-5. {? By replacing 𝐻xx⃑(𝑡) with the effective magnetic field 𝐻xx⃗(𝑡)zK~~J (which considers the dipolar fields), the time- (17) (18) varying 3rd and 5th harmonic voltages will be altered, as a result, the harmonic amplitudes after DFT will be altered. 9 García-Otero, J.; Porto, M.; Rivas, J.; Bunde, A., Influence of dipolar interaction on magnetic properties References 1. of ultrafine ferromagnetic particles. Physical review letters 2000, 84 (1), 167. 2. Andersson, J.-O.; Djurberg, C.; Jonsson, T.; Svedlindh, P.; Nordblad, P., Monte Carlo studies of the dynamics of an interacting monodispersive magnetic-particle system. Physical Review B 1997, 56 (21), 13983. 3. Wu, K.; Schliep, K.; Zhang, X.; Liu, J.; Ma, B.; Wang, J. P., Characterizing Physical Properties of Superparamagnetic Nanoparticles in Liquid Phase Using Brownian Relaxation. Small 2017, 13 (22). 4. experimental investigations. Journal of Physics D: Applied Physics 2017, 50 (8), 085005. 5. nanoparticles in liquid phase. Journal of Magnetism and Magnetic Materials 2019, 471, 394-399. Wu, K.; Tu, L.; Su, D.; Wang, J.-P., Magnetic dynamics of ferrofluids: mathematical models and Wu, K.; Su, D.; Liu, J.; Wang, J.-P., Estimating saturation magnetization of superparamagnetic 10
1912.08787	1	1912	2019-12-18T18:38:49	Confronting Interatomic Force Measurements	[ "physics.app-ph" ]	The quantitative interatomic force measurements open a new pathway to materials characterization, surface science, and chemistry by elucidating the force between 'two' interacting atoms as a function of their separation. Atomic force microscope is the ideal platform to gauge interatomic forces between the tip and the sample. For such quantitative measurements, either the oscillation frequency or the oscillation amplitude and the phase of a vibrating cantilever are recorded as a function of the tip-sample separation. These experimental measures are subsequently converted into the interatomic force laws. Recently, it has been shown that the most commonly applied mathematical conversion techniques may suffer a significant deviation from the actual force laws. To avoid assessment of unphysical interatomic forces, either the use of very small (i.e., a few picometers) or very large oscillation amplitudes (i.e., a few nanometers) has been proposed. However, the use of marginal oscillation amplitudes gives rise to another problem as it lacks the feasibility due to the adverse signal to noise ratios. Here we show a new mathematical conversion principle that confronts interatomic force measurements while preserving the oscillation amplitude within the experimentally achievable and favorable limits, i.e. tens of picometers. We anticipate that our findings will be the nucleus of reliable evaluation of material properties with a more accurate measurement of interatomic force laws.	physics.app-ph	physics	Confronting Interatomic Force Measurements Omur E Dagdeviren1,2,* 1 Department of Physics, McGill University, Montreal, Quebec, Canada, H3A 2T8 2 Department of Mechanical Engineering and Materials Science, Yale University, New Haven, Connecticut 06520, USA *Corresponding author's email: [email protected] interatomic force measurements open a new pathway to materials The quantitative characterization, surface science, and chemistry by elucidating the force between "two" interacting atoms as a function of their separation. Atomic force microscope is the ideal platform to gauge interatomic forces between the tip and the sample. For such quantitative measurements, either the oscillation frequency or the oscillation amplitude and the phase of a vibrating cantilever are recorded as a function of the tip-sample separation. These experimental measures are subsequently converted into the interatomic force laws. Recently, it has been shown that the most commonly applied mathematical conversion techniques may suffer a significant deviation from the actual force laws. To avoid assessment of unphysical interatomic forces, either the use of very small (i.e., a few picometers) or very large oscillation amplitudes (i.e., a few nanometers) has been proposed. However, the use of marginal oscillation amplitudes gives rise to another problem as it lacks the feasibility due to the adverse signal to noise ratios. Here we show a new mathematical conversion principle that confronts interatomic force measurements while preserving the oscillation amplitude within the experimentally achievable and favorable limits, i.e. tens of picometers. We anticipate that our findings will be the nucleus of reliable evaluation of material properties with a more accurate measurement of interatomic force laws. Mapping interactions between a sharp probe tip and a sample surface has become possible with the advances of scanning probe microscopy1. Among other interactions that can be gauged by employing scanning-probe-based techniques, force measurements are among the most popular choice2-7. Material characterization with force measurements dates back to the premier realizations of atomic force microscopes8. Initially, forces were recovered by gauging the deflection of the cantilever beam and with the knowledge of cantilever's spring constant9,10. Although this direct deflection measurement-based technique is still applied, the degree of locality is dictated by the dimensions of the tip's apex and the size of the contact area. Besides, this deflection-based technique also suffers from the mechanical instabilities that arise within the proximity of the sample's surface. Sudden instabilities (i.e., jumps-in) occur at the exact distance where the gradient of the attractive surface forces becomes larger than the cantilever's spring constant9,10. Utilization of cantilevers that feature spring constants much higher than the largest force gradient experienced 1 during the approach has become customary particularly for vacuum applications. While the use of stiff cantilevers eliminates mechanical instabilities, it also impedes the applicability of Hooke's law. The high spring constants reduce the cantilever deflections to values that are too small to resolve with sufficient accuracy using the current sensors. As an alternative, "dynamic" operational methodologies for atomic force microscopy (AFM) can be applied to assess interatomic forces11,12. As Figure 1 illustrates, a probe oscillates in the close vicinity of the surface. The perturbations from the harmonic oscillation of the cantilever due to the tip-sample interaction are measured as a function of the tip-sample separation. To this end, two modulation techniques are most commonly used. The frequency modulation (FM) technique is widely employed, which demodulates the resonance frequency f of the cantilever beam under the influence of the tip-sample interaction while keeping the oscillation amplitude constant13. Alternatively, the amplitude modulation (AM)-based techniques track the change of the oscillation amplitude (A) and/or the phase difference between the oscillation and excitation () while driving with a constant excitation signal11. Figure 1: Explanation of experimental procedure of interatomic force measurements. The base of the vibrating cantilever is excited with a sinusoidal signal with an oscillation amplitude ad. The cantilever oscillates with an oscillation amplitude, A. The tip's apex is at a distance, d, away from the surface when it is undeflected. The distance, D = d - A, distinguishes itself from the undeflected cantilever position. The deflection signal is demodulated. The resonance frequency shift (f) or the oscillation amplitude (A) and the phase of the cantilever signal () are recorded as a function of tip-sample separation. The interatomic forces can be recovered by employing the techniques that were developed around the millennium from the demodulated oscillation signal14-18. Equation 1 shows the general 2 form of the conversion technique to assess interatomic forces, FIA, both for AM and FM-type force spectroscopy experiments17,19,20: 𝐹IA(𝐷) = 𝜕 𝜕𝐷 ⌊2𝑐𝑧 ∫ 𝑃𝐹 [(𝑧 − 𝐷) + (√ ∞ 𝐷 𝐴 16𝜋 √𝑧 − 𝐷) + ( 𝐴3 2⁄ √2(𝑧−𝐷) )] 𝑑𝑧 ⌋ (1) In equation 1, cz is the spring constant of the cantilever. The first term in the integral, PF, is a pre- factor which changes for AM and FM-based force spectroscopy experiments17,19,20: The position of the tip is presented by z, while D is the nearest tip-sample distance and A is the oscillation amplitude as explained in Figure 1. Different terms in the parenthesis dominate the integral as a function of A and the decay length of the interatomic force, l17,21. When A < l, the first term in the parenthesis, which the term is written in red, dominates the integral. The integral is dominated by the term written in green for A > l. Subsequently, the second term (written in orange) of equation 1, which is an approximation term, dominates the integral when A ≈ l. Recently, it has been shown that the interatomic force measurements that employs equation 1 may deviate significantly from the actual force laws21,22. More specifically, sudden changes in the interatomic force laws may impede the reconstruction of the force from the measured quantities due to the limitations of renormalization schemes and approximations applied in equation 1. It was proposed that the oscillation amplitude should be adjusted depending on the distance and the degree of the sudden change in the interatomic forces to avoid problems associated with the limitations of the mathematical conversion principles21,22; however, marginal oscillation amplitudes provoke another experimental problem which is the attenuated signal to noise ratio11,12. Our numerical analysis reveals that the mathematical instabilities may also arise because of the small amplitude term of equation 1 (i.e. term written in red) in the existence of a sudden change in the interatomic forces as a function of distance. Here we show a new mathematical conversion principle that confronts the reconstruction of interatomic force laws with a significant relaxation of applicable oscillation amplitude range for a successful measurement with the experimentally achievable range of oscillation amplitudes. Lee et al. developed a general theory for the reconstruction of interatomic force laws for AM-based force spectroscopy experiments, which works for small-amplitude range without any instabilities18. However, at large oscillation amplitudes, the technique is mathematically intense and may deviate from the actual force laws18. We propose a hybrid technique that employs the general theory for the reconstruction of interatomic force laws to avoid mathematical instabilities 3 of the small amplitude term in equation 1. The hybrid technique for the reconstruction of the interatomic force laws is presented by equation 2: 𝐹IA(𝐷) = ∫ 𝑑𝑧 [ ∞ 𝐷 𝑐𝑧𝐴0 sin 𝜑(𝑧) 𝑄𝐴(𝑧) − (𝑐𝑧 − 𝑚𝑤2)] + 𝜕 𝜕𝐷 ⌊2𝑐𝑧 ∫ 𝑃𝐹 [(√ 𝐴 ∞ 𝐷 16𝜋 √𝑧 − 𝐷) + ( )] 𝑑𝑧 ⌋ (2) 𝐴3 2⁄ √2(𝑧−𝐷) In this hybrid technique, the effective mass, m, is replaced by 𝑐𝑧/(2𝜋𝑓0)2. The angular frequency term, w, equals to (2𝜋𝑓). The term f equals to drive frequency of the cantilever for AM- modulation. The technique can be extended for FM-modulation by introducing (f0 + f) for f. Also, the sinusoidal term in the summation drops for FM-modulation as the phase is kept 90 by the control electronics for a constant oscillation amplitude12. For this reason, FM modulation-based force spectroscopy using equation 2 is immune to uncertainty of the oscillation amplitude for small oscillation amplitudes which is a major advantage over the existing techniques22. To test our methodology, we followed a commonly used approach for dynamic AFM. We solved the equation of motion of a damped harmonic oscillator with external excitation and a model interatomic force23,24. We calculated A and  for AM-modulation and f for FM-modulation as a function of tip-sample separation. We then reconstructed the model interatomic force field both with equation 1 (i.e., method 1) and equation 2 (i.e., method 2). Figure 2 shows a summary of our results for four different interatomic force models. As Figure 2 illustrates, we can successfully reconstruct the force even after the sudden change in the interatomic force as a function distance, also known as the inflection point22. According to former proposals, the lower limit of the required oscillation amplitude approximates to zero for a successful force reconstruction by using method 1 as the derivative of the interatomic force model is discontinuous. As illustrated in Figure 2 a-d, method 1 suffers from mathematical instabilities after the inflection point both for AM and FM modulation-based reconstruction. Method 2, however, can successfully reconstruct the interatomic force model even in the existence of multiple inflection points while experimentally feasible oscillation amplitudes are employed. This results in a significant relaxation of the applicable oscillation amplitude range for force measurements. 4 Figure 2: The comparison of two different force reconstruction methodologies for different interatomic force models, (a) Hertz-Offset model (for details see Ref 21), (b) ramp force model, (c) triangle force model with two inflection points, (d) triangle force model with three inflection points. The reconstruction of tip-sample interaction force is presented with different colors for different oscillation amplitudes while the model force is presented in black. (a-d) Our numerical results show that method 1 suffers from mathematical instabilities. (a-d) Method 2 successfully eliminates mathematical instabilities even in the existence of multiple inflection points for oscillation amplitudes that are feasible from the experimental perspective both for AM and FM modulation-based interatomic force reconstruction. Our technique enables confronting the interatomic force measurements by eliminating mathematical instabilities for experimentally achievable oscillation amplitudes. For this reason, it has a major advantage over existing practices. Although our methodology is significantly more 5 robust than the existing techniques, a more generalized mathematical conversion principle for all oscillation amplitude ranges and all possible interatomic force laws remains an open question. Acknowledgements I would like to thank Prof. Peter Grütter and Prof. Udo D. Schwarz for fruitful discussions. Financial support from the Natural Sciences and Engineering Research Council of Canada and Le Fonds de Recherche du Québec - Nature et Technologies are gratefully acknowledged. References 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Bonnell, D. A. et al. Reviews of Modern Physics 84, 1343-1381, (2012). Lantz, M. A. et al. Science 291, 2580-2583, (2001). Sugimoto, Y. et al. Nature 446, 64-67, (2007). Ternes, M., Lutz, C. P., Hirjibehedin, C. F., Giessibl, F. J. & Heinrich, A. J. Science 319, 1066- 1069, (2008). Gross, L. et al. Science 324, 1428-1431, (2009). Weymouth, A. J., Hofmann, T. & Giessibl, F. J. Science 343, 1120, (2014). Kawai, S. et al. Science 351, 957, (2016). Binnig, G., Quate, C. F. & Gerber, C. Physical Review Letters 56, 930-933, (1986). Yongho, S. & Wonho, J. Reports on Progress in Physics 71, 016101, (2008). Burnham, N. A., Colton, R. J. & Pollock, H. M. Nanotechnology 4, 64, (1993). Garcia, R. Amplitude Modulation Atomic Force Microscopy. (Wiley-VCH, 2010). Giessibl, F. J. Reviews of Modern Physics 75, 949-983, (2003). Albrecht, T. R., Grutter, P., Horne, D. & Rugar, D. Journal of Applied Physics 69, (1991). Giessibl, F. J. Physical Review B 56, 16010-16015, (1997). Hölscher, H., Allers, W., Schwarz, U. D., Schwarz, A. & Wiesendanger, R. Physical Review Letters 83, 4780-4783, (1999). Dürig, U. Applied Physics Letters 75, 433-435, (1999). Sader, J. E. & Jarvis, S. P. Applied Physics Letters 84, 1801-1803, (2004). Lee, M. & Jhe, W. Physical Review Letters 97, 036104, (2006). Amir, F. P., Daniel, M.-J. & Ricardo, G. Nanotechnology 26, 185706, (2015). Dagdeviren, O. E., Götzen, J., Hölscher, H., Altman, E. I. & Schwarz, U. D. Nanotechnology 27, 065703, (2016). Dagdeviren, O. E., Zhou, C., Altman, E. I. & Schwarz, U. D. Physical Review Applied 9, 044040, (2018). Sader, J. E., Hughes, B. D., Huber, F. & Giessibl, F. J. Nature Nanotechnology 13, 1088-1091, (2018). Anczykowski, B., Krüger, D., Babcock, K. L. & Fuchs, H. Ultramicroscopy 66, 251-259, (1996). Hölscher, H. Journal of Applied Physics 103, 064317, (2008). 6
1801.10317	1	1801	2018-01-31T06:47:00	Highly efficient visible colloidal lead-halide perovskite nanocrystal light-emitting diodes	[ "physics.app-ph" ]	Lead-halide perovskites have been attracting attention for potential use in solid-state lighting. Following the footsteps of solar cells, the field of perovskite light-emitting diodes (PeLEDs) has been growing rapidly. Their application prospects in lighting, however, remain still uncertain due to a variety of shortcomings in device performance including their limited levels of luminous efficiency achievable thus far. Here we show high-efficiency PeLEDs based on colloidal perovskite nanocrystals (PeNCs) synthesized at room temperature possessing dominant first-order excitonic radiation (enabling a photoluminescence quantum yield of 71% in solid film), unlike in the case of bulk perovskites with slow electron-hole bimolecular radiative recombination (a second-order process). In these PeLEDs, by reaching charge balance in the recombination zone, we find that the Auger nonradiative recombination, with its significant role in emission quenching, is effectively suppressed in low driving current density range. In consequence, these devices reach a record high maximum external quantum efficiency of 12.9% reported to date and an unprecedentedly high power efficiency of 30.3 lm W-1 at luminance levels above 1000 cd m-2 as required for various applications. These findings suggest that, with feasible levels of device performance, the PeNCs hold great promise for their use in LED lighting and displays.	physics.app-ph	physics	Highly efficient visible colloidal lead-halide perovskite nanocrystal light-emitting diodes Fei Yan1‡, Jun Xing2,3‡, Guichuan Xing4, Lina Quan3, Swee Tiam Tan1, Jiaxin Zhao2, Rui Su2, Lulu Zhang2, Shi Chen2, Yawen Zhao5, Alfred Huan2, Edward H. Sargent3, Qihua Xiong2,6, and Hilmi Volkan Demir1,2,7* 1 LUMINOUS! Center of Excellence for Semiconductor Lighting and Displays, TPI-The Photonics Institute, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, 2 Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore 637371, 3 Department of Electrical and Computer Engineering, University of Toronto, 10 King's College Road, Toronto, Ontario M5S 3G4, Canada, 4 Institute of Applied Physics and Materials Engineering, University of Macau, Macao SAR 999078, China, 5 China Academy of Engineering Physics, Mianyang 621900, China, 6 NOVITAS, Nanoelectronics Centre of Excellence, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, 7 Department of Electrical and Electronics Engineering, Department of Physics, UNAM−Institute of Materials Science and Nanotechnology, Bilkent University, Ankara 06800, Turkey. ‡These authors contributed equally to this work. *e-mail: [email protected], [email protected], [email protected] Lead-halide perovskites have been attracting attention for potential use in solid-state lighting. Following the footsteps of solar cells, the field of perovskite light-emitting diodes (PeLEDs) has been growing rapidly. Their application prospects in lighting, however, remain still uncertain due to a variety of shortcomings in device performance including their limited levels of luminous efficiency achievable thus far. Here we show high-efficiency PeLEDs based on colloidal perovskite nanocrystals (PeNCs) synthesized at room temperature possessing dominant first-order excitonic radiation (enabling a photoluminescence quantum yield of 71% in solid film), unlike in the case of bulk perovskites with slow electron-hole bimolecular radiative recombination (a second-order process). In these PeLEDs, by reaching charge balance in the recombination zone, we find that the Auger nonradiative recombination, with its significant role in emission quenching, is effectively suppressed in low driving current density range. In consequence, these devices reach a record high maximum external quantum efficiency of 12.9% reported to date and an unprecedentedly high power efficiency of 30.3 lm W-1 at luminance levels above 1000 cd m-2 as required for various applications. These findings suggest that, with feasible levels of device performance, the PeNCs hold great promise for their use in LED lighting and displays. 2 Lead-halide perovskites with direct band gaps have emerged as a new material system for low-cost, large-area and light-weight optoelectronics1-21. Previous reports have confirmed promising performance for this material family in light generation13-21. With binding energy up to dozens of meV and long electron-hole diffusion length of micrometers at room temperature, carriers are almost free and delocalized in bulk perovskites16,22-24 leading to slow electron-hole bimolecular radiative recombination (a second-order process)11,24. Competing with trap-mediated nonradiative decay (a first-order process), this radiation process limits the efficiency levels11,24. Recently a strategy that enhances the photoluminescence quantum yield (PLQY) of light emitters and external quantum efficiency (EQE) of their LEDs has been developed by localizing the charge carriers in small sized lead-halide perovskite grains or quantum wells16-21. However, even though such colloidal perovskite nanoparticles can present PLQY of near unity in solution and above 60% in film, their performance levels have been still severely limited during the device operation to date19,20,25,26. Therefore, deeper understanding of device performance limits and a new way divorced from this low performance are required. Here we present highly efficient PeLEDs based on colloidal CH3NH3PbBr3 (MAPbBr3) PeNCs with efficient first-order excitonic radiation reaching a 71% PLQY in film. Due to the significant role of Auger nonradiative recombination, highly efficient emission at low charge densities is desired. By effectively suppressing Auger nonradiative recombination with balanced charge at low driving charge density levels, these PeLEDs exhibit a record high EQE of 12.9% up to date and a high power efficiency level of 30.3 lm W-1 and high brightness level above 1000 cd m-2, which indicates the feasibility of colloidal PeNCs in LED applications in terms of efficiency requirements. Luminescence dynamics of lead-halide perovskite emitters The radiation of bulk perovskites originates from slow bimolecular recombination. The Langevin theory predicts that this bimolecular recombination rate could be four orders-of-magnitude higher in perovskites, which is one of the main merits ensuring superior performance in photovoltaics11,24,27-30. The carrier kinetics in bulk lead-halide perovskites can be described by: 3 () =−−− (1) where n is the carrier density, t is the time, G is the generation rate of carrier density, k1 is the first- order trap-mediated monomolecular recombination rate constant, k2 is the free carrier bimolecular recombination rate constant, and k3 is the three-body Auger recombination rate constant11,24,27-30. Thus, under steady-state electric excitation, the luminous QY (A(n)) is given by: (2) ()= The emission efficiency (Fig. 1a) first increases with increasing n due to the domination of bimolecular recombination over the trap-mediated nonradiative recombination. However, at high carrier densities, where the Auger recombination dominates over the bimolecular recombination, the emission efficiency will decrease with increasing n. Since k2 (~10-10 cm3 s-1) and k3 (~10-28 cm6 s-1) are the intrinsic parameters of the bulk perovskites11,24,27-30, the emission efficiency strongly depends on k1 and n. Ignoring other quenching factors, with a typical reported value of k1 (~107 s-1, 100 ns), the internal efficiency of the PeLEDs is constrained to <10%, even when driving at a high carrier density of 1016 cm-3 (Fig. 1a). The bulk perovskite films have shown a long carrier lifetime (i.e., low trap states density)31 and a high mobility32. Thus, the delocalized charge carriers are also possible to transport through the thin active layer without recombination. When the PeLEDs are driven at a lower carrier density level or under a higher trap density regime (trap-related recombination time <100 ns), the nonradiative recombination could always dominate over the bimolecular recombination, which also leads to an internal QY of <10%. Therefore, the slow bimolecular recombination rate in bulk perovskites limit the PeLEDs working at a relatively low carrier density. The quadratic power dependence of the recombination rate on the carrier densities of MAPbBr3 bulk crystal (Fig.1b) confirms the bimolecular recombination in the bulk perovskite. The experimentally measured curves match well with the theoretical prediction at the low charge density level (Fig. 1a, c). It is noted that the first-order excitonic emission could compete effectively with the trap-mediated nonradiative recombination in the low carrier density range of n ≤ 1016 cm-3 with trap states density (k1 ~ 107 s-1) as shown in Fig.1a. Therefore, 4 small-size PeNCs can enhance the radiative excitonic recombination with increased exciton confinement and exciton binding energy comparing to bulk crystals with dominating free-carrier recombination. Moreover, the surrounded passivation of ligands can decrease the surface defects of PeNCs, which also leads to an increment of QY at low excitation density level. Synthesis of PeNCs and their luminescence dynamics The colloidal MAPbBr3 PeNCs were synthesized by using anti-solvent precipitation method at room temperature. The as-synthesized solution was pre-aged for 6 days before being used for LEDs fabrication. In this duration, the absorption edge and PL peak gradually red-shift until after 6 days, and the full-width-at-half-maximum (FWHM) of PL spectra keeps narrowing synchronously (Fig. 1d, Supplementary Fig. 1 ). Notably, the PLQY was significantly enhanced from 70% in solution and 35% in film to 89% in solution and 71% in film. From nuclear magnetic resonance (NMR), the composition of the MAPbBr3 NCs was found to remain stable during the aging time (Supplementary Table 1). However, as observed in X-ray diffraction (XRD) patterns, the MAPbBr3 NCs show increased crystallization during this process (Fig. 1e). The high-resolution transmission electron microscopy (HR-TEM) measurements (Fig. 1f) illustrate that the aged PeNCs still exhibit a small crystal size (< 8 nm). The lattice fringes of the MAPbBr3 PeNCs are 2.93 and 2.65 Å, which correspond to the (200) and (210) facets of the cubic phase MAPbBr3, respectively. With the passivation of ligands and solvent molecules, the low-crystallinity MAPbBr3 NCs in colloidal solution possess a high PLQY. However, the passivation effect decreases drastically in film (Supplementary Table 1), thus the defect sites become significant in quenching emission through the competition with radiation process. Though with similar passivation, the highly crystallized NCs with a low defect sites display low trap-assisted nonradiative recombination, which ensures a high PLQY in film. Even with such passivation effects, the PeNCs still shows high mobility of 0.22 cm2 V-1 S-1 for holes and 0.48 cm2 V-1 S-1 for electrons (Supplementary Fig. 2). 5 As illustrated in Fig. 1b, for MAPbBr3 PeNCs, IPL[t = 0] is nearly linearly dependent on the carrier density from 5 × 1014 to 2 × 1017 cm-3, which suggests that the emission is mainly from the first-order excitonic radiation process. Below 5 × 1014 cm-3, the trap-mediated nonradiative recombination dominates over the excitonic emission. Above 2 × 1017 cm-3, the multi-particle Auger recombination prevails. Therefore, MAPbBr3 PeNCs feature a high PLQY level under the charge carrier density from 1015 to 1016 cm-3, which matches well with the device working regime in PeLEDs (Fig. 1c)11,12. Therefore, the highly efficient first-order of excitonic radiation enables the MAPbBr3 PeNCs as a suitable candidate possibly for efficient PeLEDs. Investigation of PeLEDs As the topmost important parameter for EL device characterization, EQE, which is defined as the number of emitted photons per injected electrons, is given as the product of four different parameters: =× / × × (3) Therein the first parameter Ein indicates the charge balance factor in recombination. The second one Es/t gives the fraction of excitons that are allowed to radiate by spin statistics. The third factor Erad depicts the fraction of the intrinsic radiative efficiency of the emitter. The last one Eout determines the photon extraction efficiency in a device. In this work with the proposed devices using normal structure without any additional measurement for spin injection of charge carriers or enhanced outcoupling efficiency of radiation, thus all attention was paid onto the charge balance factor in recombination zone to reach the high device efficiency promised by MAPbBr3 PeNCs. Moreover, due to the significant role of Auger nonradiative recombination (Fig. 1a, c) which is to be effectively suppressed in the proposed MAPbBr3 PeNCs devices, a high EQE in low driving current density level was also targeted. In addition to EQE, the power efficiency, especially at high brightness level, is also a critical parameter for evaluating the potential applications of LEDs; therefore, a low driving voltage is essential. To obtain high quality PeNCs film, a classical PEDOT:PSS layer was selected as hole injection/transport layer (HIL/HTL) for all devices in this work (Fig. 2a, b, Supplementary Fig. 3, 4). 6 As a good HIL/HTL, PEODT:PSS may lead to emission quenching of PeNCs at HTL/PeNCs interface, but we did not observe a significant weakness of emission comparing to the PeNCs film deposited onto other polymer substrates, like poly-TPD, at an excitation wavelength of 365 nm. The insertion of an insulating thin-film can improve charge balance by blocking the majority carriers, but this would probably be accompanied with high driving voltage, which would lead to high power consumption and low power efficiency. To balance the injected hole, a blended layer of B3PYMPM:TPBi (1:X w/w) with adjustable electron transport capability by changing the ratio of two components served as an electron transport layer (ETL) (Fig. 2a, b). The deeper LUMO level (-3.4 eV) and high electron conductivity (4.1 × 10−7 S cm-1) of B3PYMPM can lower down the barrier for effective electron injection and transport with low driving voltage20,33,34. Due to the significant difference of the electron conductivity in TPBi and B3PYMPM33, the electron transport capability of the blended ETL can be tuned in a large range (Fig. 2c). According to Miller−Abrahams hopping model, electrons transport through a hopping process between the localized occupied and unoccupied state. In the blended ETL, TPBi (B3PYMPM) molecules play the role of an energy barrier (trap) to scatter the electron transfer between two B3PYMPM (TPBi) molecules, leading to a tunable electron transport capability35. In the single carrier device with TPBi ETL, as the low electron conductivity (4.2 × 10−10 S cm-1) and high energy barrier for electron injection caused by the high lying LUMO level of TPBi, the driving voltage is much higher than that of single carrier device with B3PYMPM ETL, which can offer a much higher level of electron conductivity and a deeper LUMO level (Fig. 2a, c)33,34. For the single carrier device with B3PYMPM:TPBi (1:X, w/w) ETL, the blended layer shows a tunable electron transport capability depending on the varied component ratio (Fig. 2c). To reach a high power efficiency a n-type doped electron injection layer was used for a low driving voltage20. Control device I with pure TPBi and control device II with pure B3PYMPM were selected as the baseline. As seen in Fig. 2d, PeLEDs show a green color emission with a nearly ideal Lambertian profile (Supplementary Fig. 5). Control device I exhibits overall performance of 15,130 cd m-2 in Lmax, 7 5.11 % in EQEmax, and 10.82 lm W-1 in ηmax (Figs. 2, 3, Table 1, and supplementary Table 2). On the other hand, control device II displays a higher Lmax of 24,410 cd m-2 and a drastically decreased driving voltage, but a lower ηmax of 5.96 lm W-1 caused by a lower EQEmax of 2.39% compared to control device I (Table 1 and Figs. 2, 3). With B3PYMPM:TPBi (1:2, w/w) ETL control device III shows the best overall performance up to now, with Lmax of 33,570 cd m-2, EQEmax of 7.91%, and ηmax of 20.18 lm W-1 (Figs. 2, 3, Table 1, and supplementary Table 3). In all devices, the influence of Es/t, Erad, and Eout can be neglected because the emitters are identical and the difference in refractive index between TPBi and B3PYMPM is also negligible. Therefore, the significant difference in EQEmax among control devices is primarily caused by different charge balance in the recombination zone. Generally, the brightness in a LED is correlated with the amount of activated emitters in the recombination zone. In all control devices, the nanocrystals distribution should be the same, hence the higher brightness means more activated emitters in the recombination zone. As the driving current density increases, more emitters were excited, leading to increment of brightness. With the highest charge supply capability of B3PYMPM, control device II, instead, gave the lowest luminous efficiency among all devices, even though its Lmax is higher than that of control device I by 61%. As the emitters were the same for all devices, the possible reason is the imbalanced charge in the recombination zone of control device II. Additionally, the red emission in Ir(piq)2acac indicator device with TPBi ETL suggested the injection of remaining hole into ETL as the maximum effective radius of Förster energy is 6 nm. It was thus deducted that hole is the major charge in the recombination zone of control device I (Supplementary Fig. 6). Otherwise, the pure PeNCs green color emission in Ir(piq)2acac indicator device with B3PYMPM ETL would have suggested there was less or no hole injection into ETL comparing to TPBi devices (Supplementary Fig. 6). With the same hole injection, the highest EQEmax suggested the best charge balance in control device III with medium electron supply ETL comparing to I and II (Fig. 2, 3). Therefore, the lowest EQEmax of control device II was ascribed to the overwhelming majority of electrons in the PeNCs emissive layer as the highest electron supply of B3PYMPM. 8 tail extended into band gap as gap states36. Moreover, as = (where σ is electron conductivity, In amorphous organic semiconductor the LUMO/HOMO level distribution was broad, of which n is the electron concentration, e is the elementary charge, µe is the electron mobility) indicates, a comparable electron mobility but significantly high electron conductivity suggest high electron concentration. Thus, here it was deducted that the amount of gap sates in B3PYMPM is much higher comparing to TPBi, which can be supported by the smaller distance between the Fermi level and LUMO level of B3PYMPM comparing to TPBi (Supplementary Fig. 7). In consequence, with a deeper LUMO level, there were more electrons injected into B3PYMPM layer even when the applied bias was less than the threshold voltage of 2.2 V comparing to TPBi (Fig. 2e). By the law of charge conservation, in a closed-loop circuit, the electrons injected into B3PYMPM layer will transit to the valence band of PeNCs by a nonradiative process, leading to leakage current (Fig. 2a, e). Moreover, under high driving voltage the excessive electrons injected into the conductive band of PeNCs will transit to the Fermi level of PEDOT:PSS by a nonradiative process as current overflow, which is another way of leakage current for control device II working in high driving current density range. The leakage of carriers which flow through the device without recombining drives down the EQE directly. The imbalanced charge also leads to an enhanced Auger nonradiative recombination. Thus, control device II presented the lowest EQEmax among all devices. As exhibited in Fig. 2e, with blended ETL, the leakage current density was suppressed significantly by 3-4 magnitudes. In most practical LEDs such leakage current was ascribed to device inherent structure. Therefore we believe that, in control device III by decreasing the amount of electron injected into the recombination zone with a blended ETL, the leakage current was suppressed and the charge balance in recombination zone was improved simultaneously. Furthermore, a thicker PeNCs layer containing more emitters but fewer pinholes will be helpful to obtain a broader recombination zone, higher luminance, and lower leakage current density. As the comparable mobility of hole and electron, the thickness change in PeNCs layer will not influence the charge balance in the recombination zone significantly. Therefore, a thicker (80 nm) PeNCs layer was 9 deposited using a mixed THF:CB (Y:1 v/v, Y=16) PeNCs colloidal solution, by means of which the champion device reached a record EQEmax of 12.9 % with the lowest leakage current (Fig. 2e, 3a, Table 1). Assuming the light outcoupling efficiency of around 20%, the estimated internal quantum yield reaches 64.5%20. The EQEmax histogram for 15 devices from 4 batches describes the distribution of EQEmax of the fabricated PeLEDs (Fig. 3d). Due to the significant difference between CB and THF colloidal solution in PeNCs film deposition (Supplementary Fig. 8, 9), the reproducibility of PeNCs film deposition using THF:CB (Y:1 v/v, Y=16) colloidal solution is moderate. To balance the EQEmax and Lmax, the peak value of 43,440 cd m-2 in Lmax, 12.9% in EQE, and 30.3 lm W-1 in η (at a luminance level of 2,069 cd m-2, which meets the requirement for lighting application of 1000 cd m-2) were observed from different champion devices (Table 1, Fig. 3). Furthermore, we measured the device half-lifetime to be around 6 min (Supplementary Fig. 10), which is still a daunting problem for PeLEDs, and there is a considerable margin to improve. Moreover, with the same device structure except a slightly changed ratio of B3PYMPM:TPBi (1:1.5, w/w), a PeLEDs with emitter of CsPbBr3 nanocrystals also offered a high performances, including a EQEmax of 12.4%, which details will be dicussed in the following work. For all devices it was found that a higher EQEmax corresponds to a lower driving current density (Jc) (Fig. 3a, e). As the device structures were different, the exponential dependence was probably related to the luminescence dynamics of PeNCs. As the rising of driving current density EQEs increases from very low initial values (Fig. 3a), which is caused by the competition of traps. Additionally, as the high mobilities of PeNCs holes and electrons meet and recombine with low probability, which is also a reason for the negligible EQE value in the low driving current density. With the increasing driving current density, the radiative excitonic recombination takes a dominating role gradually. With the further increase of the driving current density, the Auger nonradiative recombination dominates over the excitonic recombination. However, the imbalanced charges in the recombination zone, which do not exist in the photoexcitation process, boosts the nonradiative Auger process even at the low driving current density level (Fig. 3a, f). As shown by the electron single carrier 10 PeNCs device, the excessive electrons injected into the PeNCs layer leads to a significantly quenched PL even at a low electron density (n) level. Though the charge balance factors are the same in the recombination zone for control device III and champion device, the thicker emissive layer of champion device can decrease the charge density, allowing for a weak Auger nonradiative recombination compared to control device III. Conclusion In summary, we demonstrated highly efficient PeLEDs using colloidal organometal halide PeNCs. The high-quality colloidal MAPbBr3 PeNCs with efficient one-order excitonic emission was synthesized at room temperature. As the significant Auger nonradiative recombination caused by high driving current density and imbalanced charge, the PeLEDs with optimized charge balance in low driving current density level were developed and reached a record EQEmax of 12.9% up to date accompanied with high ηmax of 30.4 lm W-1 at high brightness above 1000 cd m-2. This work demonstrates a breakthrough advancement toward the ultimate goal of application of PeLEDs and lays down a systematic constructive approach for researchers on halide PeLEDs in their future works. Methods Materials and chemicals. Methylamine (MA) solution in ethanol, formamidine acetate, hydrobromic acid (HBr), hydroiodic acid (HI), lead bromide (PbBr2), lead iodide (PbI2), octylamine, anhydrous N,N-dimethylformamide (DMF), γ-butyrolactone, tetrahydrofuran (THF), chlorobenzene (CB) were all purchased from Sigma-Aldrich. PEDOT:PSS (AI 4083) was purchased from Heraeus. 1,3,5-tris(N- phenylbenzimiazole-2-yl)benzene (TPBi), cesium carbonate (Cs2CO3), 4,6-Bis(3,5-di(pyridin-3- yl)phenyl)-2-methylpyrimidine (B3PYMPM), Poly(4-butylphenyldiphenylamine) (poly-TPD) and Bis(1-phenylisoquinoline)(acetylacetonate)iridium(III) (Ir(piq)2(acac)) were purchased from Lumtec. The colloidal zinc oxide (ZnO) nanoparticles (N-11) was purchased from Avantama. All chemicals 11 were used as received. Lead-halide perovskite preparation. MABr: CH3NH2 solution in ethanol was reacted with HBr in water with excess methylamine at 0 °C. The precipitate was recovered via evaporation at 60 °C. The yellowish raw product was washed with diethyl ether for 5 times and dried at 60 °C in a vacuum oven for 24 h. MAPbBr3 PeNCs: 0.225 mmol MABr and 0.15 mmol PbBr2 were dissolved in 0.5 mL anhydrous DMF and 8.5 mL γ-butyrolactone mixed solution with 15 μL octylamine to form a halide perovskite precursor solution. 0.3 mL of the perovskite precursor solution was dropped into 5 mL toluene with vigorous stirring. The colloidal solution was stored in sealed vial for one week before using. After ageing one week, we can get the perovskite NCs with higher PLQY. Then, the colloidal solution was centrifuged at 5000 rpm for 5 min to discard the aggregated precipitates and the PeNCs was obtained after centrifuged at 14,800 rpm. The PeNCs product was dispersed in solvent to form the colloidal solution with 5 mg/mL concentration for LEDs fabrication. MAPbBr3 bulk crystal: 1 mmol MABr and 1 mmol PbBr2 were dissolved in 0.5 mL DMF and 0.5 mL γ-butyrolactone mixed solution and stored in open vial. The precursor solution was kept in oven at 60 °C for 24 h for crystal growing. The separated bulk crystals were collected for measurement (Supplementary Fig. 11). PeNCs characterization. The structure of as-grown samples was characterized using X-ray diffraction (XRD, Bruker D8 Advanced Diffractometer with Cu Kα radiation) and transmission electron microscopy (TEM) with energy-dispersive X-ray spectroscopy (TEM; subångström-resolution, aberration-corrected TITAN G2 60-300, operated under 60 kV). UV-vis absorption was measured using LAMBDA 950 UV/Vis/NIR spectrophotometer and PL spectra were recorded through a micro- 12 Raman spectrometer (Horiba, LABRAM HR800) with a single-grating setup and a 442 nm solid state laser were used as the excitation laser. The morphology and structure of the LED device were characterized by atomic force microscopy (AFM, Cypher ES SPM) in the contacting mode and field- emission scanning electron microscopy (FE-SEM, JEOL JSM-7001F). UPS measurements were performed on the samples in a home-built UHV multi-chamber system with base pressure better than 1 × 10-9 torr. NMR was measured on Bruker 500MHz spectrometer and DMSO-D6 was used as solvent to dissolve the sample. The UPS source is from a helium discharge lamp (hν = 21.2 eV). The photoelectrons were measured by an electron analyzer (Omicron EA125). PLQY measurement. PLQY measurements of colloidal PeNCs samples dispersed in THF solution and thin-film samples deposited onto sapphire substrate using the identical colloidal solution were carried out by coupling a Quanta-Phi integrating sphere to the Horiba Fluorolog system with optical fiber bundles. Both excitation and emission spectra were collected for the two cases of the sample directly illuminated by the excitation beam path in the integrating sphere and the empty sphere itself. A monochromatized Xe lamp was used as excitation source with wavelength of 400 nm and power of 1 mW cm-2. Transient photoluminescence. For femtosecond optical spectroscopy, the laser source was a Coherent LibraTM regenerative amplifier (50 fs, 1 KHz, 800 nm) seeded by a Coherent VitesseTM oscillator (50 fs, 80 MHz). 800 nm wavelength laser pulses were from the regenerative amplifier while laser pulses at 400 nm were obtained with a BBO doubling crystal. The laser pulses (circular spot, diameter 3.5 mm) were directed onto the films. The emission from the samples was collected at a backscattering angle of 150° by a pair of lenses into an optical fiber that was coupled to a spectrometer (Acton, Spectra Pro 2500i) and detected by a charge coupled device (Princeton Instruments, Pixis 400B). The PL at different excitation intensity levels was recorded under the same exact experimental conditions except for the excitation intensity, which was controlled by adjusting the neutral density 13 filters. Under the assumption of negligible nonlinear absorption and negligible sample damage, the relative PLQY at different excitation intensity was obtained by normalizing the collected PL intensity with the corresponding excitation light intensity. Transistor testing. The transistors performance was measured to extract mobility with a Keithley 4200 semiconductor parameter analyzer in air at room temperature. A Si/SiO2 substrate with patterned gold electrodes was selected as a substrate to make a bottom gate field effect transistor, and a PeNCs film was deposited on top as the active layer by spin-casting. PeLEDs fabrication. The patterned ITO-glass substrates were cleaned with de-ionized water, acetone and isopropanol in sequence, followed by oven-drying at 80 C for 2 h and O2 plasma treatment (FEMTO) for 15 min. PEDOT:PSS was deposited onto the substrate by spin-casting at 4,000 rpm for 60 s, and annealed at 120 °C for 30 min in air. Colloidal PeNCs solution was spin-coated onto PEDOT:PSS to form the PeNCs layer. The remaining functional layers and aluminum electrodes were fabricated by thermal evaporation in a single run at a base pressure of <2 × 10-4 Pa without breaking the vacuum. All the organics were evaporated at a rate of around 0.1-0.2 nm s-1, and the aluminum electrodes were evaporated at a rate of 0.8-1 nms-1. A shadow mask was used to define the cathode. All devices were completed with rudimentary encapsulation in glovebox for characterization in air at room temperature. The PEDOT:PSS film here serves as the ITO modification layer to enhance the injection and transport of hole. The pure TPBi and B3PYMPM film serve as electron transport layer (ETL). The TPBi:Cs2CO3 and B3PYMPM:Cs2CO3 film serve as the n-type doped electron injection layer (EIL) which can enhance the electron injection and decrease the driving voltage effectively20. All blended function layers were deposited using multi-source co-evaporation. The lowest unoccupied molecular orbital (LUMO) and the highest occupied molecular orbital (HOMO) of all materials were obtained from references20,34,37. The valance band and conductive band of PeNCs were obtained by ultraviolet photoelectron spectroscopy (UPS) and absorption spectrum (Supplementary Fig. 1, 3). 14 Electron-only devices fabrication. The devices structures are as follows: Device 1: ITO/ZnO (40 nm)/TPBi (30 nm)/TPBi:Cs2CO3 (10:1 w/w, 20 nm)/Al (100 nm). Device 2: ITO/ZnO (40 nm)/TPBi:B3PYMPM (2:1 w/w, 30 nm)/B3PYMPM:Cs2CO3 (10:1 w/w, 20 nm)/Al (100 nm). Device 3: ITO/ZnO (40 nm)/TPBi:B3PYMPM (1:2 w/w, 30 nm)/B3PYMPM:Cs2CO3 (10:1 w/w, 20 nm)/Al (100 nm). Device 4: ITO/ZnO (40 nm)/ B3PYMPM (30 nm)/B3PYMPM:Cs2CO3 (10:1 w/w, 20 nm)/Al (100 nm). Device with PeNCs layer: ITO/ZnO (40 nm)/ MAPbBr3 PeNCs (40 nm)/B3PYMPM (30 nm)/B3PYMPM:Cs2CO3 (10:1 w/w, 20 nm)/Al (100 nm). ZnO nanoparticle film was deposited onto the substrate by spin-casting at 3,000 rpm for 60 s, and annealed at 150 °C for 10 min in air. Except the deposition of ZnO nanoparticles film, the device fabrications were the same with PeLEDs. MAPbBr3 PeNCs layer deposition. Control devices: A THF solvent dispersed PeNCs colloidal solution was selected for the PeNCs layer deposition with an optimized rotational speed of 1,500 rpm (Supplementary Table 2 and Supplementary Fig. 6). Champion devices: A mixed solvent of THF:CB (16:1 v/v) was used to disperse PeNCs to obtain a uniform colloidal solution. A programmable spin-casting was developed for deposition of PeNCs film, which consists of two steps: step 1 was chosen as 400 rpm for 3 s; and step 2 was 1,500 rpm for 60 s. PeLEDs characterization. With a nearly ideal Lambertian emission, the voltage independent EL spectra of the PeLEDs were measured using a Photo Research Spectra Scan spectrometer PR705 at different brightness, while the luminance-current-voltage characteristics of the devices were measured simultaneously with a programmable Agilent B2902A sourcemeter and a Konica-Minolta LS-110 luminance meter in air at room temperature (Supplementary Fig. 5b). The lifetime test was carried out by a programmable sourcemeter of Agilent B2902A and a calibrated Si-photodetector (Thorlabs, FDS- 15 1010CAL). PL of electron single carrier device: PL spectra were recorded through a micro-Raman spectrometer (Horiba, LABRAM HR800) with a single-grating setup and a 442 nm solid state laser were used as the excitation laser. The bias was applied by Agilent B1500A semiconductor device parameter analyzer. Acknowledgements This research is supported by the National Research Foundation, Prime Minister's Office, Singapore under its competitive Research Programme (CRP Award No. NRF-CRP14-2014-03). H.V.D. gratefully acknowledges the financial support from NRF Investigatorship grant NRF-NRFI2016-08 and additional support from TUBA. Q.H.X. acknowledges financial support from Singapore National Foundation via an Investigatorship Award (NRF-NRFI2015-03), and Singapore Ministry of Education through grants (MOE2015-T2-1-047 and MOE2015-T1-001-175). G.H.X. acknowledges financial support from the Science and Technology Development Fund from Macau SAR (FDCT-116/2016/A3) and Start-up Research Grant (SRG2016-00087-FST) from Research & Development Office at University of Macau. Author contributions H.V.D., Q.H.X., and E.H.S. planned and guided the work. F.Y. designed and fabricated the devices, performed the device experiments, analyzed the data with assistance of S.T.T., and discussed the results with H.V.D. J.X designed and synthesized the perovskite materials and performed the material experiments and data analyses with assistance of J.X.Z, L.L.Z. and L.N.Q., and discussed the results with Q.H.X. The luminescence lifetime measurement made by R. S. The luminescence dynamics study was performed by G.C.X. The TEM measurement was made by Y.W.Z. The UPS measurement was made by S.C. and A. H. All authors discussed and commented on the manuscript. 16 Additional information Supplementary information is available in the online version of the paper. Reprints and permissions information is available online at www.nature.com/reprints. Correspondence and requests for materials should be addressed to H. V. D., Q. H. X., E.H.S. Competing financial interests The authors declare no competing financial interests. References 1. Lee, M. M., Teuscher, J., Miyasaka, T., Murakami, T. N. & Snaith, H. J. Efficient hybrid solar cells based on meso-superstructured organometal halide perovskites. Science 338, 643–647 (2012). 2. Heo, J. H. et al. Efficient inorganic-organic hybrid heterojunction solar cells containing perovskite compound and polymeric hole conductors. Nat. Photon. 7, 486–491 (2013). 3. Grätzel, M. The light and shade of perovskite solar cells. Nat. Mater. 13, 838–842 (2014). 4. Zhou, H. et al. Interface engineering of highly efficient perovskite solar cells. Science 345, 542– 546 (2014). 5. Liu, D. & Kelly, T. L. Perovskite solar cells with a planar heterojunction structure prepared using room-temperature solution processing techniques. Nat. Photon. 8, 133–138 (2014). 6. Kazim, S., Nazeeruddin, M. K., Grätzel, M. & Ahmad, S. Perovskite as light harvester: a game changer in photovoltaics. Angew. Chem. Int. Ed. 53, 2812–2824 (2014). 7. Saidaminov, M. I. et al. Planar-integrated single-crystalline perovskite photodetectors. Nat. Commun. 6, 8724 (2015). 8. Jeon, N. J. et al. Compositional engineering of perovskite materials for high-performance solar cells. Nature 517, 476–480 (2015). 9. Stranks, S. D. & Snaith, H. J. Metal-halide perovskites for photovoltaic and light-emitting devices. Nat. Nanotech. 10, 391–402 (2015). 17 10. Chin, X. Y., Cortecchia, D., Yin, J., Bruno, A. & Soci, C. Lead iodide perovskite light-emitting field-effect transistor. Nat. Commun. 6, 7383 (2015). 11. Xing, G. et al. Low-temperature solution-processed wavelength-tunable perovskites for lasing. Nat. Mater. 13, 476–480 (2014). 12. Zhu, H. M. et al. Lead halide perovskite nanowire lasers with low lasing thresholds and high quality factors. Nat. Mater. 14, 636–642 (2015). 13. Xiao, Z. et al. Efficient perovskite light-emitting diodes featuring nanometre-sized crystallites. Nat. Photon. 11, 108–115 (2017). 14. Tan, Z. et al. Bright light-emitting diodes based on organometal halide perovskite. Nat. Nanotech. 9, 687–692 (2014). 15. Wang, J. et al. Interfacial control toward efficient and low-voltage perovskite light-emitting diodes. Adv. Mater. 27, 2311–2316 (2015). 16. Cho, H. et al. Overcoming the electroluminescence efficiency limitations of perovskite light- emitting diodes. Science 350, 1222–1225 (2015). 17. Li, G. et al. Efficient light-emitting diodes based on nanocrystalline perovskite in a dielectric polymer matrix. Nano Lett. 15, 2640–2644 (2015). 18. Wang, N. et al. Perovskite light-emitting diodes based on solution-processed self-organized multiple quantum wells. Nat. Photon. 10, 699–704 (2016). 19. Yuan, M. et al. Perovskite energy funnels for efficient light-emitting diodes. Nat. Nanotech. 11, 872–877 (2016). 20. Xing, J. et al. High-efficiency light-emitting diodes of organometal halide perovskite amorphous nanoparticles. ACS Nano 10, 6623–6630 (2016). 21. Byun, J. et al. Efficient visible quasi-2D perovskite light-emitting diodes. Adv. Mater. 28, 7515– 7520 (2016). 22. Yang, Y. et al. Observation of a hot-phonon bottleneck in lead-iodide perovskites. Nat. Photon. 10, 53–59 (2016). 18 23. Saba, M. et al. Correlated electron-hole plasma in organometal perovskites. Nat. Commun. 5, 5049 (2014). 24. Johnston, M. B. & Herz, L. M. Hybrid perovskites for photovoltaics: charge-carrier recombination, diffusion, and radiative efficiencies. Acc. Chem. Res. 49, 146–154 (2016). 25. Huang, H. et al. Colloidal lead halide perovskite nanocrystals: synthesis, optical properties and applications. NPG Asia Mater. 8, e328 (2016). 26. Kim, Y. H. et al. Highly efficient light-emitting diodes of colloidal metal–halide perovskite nanocrystals beyond quantum size. ACS Nano,11, 6586–6593 (2017). 27. Herz, L. M. Charge-carrier dynamics in organic-inorganic metal halide perovskites. Annu. Rev. Phys. Chem. 67, 65–89. (2016). 28. Wehrenfennig, C., Eperon, G. E., Johnston, M. B., Snaith, H. J. & Herz, L. M. High charge carrier mobilities and lifetimes in organolead trihalide perovskites. Adv. Mater. 26, 1584–1589 (2014). 29. Savenije, T. J. et al. Thermally activated exciton dissociation and recombination control the carrier dynamics in organometal halide perovskite. J. Phys. Chem. Lett. 5, 2189–2194 (2014). 30. Wehrenfennig, C., Liu, M. Z., Snaith, H. J., Johnston, M. B. & Herz, L. M. Charge-carrier dynamics in vapour-deposited films of the organolead halide perovskite CH3NH3PbI3−xClx. Energy Environ. Sci. 7, 2269–2275 (2014). 31. Noel, N. K. et al. Enhanced photoluminescence and solar cell performance via Lewis base passivation of organic-inorganic lead halide perovskites. ACS Nano 8, 9815–9821 (2014). 32. Stoumpos, C. C., Malliakas, C. D. & Kanatzidis, M. G. Semiconducting tin and lead iodide perovskites with organic cations: phase transitions, high mobilities, and near-infrared photoluminescent properties. Inorg. Chem. 52, 9019–9038 (2013). 33. Lee, J. H. & Kim, J. J. Electron injection and transport for high performance inverted organic light-emitting diodes. J. Inf. Display 14, 39–48 (2013). 34. Sasabe, H. et al. Influence of substituted pyridine rings on physical properties and electron mobilities of 2-methylpyrimidine skeleton-based electron transporters. Adv. Funct. Mater. 21, 19 336–342 (2011). 35. Li, C., Duan, L., Sun, Y., Li, H. & Qiu, Y. Charge transport in mixed organic disorder semiconductors: trapping, scattering, and effective energetic disorder. J. Phys. Chem. C 116, 19748–19754 (2012). 36. Brütting, W. Physics of Organic Semiconductors. Ch. 10 (WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim, 2005). 37. Malinkiewicz, O. et al. Perovskite solar cells employing organic charge-transport layers. Nat. Photon. 8, 128–132 (2014). 20 Figure Caption Figure 1. Luminescence study and synthesis of CH3NH3PbBr3 emitters. a, The calculated injected carrier density dependence of luminescence quantum yield for excitonic emission (= , = , = ) and bimolecular emission (= , = , = ) in different dimensional CH3NH3PbBr3 emitters. b, Photon- injected charge carrier density dependence of the initial time PL intensity (IPL(t = 0)). The quadratic behavior IPL[t=0] confirms the bimolecular recombination in bulk perovskite. c, Relative PLQY as a function of the photon-injected charge carrier density. d, PL spectra of colloidal PeNCs solution aging for 6 days. Peak emission wavelengths redshift: 506.5, 512.5, 514.6, 516.4, 517.8, 518.5 nm. The FWHM decreases: 32.7, 25.7, 24.9, 24.8, 24.7, 24.7 nm. e, XRD pattern migration of PeNCs in 6 days. f, TEM and high-resolution TEM image of the as-synthesized PeNCs. Scale bars: 50 and 2 nm. d (200) = 2.93 Å, d (210) = 2.65 Å. Figure 2. Device structure and L-I-V curves of devices. a, Device structure with energy level alignment. b, SEM of device cross-section showing the multi-layer structure with distinct contrast. Scale bar: 100 nm. c, I-V curves of electron single carrier devices with different electron transport layers. d, Normalized PL spectrum of PeNCs film and normalized EL spectra of champion PeLEDs at different brightness levels from 500 to 15,000 cd m-2 around. The peak emission wavelength is 524 nm and the FWHM is 22 nm. e, I-V curves of all PeLEDs. f, Luminance-Voltage dependences of devices. Figure 3. Comparison of devices. a, EQE-current density characteristics. b, EQE-voltage dependences. c, Power efficiency-luminance correlations. d, The EQEmax histogram for 15 champion devices from 4 batches. e, The dependencies between EQEmax and the corresponding driving current density (scatters: experiment result; solid line: simulation by linear fitting). f, The PL intensity of 21 MAPbBr3 layer in an electron single carrier device. no is the density of electron injected into PeNCs layer when the current density is 1.0 × 10-2 mA cm-2 (scatters: experiment result; solid line: simulation by linear fitting). Table 1. Key parameter comparison of PeLEDs. Device Structure Von (V) Lmax (cd m-2) EQEmax (%) ηmax (lm W-1) Control device I Control device II Control device III Champion device ITO / PEDOT:PSS (40 nm) / MAPbBr3 PeNCs (40 3.3 nm) / TPBi (30 nm) / TPBi: Cs2CO3 (10:1 w/w, 20 nm) 3.4 / Al (100 nm). ITO / PEDOT:PSS (40 nm) / MAPbBr3 PeNCs (40 nm) / B3PYMPM (30 nm) / B3PYMPM: Cs2CO3 (10:1 w/w, 20 nm) / Al (100 nm). ITO / PEDOT:PSS (40 nm) / MAPbBr3 PeNCs (40 nm) / B3PYMPM:TPBi (1:2 w/w, 30 nm) / B3PYMPM: Cs2CO3 (10:1 w/w, 20 nm) / Al (100 nm). ITO / PEDOT:PSS (40 nm)/ MAPbBr3 PeNCs (80 nm) / B3PYMPM:TPBi (1:2 w/w, 30 nm) / B3PYMPM: Cs2CO3 (10:1 w/w, 20 nm) / Al (100 nm). 3.6 2.7 2.6 2.7 3.0 2.9 3.0 3.1 3.0 2.9 15,130 16,470 14,190 24,410 23,290 23,120 33,570 30,380 32,730 22,830 43,400 25,410 5.11 5.06 4.81 2.39 2.28 2.19 7.91 6.75 7.25 12.9 7.53 10.67 10.82 7.03 7.49 5.96 6.21 5.62 20.18 17.92 18.47 26.07 17.56 30.3 Von: turn on voltage @ 1 cd m-2; Lmax: maximum luminance; EQEmax: maximum external quantum efficiency; ηmax: maximum power efficiency. 22 1.0 0.8 0.6 0.4 0.2 Excitonic emission LED working range Bimolecular emission b ) . u . a ( ) 0 = y x1.1 ± 0.1 t ( L P I c ) d e z i l a m r o N ( Y Q L P e v i t a e R l PeNCs Film Bulk Crystal 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 y x2.1 ± 0.1 PeNCs Film Bulk Crystal 0.0 1011 1.0 0.8 0.6 0.4 0.2 0.0 1015 1013 Charge Density (cm-3) 1017 1019 1021 0 day 6 days 480 500 Wavelength (nm) 520 540 e ) . u . a ( y t i s n e t n I 10 560 a ) . u . a ( y c n e c i f f i E e v i t i a d a R d ) . u . a ( L P d e z i l a m r o N Figure 1 1019 1015 1016 1017 Charge Density (cm-3) ) 0 1 1 ( ) 0 0 1 ( ) 0 0 2 ( ) 0 1 2 ( ) 1 1 2 ( 1015 1016 1018 Charge Density (cm-3 ) 1017 1018 1014 f 6 days 2 days 0 day 20 40 2 Theta (degree) 30 50 23 b 3 2 O C s C -2.9 Al -4.1 O T I -4.7 : S S P T O D E P -5.1 -3.4 s C N e P -5.8 -2.8 i B P T -6.2 -3.4 M P M Y P 3 B -6.9 c ) 2 - m c A m ( y t i s n e D t n e r r u C 1000 800 600 400 200 0 0 2 4 TPBi:B3PYMPM (1:0 w/w) TPBi:B3PYMPM (2:1 w/w) TPBi:B3PYMPM (1:2 w/w) TPBi:B3PYMPM (0:1 w/w) 10 12 14 16 8 6 Voltage (V) ) f 10000 1000 100 10 1 0.1 0.01 m d c ( e c n a n m u L 2 - i Control Device I Control Device II Control Device III Champion Device 2 3 4 5 6 7 8 9 Voltage (V) a ) V e ( E 0 -1 -2 -3 -4 -5 -6 -7 -8 d ) . u . a ( y t i s n e t n I 1.0 0.8 0.6 0.4 0.2 e ) 2 - m c A m ( y t i s n e D t n e r r u C 1000 100 10 1 0.1 0.01 1E-3 1E-4 1E-5 Control Device I Control Device II Control Device III Champion Device 0 2 4 6 Voltage (V) 8 10 24 0.0 450 475 500 525 550 575 600 Wavelength (nm) Figure 2 Control Device I Control Device II Control Device III Champion Device a ) % ( E Q E 16 14 12 10 8 6 4 2 0 1E-3 0.01 0.1 1 10 100 d i e c v e D f o r e b m u N Current Density (mA cm-2) 6 5 4 3 2 1 0 6 7 9 8 10 Max EQE (%) 11 12 b ) % ( E Q E e ) % ( x a m E Q E 14 12 10 8 6 4 2 0 14 12 10 8 6 4 2 0 Control Device I Control Device II Control Device III Champion Device 0 1 2 3 4 5 6 7 8 9 Voltage (V) Slope= - 7.16958 Control Device I Control Device II Control Device III Champion Device 10 1 Luminance (cd m-2) 100 1000 10000 Slope = -7765.050 c ) 1 - W m l ( y c n e c i f f i E r e w o P 30 25 20 15 10 5 0 0.1 f ) . u . a ( y t i s n e t n I L P 10 Current Density (mA cm-2) 100 1 10 1000 10000 100 n/n0 Figure 3 25
1901.06927	1	1901	2019-01-21T13:29:55	Laser-based underwater transfer of radio-frequency clock signal with electronic phase compensation	[ "physics.app-ph", "physics.ins-det" ]	We demonstrate a laser-based underwater transfer of radio frequency clock signal with a phase compensation technique. With this frequency transfer scheme, a 100 MHz signal has been transferred over 5 m underwater link. Timing jitter power spectral density, timing fluctuations and instabilities were all measured to evaluate the quality of the transferred frequency signal. The experimental results show the root-mean-square (RMS) timing fluctuations of the transferred frequency signal over 5 m underwater link with and without phase compensation are 2.1 ps and 9.6 ps within 5000 s, respectively. The relative fractional frequency instabilities are on the order of 5E-13 at 1 s and 7E-16 at 1000 s for phase-compensated link, and 2E-12 at 1 s and 7E-15 at 1000 sfor uncompensated link, respectively. The proposed frequency transfer scheme has a potential application that disseminating Cs and H-master clocks signal over the underwater transmission link since the stability of the transmission link is superior to these atomic clocks.	physics.app-ph	physics	Laser-based underwater transfer of radio-frequency clock signal with electronic phase compensation DONG HOU,1,* GUANGKUN GUO,1 DANIAN ZHANG,1 AND FUYU SUN2 1Time & Frequency Research Center, School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China 2National Time Service Center, Chinese Academy of Science, Xi'an, Shanxi 710600, China *Corresponding author: [email protected] Received 12 January 2019; revised XX Month, 2019; accepted XX Month 2019; posted XX Month 2019 (Doc. ID XXXXX); published XX Month 2019 We demonstrate a laser-based underwater transfer of radio frequency clock signal with a phase compensation technique. With this frequency transfer scheme, a 100 MHz signal has been transferred over 5 m underwater link. Timing jitter power spectral density, timing fluctuations and instabilities were all measured to evaluate the quality of the transferred frequency signal. The experimental results show the root-mean-square (RMS) timing fluctuations of the transferred frequency signal over 5 m underwater link with and without phase compensation are 2.1 ps and 9.6 ps within 5000 s, respectively. The relative fractional frequency instabilities are on the order of 5 × 10−13 at 1 s and 7 × 10−16 at 1000 s for phase-compensated link, and 2 × 10−12 at 1 s and 7 × 10−15 at 1000 s for uncompensated link, respectively. The proposed frequency transfer scheme has a potential application that disseminating Cs and H-master clocks signal over the underwater transmission link since the stability of the transmission link is superior to these atomic clocks. © 2018 Optical Society of America OCIS codes: (060.2605) Free-space optical communication; (010.4455) Oceanic propagation; (120.0120) Instrumentation, measurement, and metrology; (270.2500) Fluctuations, relaxations, and noise. 1. INTRODUCTION http://dx.doi.org/10.1364/PRxxxxxx In the last decades, the underwater wireless communication has attracted a remarkable attention as it is capable of providing a higher flexibility and cost-effectiveness than submarine optical fiber communication which can only be used in the data transmission between the static objects in the water [1-3]. Up to now, the underwater wireless communication has been widely used in the data transmission between various underwater vehicles, sensors and observatories [3, 4]. Generally, the traditional wireless underwater communication includes, underwater acoustic and microwave methods [5-8]. Although the current available acoustic scheme which uses acoustic waves for underwater communication can reach a long distances (ranging larger than few hundred meters), it is still limited by the low bandwidth, high transmission losses, time varying multipath propagation, and high latency [5]. The transmission of microwave have a high-bandwidth and high propagation speed. However, the microwave signal is highly attenuated by the water, which significantly limits the distance of the underwater communication (ranging only few meters) [8]. To solve the problem that high-bandwidth and long- distance are both required in the underwater wireless communication, another method of transferring optical signal over underwater link was proposed [3, 9]. In this underwater optical wireless communication, the optical signal is able to be transferred as far as tens or hundred meters. In the past years, the interest towards optical wireless transmission over underwater link has increased as it is capable of providing high data rates with low power over long-distance underwater link. Some prior works show that the high-speed wireless underwater transmission links with blue and green visible lasers have been achieved [10-17]. As the development of the wireless underwater communication, distribution of timing and frequency signal between the underwater objects becomes more and more important, which has many potential applications in underwater metrology [3, 18, 19], for example, submersible synchronization, underwater navigation, underwater sensing, and etc. The timing and frequency signals has been able to be transferred over atmospheric link. Many frequency dissemination experiments show that atmospheric transfers of optical and radio frequency signals with ultra-low timing deviation have been achieved, where phase compensation schemes were proposed to suppress the turbulence-affected timing fluctuation over free-space optical link [20- 24]. However, to the best of our knowledge, the transfer of timing and frequency signal over underwater link has not be reported yet. In this paper, a laser-based underwater transfer of a radio- frequency signal with a phase compensation scheme is demonstrated. In our underwater frequency transfer experiment, we disseminated a 100 MHz signal over a 5 m underwater link. The experimental results show that the RMS timing fluctuations of the transferred frequency signals with and without phase compensation are 2.1 ps and 9.6 ps in 5000 s, respectively; and the relative fractional frequency instabilities of the transmission links with and without phase compensation are on the order of 5 × 10-13 at 1 s and 7 × 10-16 at 1000 s, and the order of 2 × 10-12 at 1 s and order of 7 × 10-15 at 1000 s respectively. 2. SCHEMATIC OF UNDERWATER FREQUENCY TRANSFER WITH PHASE COMENSATION In order to transfer a radio-frequency signal from a transmitter to a receiver via an optical carrier over underwater link, the most convenient scheme includes three steps [3], which are directly loading the radio-frequency signal onto the optical carrier, then transferring the light to remote receiver via an direct underwater optical link, and lastly recovering this radio-frequency signal by a photodetector. This process is almost the same as the underwater optical wireless communication, where the optical light with encoded data is directly transmitted from transmitter to receiver via a visible underwater link. In this case, the bit error rate of the decoded data is determined by the signal-to-noise ratio (SNR) of the transmitted signal, and just slightly influenced by the timing fluctuation or drift. However, in a frequency transfer system, the stability of the transmitted radio-frequency signal is directly determined by the timing fluctuation and drift of the optical signal. The three steps of the underwater transmission of optical signal presented above could introduce excess timing fluctuation and drift into the signal, resulting in the stability degradation of the original radio-frequency signal. However, the strong timing fluctuation and drift introduced by the water turbulence and temperature drift is much higher than that introduced by laser source and the photo-detection electronics. Therefore, for an underwater transfer of a radio-frequency signal in the water environment, the timing fluctuations and frequency instability of the transmission link is dominantly caused by the water turbulence and temperature drift rather than on the transmitter and receiver [25-27]. Literatures reveal that the water turbulence induces variation of the refraction index [28], which could directly lead to excess phase noise or timing fluctuation in the transmitted frequency signal. Here, the spectrum density of underwater optical turbulence can be expressed as (cid:2004)(cid:3041)(cid:3038)(cid:4666)(cid:2018)(cid:4667)= (cid:1837)(cid:2871)(cid:2018)(cid:2879)(cid:2869)(cid:2869)/(cid:2871)[28], where K3=Χε-1/3 (Χ expresses the strength of temperature gradient and ε is the kinetic energy dissipation rate) is the constant that determines the turbulence strength. For underwater conditions, the value of K3 ranges from 10-14 to 10-8 m-2/3. Note that, κ is the scalar spatial frequency (in rad/m). Therefore, the longer the transmission distance is, the higher the excess timing fluctuation of transmitted frequency signal is introduced. In order to improve the quality of the frequency transfer significantly, the timing fluctuations should be suppressed by a phase compensation technique. In this paper, we propose a scheme of laser-based underwater frequency transfer with a low timing fluctuation using a phase compensation technique. Figure 1 shows the schematic of our laser-based underwater radio- frequency transfer with an electronic phase compensation technique. The optical carrier is provided by a continuous wave (CW) 520 nm diode laser. A highly-stable frequency source, is phase-shifted by a voltage-controlled phase shifter first, and then directly loaded onto the optical carrier via a current amplitude-modulation (AM) scheme. The modulated laser beam is launched into the water from the transmitter. On the receiver, half of the beam is reflected by a half-mirror, and the remaining beam is converted to a radio-frequency signal via a high- speed photodiode for users. The returning beam which transmits the same optical path along the forward underwater transmission link, is detected by another high-speed photodiode on the transmitter. The detected microwave signal from the photodiode is phase-shifted by an identical electronic phase shifter, and then compared with the reference source signal on a phase detector, to generate an error signal which includes the information of timing fluctuation affected by water turbulence. This error signal is fed-back to the two identical phase shifters via a (proportional-integral) PI servo controller, to adjust the phase delay on each phase shifter. When the servo loop is closed, the timing fluctuations affected by the water turbulence can be corrected by the phase compensation technique. Note that, the phase shifters produce the same phase delays, because they are controlled by the same error signal simultaneously. The mechanism of the phase compensation will be explained in detail below. the receiver signal the returned microwave Fig. 1. Schematic of the underwater frequency transfer with an electronic phase compensation technique. PS: phase shifter, PD: photodiode, HM: half-mirror, PI: Proportion-Integration controller. As shown in Fig. 1, we assume that the frequency source generates a radio-frequency signal with an initial phase φ0. On the transmitter, this signal is phase-shifted with φc via a phase shifter first, and then delivered to the receiver over an underwater link. Assuming the water turbulence introduces a phase fluctuation φp to the transmitted signal over the one-trip transmission link, the total phase delay of the recovered microwave signal on is given by φtotal=φ0+φc+φp. With a half-mirror, half of beam is reflected to the transmitter along the almost same optical path, which will introduces a same turbulence-affected phase fluctuation φp. In this case, the returning microwave signal detected by the photodiode has a phase delay φ0+φc+2φp due to the twice turbulence effect. After passing through an identical phase-shifter, another phase delay φc is introduced to the returned microwave signal. Therefore, the phase delay of is given by φreturned=φ0+2φc+2φp. After eliminating the initial phase φ0 by comparing it with the reference signal, we get an intermediate frequency signal with the phase error information 2(φc+φp). Here, this error signal is used to control the two identical phase shifters via a PI servo controller, for compensating the turbulence-affected phase φp. When the servo loop is closed, the phase error 2(φc+φp) equals zero, and consequently, the timing fluctuation affected by water turbulence for the recovered radio-frequency signal on receiver will be corrected as φc = -φp. With this electronic phase compensation technique, the timing fluctuation performance and stability of our underwater frequency transfer will be improved significantly, compared to the direct transmission link without phase compensation Figure 2 shows the experimental setup of our underwater frequency transfer system using a visible diode laser with the electronic phase compensation. This experiment setup was built in a windows-opened laboratorial room. Before the experiment was conducted, we built a water tank with a length of 3.1 m, a width of 0.35 m, and a height of 0.35 m. The water tank was fixed on a sturdy and flat desk, and then we filled it with fresh water. On the transmitter, the optical carrier is provided by a commercial 520 nm green diode laser with an output power of 30 mW, which has a much lower attenuation than red and infrared lasers in water. A 100 MHz microwave signal with a power of 20 mW, generated from a stable signal source (Agilent, E4421B), is directly loaded onto the laser with an amplitude modulation (AM) method. The laser beam is coupled to a collimator, and then launched into the water tank directly. At 2.5 m far away remote site, the beam was sent back to the receiver by a silver-coated two-inch mirror (M1), which forms a total 5 m one-way transmission link in water. 3. EXPERIMENTAL SETUP 4. EXPERIMENTAL RESULTS AND DISSCUSSION Fig. 2. Experimental setup of the underwater frequency transfer with an electronic phase compensation technique. PS: phase shifter, CW: Continuous-wave, PHD: phase detector, PI: proportion-Integration controller, AMP: amplifier. On the receiver, half of the transmitted beam is reflected to transmitter along the same optical path by a half-mirror (M2). The reflected beam with power of about 1 mW is collected by The frequency transfer experiment was conducted in a normal another collimator and tightly focused onto a high-speed Si day. In the experiment, the two collimator were placed as close photodiode on the transmitter. The converted 100 MHz radio- as possible on the transmitter to obtain an almost identical frequency signal with twice timing fluctuations was amplified, turbulence effect. The real photo of the underwater frequency phase shifted, and then mixed with the RF reference source transfer experiment is shown in Fig.3. Since the phase signal to generate the phase error signal. This error signal is compensation technique can suppress the additional timing then fed-backed to the two identical phase shifters (Mini- fluctuations caused by water turbulence, we believe the quality circuit, JSPHS-150) via a fast PI servo controller, to compensate of the transferred 100 MHz frequency signal transfer should be the timing fluctuations affected by the water turbulence. On the improved significantly compared to the test without phase receiver, with another high-speed Si photodiode, the remaining compensation. Here, we measured the timing jitter power half of beam with power of about 6 mW is also converted to a spectral density (PSD) and timing fluctuation of the transferred radio-frequency signal. This recovered signal, is also amplified, 100 MHz radio-frequency signals with and without phase filtered and mixed with the frequency reference source, to compensation. produce a DC output. After low pass filtering, the DC signal is delivered to a digital voltage meter (Keysight, 34461A) and a FFT analyzer (HP, HP35665), for estimating the quality of the frequency transmission. Note that, our experimental setup is in an open-window laboratorial environment, and the water tank is not covered. Therefore, in this setup, the water turbulence and temperature drift has an obvious effect on the phase of the transmission signal. In addition, we collected the optical light on the photodetectors as much as possible to acquire the highest signal-to-noise ratio (SNR). Real photo of experiment setup for underwater Fig. 3. frequency transfer. Fig. 4. Timing jitter (PSD) results for underwater frequency transfer. Curve (i): Without phase compensation. Curve (ii): With phase compensation. Curve (iii): Result for short link as a measurement floor The timing jitter PSD results are shown in Fig. 4. Curve (i) and (ii) are the timing jitter PSD data of excess timing jitter of the transferred 100 MHz microwave signal without and with phase compensation, respectively. Curve (iii) shows the background noise floor of the short link (shown Fig. 2), where the modulated light was directly delivered from transmitter to rewefreflufreco eceiver, and not e can see the tim equency. This uctuation affect equency, and ompensation tec t transmitted in ming jitter PSD indicates the ted by water t it can be chnique. n water. From C is suppressed b e most of sh urbulence is b suppressed w Curve (i) and (i ii), set ng set ase below 5 Hz offs hort-term timin elow 5 Hz offs with our pha g. 5. Timing fluc FigCucoflo urve (i): Withou ompensation. Cur oor. Sample rate i The timing flu urve was recor ming fluctuatio ecovered from nd its RMS timi urve (ii) shows equency signal ith phase comp bout 2.1 ps with uctuation of th easurement flo as not transm ming fluctuatio ectronic noise ming fluctuatio pproximately 1. etween the tr ompensation, ompensation te ming fluctuation The instability me-frequency alculated the Al stribution link. ansferred radi llan Deviation r ompensation, w g. 5. From Cu equency transm stability of 2 × the relative Al ith phase com frequency transf fer. ase ent for underwater f ctuation results f (ii): With pha ensation. Curve ut phase comp k as a measureme ult for a short link rve (iii): The resu d for all curves s 1 point/second n Fig. 5, and ea ts are shown in ctuation result ch he nal on, ve (i) shows th seconds. Curv rded for 5000 frequency sign 0 MHz radio-f on of the 100 se compensatio k without phas the direct link s. ps within 5000 roximately 9.6 p ng drift is appr e 100 MHz radi ctuation of the s the timing flu io-nk to ng he ght his nd he is on ase ase suppressed lin om the timing l recovered fro drift is reduced ts RMS timing d pensation, and it asured the timin re, we also mea hin 5000 s. Her short link as th ansfer with a s e frequency tr the optical lig Fig. 2), where oor (shown in itted in the un k. Therefore, th nderwater link ibuted by the on is just attri laser noise an ic system. Curv ve (iii) shows th of our photoni MS timing drift link, and its RM n of the short 000 s. Based on n the compariso .3 ps within 50 nks with and ransmission lin without pha the p that proposed pha we believe ively suppresse ed the long-ter echnique effecti rm r a so cy he ve ase n. of the transmis ry important for ssion link is ver system. Ther distribution efore, we al s for the under rwater frequen llan Deviations ws the instabilit . Figure 6 show ty results for th ) is the relativ gnal. Curve (i) o-frequency si ransferred sign result for the tr nal without pha in ted out from the which is calculat e sampled data ters underwat ows the 5 me urve (i), it sho ter s a thout phase com mission link wit mpensation has 1000 s. Curve ( (ii) nal ers d 7 × 10-15 for 1 10-12 for 1 s an ransferred sign result for the tr llan Deviation r and mpensation, a it shows the 5 mete cutimreanCufrewiabflumwatimeletimapbecocotimtimcaditraAlcoFifreinis wi underw with phase link w transmis frequen water ncy ssion and 7 × 10-16 × 10-13 for 1 s a nstability of 5 × nsation has a in compen we find that for 100 00 s. With com mparison of Curv ve (i) and (ii), transferred underwater insta lo the ong-term ability of the ced for about frequen ncy signal with h phase compen nsation is reduc magnitude at half or rder of magnitu ude at 1 s and one order of m ) shows the 1000 s s due to the ti iming suppress sion. Curve (iii link without ned via a short measur rement floor, w which was obtai at, this curve Fig. 2). Note tha smission (see F any un nderwater trans urement, and is mere ely the lower b bound of the ins stability measu urce and the it is jus st limited by the e stability of th he frequency so r turbulence electro onic noise. This s is because the e most of water his proposed effect w was cancelled. The instability achieved by th phase system w with r transfer ency freque water underw nsation may b short-range compen e quite adequa ate with some underw water time-fre equency distri ibution applic cations. For exampl le, with compar rison with insta ability of our tra ansfer result A) [29] or H-m s clock (5071A and a commercial Cs master clock 2010, standar rd version) [3 30], we can fi (MHM nd that the ility of our tran nsmission link i is lower than th he Cs and H- instabi master r clocks. This in ndicates that dis sseminating a C Cs or H clock the signals s in water e environment by using e proposed r is feasible. underw me in this pape y transfer schem water frequency Fig. 6. NCLUSIONS Instability res Relative e Allan deviation and refe ference signal wi deviatio on with phase com as meas surement floor. 5. CONWe hav ve demonstrate over an n underwater li The RM MS timing fluct over a 5-m long unde measur red to be appr fraction nal frequency i and of f7 × 10-17 at 1 transm mission link with to atom mic clocks, for implies s the atomic clo water link in a underw underw water transmiss this is just limited by sults for underw n between trans ithout phase com mpensation; (iii) ed a laser-base ink with a phas tuation for a 10 rwater link wit roximately 2.1 instability on th 1000 s. The ins h the distance r example, Cs o ock is able to b short-range. Al sion link in our y the power of water frequency sferred radio-freq mpensation; (ii) R Allan deviation f ed radio-freque se compensatio 00 MHz freque th phase compe ps within 50 he order of 5 × stability of the of a few meter or H-master cl be directly distr lthough the dis r experiment is the laser and y transfer, (i) quency signal Relative Allan for a short link ency transfer on technique. ency transfer ensation was 00 s with a × 10-13 at 1 s underwater s is superior locks, which ributed over stance of the few meters, the physical size of the water tank. In the future, a lower timing fluctuation and longer distance underwater frequency transfer will be achieved by loading a microwave signal with higher frequency and using high power laser source. Funding Information. National Natural Science Foundation of China (NSFC) (61871084, 61601084); References 1. M. N. Kojima, T, Yabuta, Y. Negishi, K. Iwabuchi, O. Kawata, K. Yamashita, Y. Miyajima, and N. Yoshizawa, "Submarine optical fiber cable: development and laying results," Appl. Opt. 21, 815 -- 821 (1982). 2. B. Nyman, "Flexibility in Submarine Fiber Optic Networks," J. Opt. Commun. Netw. 7, 553 -- 557, (2015). 3. H. Kaushal and G. Kaddoum, "Underwater Optical Wireless Communication," IEEE ACCESS 4, 1518 -- 1547, (2016). 4. S. Arnon, "Underwater optical wireless communication network," Opt. Eng. 49, 015001, (2010). 5. M. Stojanovic, "Recent advances in high-speed underwater acoustic communications," IEEE J. Oceanic Eng. 21, 125 -- 136, (1996). 7. A. 6. M. Stojanovic and J. Preisig, "Underwater Acoustic Communication Channels: Propagation Models and Statistical Characterization," IEEE Commun. Mag., 47, 84 -- 89, (2009). I. Al-Shamma'a, A. Shaw, and S. Saman, "Propagation of electromagnetic waves at MHz frequencies through seawater," IEEE Trans. Antennas Propagat. 52, 2843 -- 2849, (2004). 8. X. Che, I. Wells, G. Dickers, P. Kear, and X. Gong, "Re-Evaluation of RF Electromagnetic Communication in Underwater Sensor Networks," IEEE Commun. Mag. 48, 143 -- 151, (2010). 9. L. J. Johnson, F. Jasman, R. J. Green, and M. S. Leeson, "Recent advances in underwater optical wireless communications," Underwater Technol. 32, 167 -- 175, (2014). 10. L. J. Mullen, A. J. Vieira, P. R. Herczfeld, and V. M. Contarino, "Application of radar technology to aerial lidar systems for enhancement of shallow underwater target detection," IEEE Trans. Microwave Theory Tech. 43, 2370 -- 2377, (1995). 11. J. B. Snow, J. P. Flatley, D. E. Freeman, M. A. Landry, C. E. Lindstrom, J. R. Longacre, and J. A. Schwartz et al., "Underwater propagation of high- data-rate laser communications pulses," Proc. SPIE, 1750, 419 -- 427, (1992). 12. F. Hanson and S. Radic, "High bandwidth underwater optical communication," Appl. Opt. 47, 277 -- 283, (2008). 13. B. Cochenour, L. Mullen, and J. Muth, "Modulated pulse laser with pseudorandom coding capabilities for underwater ranging, detection, and imaging," Appl. Opt. 50, 6168 -- 6178, (2011). 14. F. Pellen, V. Jezequel, G. Zion, and B. L. Jeune, "Detection of an underwater target through modulated lidar experiments at grazing incidence in a deep wave basin," Appl. Opt. 51, 7690 -- 7700, (2012). 15. B. Cochenour, L. Mullen, and J. Muth, "Temporal Response of the Underwater Optical Channel for High-Bandwidth Wireless Laser Communications," IEEE J Oceanic Eng. 38 730 -- 742, (2013). 16. X. Liu, S. Yi, X. Zhou, Z. Fang, Z. Qiu, L. Hu, C. Cong, L. Zheng, R. Liu, and P. Tian, "34.5 m underwater optical wireless communication with 2.70 Gbps data rate based on a green laser diode with NRZ-OOK modulation," Opt. Express 25, 27937 -- 27947, (2017). 17. X. Liu, S. Yi, X. Zhou, S. Zhang, Z. Fang, Z. Qiu, L. Hu, C. Cong, L. Zheng, and R. Liu, "Laser-based white-light source for high-speed underwater wireless optical communication and high-efficiency underwater solid- state lighting," Opt. Express 26, 19259 -- 19274, (2018). 18. J.Liu, Z. Zhou, Z. Peng, J. Cui, M. Zuba, and L. Fiondella, "Mobi-Sync: for Mobile Underwater Sensor Efficient Time Synchronization Networks," IEEE Trans. Parallel Distrib. Syst. 24, 406 -- 416, (2013). 19. J. Liu, Z. Wang, J. Cui, S. Zhou, and B. Yang, "A Joint Time Synchronization and Localization Design for Mobile Underwater Sensor Networks," IEEE Trans. Mobile Comput. 15, 530 -- 543, (2016). 20. B. Sprenger, J. Zhang, Z. Lu, and L. Wang, "Atmospheric transfer of optical and radio frequency clock signals," Opt. Lett. 34 965 -- 967 (2009). 21. F. R. Giorgetta, W. C. Swann, L. C. Sinclair, E. Baumann, I. Coddington, and N. R. Newbury, "Optical two-way time and frequency transfer over free space," Nat. Photonics 7, 435 -- 439 (2013). 22. S. Chen, F. Sun, Q. Bai, D. Chen, Q. Chen, and D. Hou, "Sub-picosecond timing fluctuation suppression in laser-based atmospheric transfer of microwave signal using electronic phase compensation," Opt. Commun. 401, 18 -- 22 (2017). 23. F. Sun, D. Hou, D. Zhang, J. Tian, J. Hu, X. Huang, and S Chen, "Femtosecond-level timing fluctuation suppression in atmospheric frequency transfer with passive phase conjunction correction," Opt. Express 25, 21312 -- 21320 (2017). 24. L. Sinclair, H. Bergeron, W. Swann, E. Baumann, J. Deschenes, and N. R. Newbury, "Comparing Optical Oscillators across the Air to Milliradians in Phase and 10(-17) in Frequency," Phys. Rev. Lett. 120, 050801, (2018). 25. L. Mullen, D. Alley, and B. Cochenour, "Investigation of the effect of scattering agent and scattering albedo on modulated light propagation in water," Appl. Opt. 50, 1396 -- 1404, (2011). 26. A. Luchinin and M. Kirillin, "Temporal and frequency characteristics of a narrow light beam in sea water," Appl. Opt. 55, 7756 -- 7762, (2016). 27. B. Cochenour, K. Dunn, A. Laux, and L. Mullen, "Experimental measurements of the magnitude and phase response of high-frequency modulated light underwater," Appl. Opt. 56, 4019 -- 4024, (2017). 28. W. Liu, Z. Xu, and L. Yang, "SIMO detection schemes for underwater optical wireless communication under turbulence," Photon. Res. 3, 48 -- 53, 2015. 29. Microsemi, https://www.microsemi.com/document- "DS-5071a", portal/doc_download/133269-5071a-datasheet, (2014). 30. Microsemi, "MHM 2010", https://www.microsemi.com/document- portal/doc_download/133261-mhm-2010-datasheet, (2011).
1806.08839	1	1806	2018-06-22T20:12:53	Quantitative principles for precise engineering of sensitivity in carbon-based electrochemical sensors	[ "physics.app-ph" ]	A major practical barrier for implementing carbon-based electrode arrays with high device-packing density is to ensure large, predictable, and homogeneous sensitivities across the array. Overcoming this barrier depends on quantitative models to predict electrode sensitivity from its material structure. However, such models are currently lacking. Here, we show that the sensitivity of multilayer graphene electrodes increases linearly with the average point defect density, whereas it is unaffected by line defects or oxygen-containing groups. These quantitative relationships persist until the electrode material transitions to a fully disordered sp2 carbon, where sensitivity declines sharply. We show that our results generalize to a variety of graphene production methods and use them to derive a predictive model that guides nano-engineering graphene structure for optimum sensitivity. Our approach achieves reproducible fabrication of miniaturized sensors with extraordinarily higher sensitivity than conventional material. These results lay the foundation for new integrated electrochemical sensor arrays based on nano-engineered graphene.	physics.app-ph	physics	Quantitative principles for precise engineering of sensitivity in carbon-based electrochemical sensors Ting Wu,1† Abdullah Alharbi,1† Roozbeh Kiani2,3, Davood Shahrjerdi1,4∗ 1Electrical and Computer Engineering, New York University, Brooklyn, NY 11201 2Center for Neural Science, New York University, New York, NY, 10003, USA 3Neuroscience Institute, New York University Langone Medical Center, New York, NY 10016 4Center for Quantum Phenomena, Physics Department, New York University, NY 10003 ∗To whom correspondence should be addressed; E-mail: [email protected]. †These authors contributed equally to this work. A major practical barrier for implementing carbon-based electrode arrays with high device-packing density is to ensure large, predictable, and homogeneous sensitiv- ities across the array. Overcoming this barrier depends on quantitative models to predict electrode sensitivity from its material structure. However, such models are currently lacking. Here, we show that the sensitivity of multilayer graphene elec- trodes increases linearly with the average point defect density, whereas it is un- affected by line defects or oxygen-containing groups. These quantitative relation- ships persist until the electrode material transitions to a fully disordered sp2 carbon, where sensitivity declines sharply. We show that our results generalize to a vari- ety of graphene production methods and use them to derive a predictive model that guides nano-engineering graphene structure for optimum sensitivity. Our ap- proach achieves reproducible fabrication of miniaturized sensors with extraordi- narily higher sensitivity than conventional material. These results lay the foun- dation for new integrated electrochemical sensor arrays based on nano-engineered graphene. 1 The ease of fabrication and operation of carbon-based electrochemical sensors gives them the potential to enable a new class of integrated sensor systems with wide-ranging applications from drug development to clinical diagnostics. To support these applications, the sensor sys- tem requires high spatial density (i.e., a dense packing of miniaturized sensors) and consistent operations across the sensor array (i.e., sensors with predictable and homogeneous sensitiv- ity). Moreover, high-precision applications additionally require electrodes with high sensitivity. Although the availability of advanced fabrication techniques would allow miniaturization of carbon-based electrochemical sensors (1–4), satisfying the low variability and predictability requirements of sensitivity across a dense sensor array remains a difficult challenge. The sensi- tivity of this family of sensors is tied to the structural properties of the electrode material (5–7). However, it is natural for the material to have atomic-level structural inhomogeneity, which can cause variability in the electrode sensitivity among sensors. Due to the random spatial dis- tribution of the structural inhomogeneities in the electrode material, this variability becomes more pronounced with reducing the sensor size. Further, no method is currently available for predicting the electrode sensitivity from its material structure. The current limitations in pro- ducing predictable and homogeneous array of electrodes with high sensitivity have prevented the implementation of highly integrated sensor systems based on carbon-based electrochemical sensors. To account for the variability and also determine the electrode sensitivity, the common prac- tice is to calibrate each sensor through post-manufacturing measurements, which involves cre- ating "calibration curves" by measuring the sensor response to known concentrations of ana- lytes (8). Although this strategy is applicable for dealing with individual devices or a small sensor array, it is highly inefficient for the implementation of large-scale integrated sensor sys- tems. A more tractable approach is to precisely engineer the material synthesis in order to reliably yield miniaturized electrodes with predictable and homogeneous sensitivities. The ef- 2 ficacy of this approach, however, hinges on an understanding of the quantitative relationship between electrode sensitivity and the structural properties of the material. Such a quantitative model is currently lacking. Our study aims to provide this model. Among different forms of carbon, electrodes from sp2 carbon nanomaterials (e.g., carbon nanotubes, graphene, and reduced graphene oxide) have shown high sensitivity (9–12), making them promising for sensing applications that might require high precision measurements of an analyte concentration. The sensitivity of those electrodes has been often attributed to structural defects such as edge plane sites (9, 13, 14) or oxygen-containing functional groups (10, 11). However-and this attests to the challenging nature of the problem-no study has yet directly tested and quantified the effect of different defect types on the sensitivity of carbon-based elec- trodes. Overcoming this problem will result in a quantitative understanding of how structural defects influence electrode sensitivity-a crucial building block for a desired predictive model. To address this problem, we undertake a systematic study of sensitivity in miniaturized carbon-based electrochemical sensors consisting of multilayer graphene electrodes. We used multilayer graphene as a model system due to the ease of defect engineering (15) and the avail- ability of experimentally-established Raman-based models for quantifying defects (16–20). Our tight control over the sensor fabrication process, together with the accuracy and reproducibility of our material characterizations and sensor measurements, have allowed us, for the first time, to experimentally determine the quantitative relationships between various defect types and the electrode sensitivity in graphene-based electrochemical sensors. We find that the electrode sensitivity is amplified in linear proportion to the average density of point defects. In contrast, the average density of line defects or oxygen-containing groups has no effect on the electrode sensitivity. Once the concentration of point defects exceeds a threshold value, which indicates the onset of transition to a fully disordered sp2 carbon material, electrode sensitivity decreases rapidly. Based on these findings, we derive a predictive model 3 for engineering the structural properties of carbon-based electrodes on an atomic level to match a desired sensitivity. As a practical example for the utility of our model and methods, we nano- engineer electrodes with unprecedented sensitivity for in vitro measurements of dopamine. Our engineered electrodes show up to 20 times higher sensitivity to dopamine than conventional carbon fiber (CF) electrodes (21–23). Moreover, we show that the sensitivity of our engineered electrodes match the predictions of our quantitative model. Finally, we show that our model consistently applies to graphene-based sensors produced through different synthesis methods, promising wide applicability of our findings for prediction and engineering of the sensitivity of carbon-based electrochemical sensors. Results Enhanced electrode sensitivity in multilayer graphene. To evaluate the link between the atomic structure of multilayer graphene and the sensitivity of electrodes made of it, we per- formed electrochemical sensing of biochemical molecules using fast-scan cyclic voltamme- try (FSCV). Owing to its good ionic specificity and sub-second detection ability, FSCV with carbon-based electrodes has been used extensively for measuring biochemical molecules in chemically complex environments such as the brain (9, 21, 24, 25). We initially constructed FSCV electrodes using multilayer graphene grown by chemical vapor deposition (CVD) and measured their sensitivity for neurochemical molecules in vitro. CVD graphene films typically have different amounts of sp2-hybridized defects, due to minor differences in the production method, apparatus, or even the granular structure of the growth substrate (26, 27). To increase the diversity of different defect densities in our sensor electrodes, we obtained several batches of CVD multilayer graphene films grown on nickle foils. To fabricate electrodes, we transferred the CVD multilayer graphene films with an aver- age thickness of 30 nm onto SiO2/Si substrates using standard chemical layer-transfer pro- 4 cesses (28–30). Using nanofabrication, we then made miniaturized sensor electrodes with a planar geometry, shown schematically in Fig. 1a. The details of the fabrication process are given in Supporting Information. We used a similar process for fabricating all devices dis- cussed in this paper. We designed the fabrication process around two key factors. The first one is to avoid creating unintentional defects in multilayer graphene during the sensor fabrication. This consideration is particularly important when making electrodes from defect-engineered multilayer graphene, discussed later. To do so, during the fabrication process, we protected the sensing region of the electrode with a thin metal layer (Cr/Au; 5 nm/50 nm). Second, for analyzing the sensor response in our study, we used the area-normalized sensitivity. We defined the sensing region of the electrodes accurately using an SU8 encapsulating layer. This layer also protected the metal contact from exposure to the electrolyte solution. To perform the sens- ing experiments, we removed the protective metal stack and mounted a fluidic chamber on the samples. Fig. 1b shows the top-view scanning electron microscopy (SEM) image of an example sensor array. For comparison, we also fabricated electrodes from conventional CFs (Fig. S3), commonly used in FSCV measurements of neurochemicals in the brain (22, 31). We characterized the sensitivity of the fabricated sensors through FSCV measurements of dopamine-an important neuromodulator for action-selection and reward-motivated behav- ior (32–34). During the FSCV measurement, dopamine undergoes a redox reaction (Fig. 1d), where it is oxidized to dopamine-o-quinone (DOQ) by a voltage ramp-up applied to the elec- trode (see Fig. 1a). The amplitude of the resulting oxidation current is a measure of the dopamine concentration. The voltage subsequently ramps down, causing the DOQ molecules to be reduced back to dopamine, which gives rise to a reduction current. FSCV estimates dopamine concentration based on the magnitude of the oxidation current. Electrode sensitiv- ity represents the change in the peak of the oxidation current (ip,ox) per unit concentration of a biomolecule (Fig. S13c-d). We defined unit concentration as µM and the area-normalized 5 sensitivity, SA, as ip,ox at 1 µM divided by electrode area. Because the amplitude of the electro- chemical current is proportional to the geometric surface area of the sensors, normalization of sensitivity by sensor area enables comparison of sensors with diverse sizes. Surface roughness increases the geometric surface area and can potentially bias the area-normalized sensitivity. Therefore, we estimated the total surface area of our multilayer graphene sensors by performing atomic force microscopy (AFM) and measuring surface roughness before FSCV experiments (Fig. 1c, Fig. S16-S17). As a result, our area-normalized sensitivity is independent of the sensor geometry and reflects the inherent sensing property of the electrode material. Fig. 1e shows the area-normalized electrochemical current (IEC) curves for four example electrodes (three CVD and one CF) in response to a 1 µM dopamine solution. The green circles on the curves denote ip,ox. This plot and the FSCV measurements of the other CVD electrodes (Fig. S14) demonstrate substantial variations in electrode sensitivity. Many sensors were minimally responsive to dopamine molecules and a few showed noticeably higher SA than the CF devices. We hypothesized that diversity of structural defects of the sensing material was crucial for explaining the wide range of observed electrode sensitivities. Quantifying structural defects in carbon-based electrodes. The ability to distinguish dif- ferent types of defects and quantify their amounts in the electrode material not only is critical for revealing the connection between structural defects and the electrode sensitivity, but might also lead to a predicative framework for the sensitivity-targeted engineering of the sensor mate- rial. While the types of structural defects are diverse, one simple way to classify them is based on their dimensionality. For example, defects in materials with a two-dimensional lattice, such as graphene, are either zero-dimensional (point defects) or one-dimensional (line defects). Ex- amples of point defects in graphene are vacancies (35–37), Stone-Wales defects (36, 37), and dopants (38). On the other hand, crystallite borders (17) and extended dislocations (36) are 6 examples of line defects. Point and line defects are often simultaneously present in synthetic graphene-based materials, as shown schematically in Fig. 2a. Raman spectroscopy has been experimentally established as a useful non-destructive method for identifying and quantifying sp2-hybridized defects in graphene based on their dimensionality (16–20). Fig. 2b shows representative Raman spectra for a few CVD samples and a CF electrode. The distinct peaks in the Raman spectrum of multilayer graphene films are well-studied (39,40). The G peak appears at about 1576 cm−1 and signifies the sp2 hybridization of carbon atoms. The D peak arises from the breathing modes of aromatic carbon rings and signifies sp2-hybridized defects. The 2D peak is the second-order of the D peak, which is present only in fully sp2- bonded carbon materials. Changes of these peaks in Fig. 2b (from bottom to top) illustrates the gradual transition of the film structure (i) from a highly ordered multilayer graphene to a disordered nanocrystalline graphite and (ii) finally to a fully disordered sp2 carbon material. In stage (i), the D peak intensity increases monotonically and the 2D peak is visible in the Raman spectra. Upon transition into stage (ii), the 2D peak becomes noticeably broad and its intensity weakens dramatically. The combination of our CVD and CF electrodes covered the whole spectrum of the graphene amorphization trajectory. We applied a recently advanced theoretical method (20) for extracting the amounts of point and line defects from the measured Raman spectra of our sensor samples (see Supporting In- formation). The results of this method have been previously validated by scanning tunneling microscopy (20), illustrating its ability to unambiguously distinguish point and line defects in graphene-based samples. This theoretical method relies on numerical simulations based on the area ratio of the D and G peaks and the line width of the G-band to derive the average crystal- lite size (La) and the average distance between point defects (LD) within the spot size of the Raman laser. The details of our La and LD calculations are given in Supporting Information. We note that this methodology for quantifying defects is independent of the production method 7 of multilayer graphene, making it suitable for our study involving CVD, graphitized, and CF materials. Since the location of defects on a sensor electrode is random, we estimated the concentration of each defect type on a sensor electrode by measuring the number of defects averaged over the sensor surface area. To do so, we first obtained the spatial Raman maps of our sensor electrodes and quantified La and LD at each Raman spot (see Supporting Information). Fig. 2c shows the spatially resolved distributions of La and LD for an example CVD electrode. The mean values from the La and LD distributions were then used for estimating the average density of point de- 2) in our sensor samples. This methodology allows −2) and average crystallite area (La fects (LD us to analyze the relationship between the area-normalized sensitivity and the concentration of each defect type in our carbon-based electrodes. −2 and La Fig. 2d shows the scatter of LD 2 for our CVD sensor samples, highlighting the large diversity of line and point defects in our candidate sensor samples. In this plot, the gray box (in the lower right corner) marks the region, where Raman lacks accuracy for estimating point and line defect concentrations, because the expected LD and La values are beyond the upper detection limits of Raman. The Raman spectra of samples that fall in this region typically do not show a visible D peak. We refer to those samples as pristine. Moreover, past Raman studies of defective graphene suggest that the onset of stage (ii) of the amorphization trajectory occurs at LD of about 4-5 nm (19). This length scale is comparable to the localization length of the disorder-induced Raman D band at 300 K (41). The yellow shading in Fig. 2d marks the stage (ii), where the CF sample is. In contrast, our CVD sensor samples were in stage (i) of the amorphization trajectory. Effect of defects on electrode sensitivity. To reveal the quantitative effect of defects on mul- tilayer graphene sensors in stage (i) of the amorphization trajectory, we made a contour plot 8 of the area-normalized sensitivity (SA) as a function of the average crystallite area (La 2) and −2), shown in Fig. 3a. The plot shows that electrodes with similar density of point defects exhibited nearly similar SA, regardless of their crystallite size. the average point defect density (LD Further, SA was amplified with increasing the density of point defects. Next, we analyzed the relationship between SA and the point defect concentration (i.e., n0D = L−2 D ). Plotting SA as a function of the point defect concentration indicated a linear relation between the area-normalized sensitivity and the point defect density when the sensing material was in stage (i), as shown in Fig. 3b. From the linear fit to the data, we found that the x-intercept occurs at a point defect density of 1.6 × 1011 cm−2. We denoted this point defect concentra- tion as n∗ 0D. From these observations, we then derived an empirical model that represents the experimentally measured SA for dopamine when the electrode material was in stage (i) of the amorphization trajectory. This model, given below, is valid for point defect concentrations greater than n∗ 0D: SA = (6.4 ± 0.2) × 10−11(n0D − n∗ 0D) (1) where SA and n0D have units of pA.µm−2.µM−1 and cm−2, respectively. We note that in our sensing experiments, electrodes containing a point defect concentration below n∗ 0D, including those made from pristine graphene, are minimally responsive, i.e., their sensitivity was below the measurable limit of our readout system (the magenta dashed line in Fig. 3b). From the data, we also found that upon transition into stage (ii), where the CF electrode is, SA no longer follows the linear trend. On the basis of the above observations, we made the following hypotheses regarding the sensitivity-targeted engineering of the graphene synthesis process. First, the synthesis process does not need to be optimized for line defects, since they had no observable effect on the elec- trode sensitivity of CVD sensor samples. Second, the quantitative relationship between SA and 9 the density of point defects in stage (i), quantitatively described by equation (1), can be used as a predictive model for tuning the sensitivity of electrodes. And finally, the sensitivity degrades as the material structure transitions into stage (ii). An important prediction of these hypotheses is that SA can be maximized by maximizing the density of point defects in multilayer graphene before transitioning to a fully disordered sp2 carbon material. Engineered electrodes with predictable sensitivity. We directly tested the validity of our hypotheses by developing a process that allowed us to engineer multilayer graphene films with different amounts of point and line defects. Of the various approaches for producing multi- layer graphene (26, 29, 42–45), we adapted a method based on metal-induced transformation of amorphous carbon to multilayer graphene using a thin nickel catalyst (46, 47). Our process involved creating diamond-like carbon (DLC) islands directly on SiO2/Si substrates and graphi- tizing them at temperatures between 1000 and 1100 ◦C, as shown in Fig. 4a (see Supporting Information for details of the graphitization process). We probed the graphitic structure of the samples using high-resolution x-ray photoelectron spectroscopy (XPS) (Fig. 4b). Curve fitting of the XPS data indicated the sp2 nature of the carbon-carbon bonds. The 2D peaks in Raman spectra of the films (Fig. 4c) further confirmed the multilayer graphene growth. We observed that the structural properties of the multilayer graphene films made from this method sensitively depended on the nickel thickness, annealing temperature, and growth time. By tuning these pa- rameters, guided by Raman analysis, we could reliably and reproducibly generate samples with desired amounts of structural defects. We refer to these multilayer graphene samples as the graphitized (GR) samples. In our experiments, we created three sets of graphitized sensor samples. A first group that contained the same amounts of point defects as the CVD samples. This was to reproduce the sensitivity of those CVD electrodes and hence to demonstrate that our experimental predictive 10 model of electrode sensitivity is independent of the graphene production method. A second group of samples had significantly higher amounts of point defects than the CVD devices with- out transitioning to stage (ii). This set of GR samples not only allowed us to investigate the validity of our predictive model for the range of point defect concentrations that was not cov- ered by the CVD sensor samples, but also enabled us to explore the upper limit of SA in stage (i). We also created a third group of GR samples, which were in stage (ii) of the amorphization trajectory of graphene. These samples allowed us to confirm the drop in SA once the structure of the electrode material transforms into a fully disordered sp2 carbon material. Fig. 4c shows the representative Raman spectra of a few GR samples and their corresponding SA for dopamine. In Fig. 4d, we show the summary of average point and line defect concentrations for our engi- neered GR samples. In this plot, we have also included the CVD and CF samples to facilitate the distinction between the three groups of the GR samples. In addition to regions of overlap with the CVD and CF samples, note the region where the GR samples were engineered to have a significantly higher point defect density than the CVD samples without transitioning to stage (ii), the yellow shading in the plot. We used miniaturized electrodes fabricated from defect-engineered graphitized films to per- form FSCV measurements of dopamine (Fig. S15). Fig. 4e shows the contour plot of SA for the graphitized and CVD electrodes, where the multilayer graphene sensor electrodes are in stage (i). The plot shows that like the CVD samples, the sensitivity of graphitized electrodes in- creased with point defect density and was independent of the crystallite size. Plotting the SA as a function of point defect density (see Fig. 4f) revealed three critical results. First, the SA of the graphitized sensors followed the same linear trend as the CVD samples, confirming that the av- erage density of point defects was the main predictor of the SA for multilayer graphene films in stage (i). The exact method of production of the multilayer graphene film did not matter as long as the point defect concentration remained the same, hence confirming that our predictive model 11 in equation (1) can be generalized to other graphene production methods. Second, by increas- ing the density of point defects in stage (i), our model predicts that one could maximize SA. Indeed, we achieved a remarkably high SA of 177 pA.µm−2.µM−1 in response to dopamine, which is about 20 times higher than the sensitivities reported for CF electrodes in past stud- ies (48, 49). Third, the transition of the multilayer graphene into a fully disordered sp2 carbon material caused a rapid degradation of SA. Our data indicates that SA began to degrade at LD of about 5-6 nm, which coincides with the onset of transition to stage (ii). Conventional CF electrodes that are used for studying neurochemical changes in the brain (21, 23, 48, 50) are usually in this regime. Our finding explains the fundamentally small sensitivity of these elec- trodes and suggests a practical method to significantly boost the electrode sensitivity for such measurements. To ensure our results were not limited to a particular molecule, we also performed in vitro FSCV measurement of serotonin neurotransmitters using the multilayer graphene electrodes in stage (i). The results confirmed that the linear increase of electrode sensitivity with increas- ing average point defect density generalized to serotonin and was, therefore, a property of the electrode not the measured analyte (Fig. S18-S19). Effect of oxygen-containing functional groups. Previous studies hypothesized a direct cor- relation between the electrode sensitivity and the amount of oxygen-containing functional groups available on the surface of carbon electrodes (9, 51, 52). To examine the role of oxygen func- tional groups, we performed XPS measurements on multiple multilayer graphene sensor elec- trodes with markedly different area-normalized sensitivity. The electrodes were chosen from both CVD and graphitized sensor samples and the XPS measurements were performed imme- diately after the FSCV experiments in an ultra high vacuum (UHV) environment. By analyzing the XPS data, shown in Fig. 5, we ruled out a significant role of oxygen-containing functional 12 groups in amplifying the sensitivity of our electrodes made from multilayer graphene. Hence, we conclude that point defects are the most critical factor for amplifying the sensitivity of graphene-based electrodes that are in stage (i) of the amorphization trajectory. Discussion Our findings establish fundamental principles for predicting the sensitivity of graphene-based electrochemical sensors based on structural defects. Using these principles, we devised a quan- titative methodology for reproducible fabrication of homogeneous sensors with optimized sen- sitivity. The remarkably high sensitivity of our miniaturized electrodes are due to the maxi- mization of point defect concentration while keeping the electrode material in stage (i) of the amorphization trajectory of graphene. The density of line defects and oxygen-containing functional groups appear to have minimal bearing on the electrode sensitivity. This observation simplifies the electrode manufacturing process by removing the need for monitoring and optimization of line defects and oxygen- containing groups. Further, it contradicts previous speculations that non-carbon functional groups may be key to the sensitivity of carbon-based electrodes (10, 11). However, we do not rule out the possibility that one may further increase the sensitivity of our nanoengineered electrodes by adding special functional groups that amplify the effect of point defects. It is currently unclear why point defects have such a critical role in determining the sen- sitivity of multilayer graphene electrodes. We speculate that these defects provide an optimal environment for electron transfer during the redox reaction by (i) fixing the chemical potential (Fermi energy) at the Dirac point (53,54), and (ii) simultaneously increasing the local density of states at about the Dirac point in proportion to their concentration in stage (i) (15,55). However, the physical principles that govern this effect should yet be elucidated. Further, studying the correlation between the specific sub-types of point defects and the electrode sensitivity might 13 be useful in further refining the theory of the sensor operation. Our discovery has far-reaching practical implications. By providing guidelines for opti- mizing the sensitivity of carbon-based electrochemical sensors, we enable significant improve- ments in a wide range of applications from a next generation of neural probes to multiplexed lab-on-a-chip sensing platforms, or more generally, wherever accurate recording of biochemical compounds using a dense array of sensors is essential. Specifically, our quantitative nanoengi- neering methodology can ensure fabrication of a dense array of sensors with predictable and homogeneous sensitivity, thus making such nanoengineered carbon-based electrodes suitable for compact, multi-channel sensor systems required in large-scale applications. Further, our nanoengineered electrodes overcome an existing obstacle for industrial-scale fabrication of reli- able sensors with reproducible electrode sensitivity. Current methods for fabrication of carbon- based electrodes are not developed to optimize the density of point defects. Consequently, they are largely dependent on post-manufacturing measurement for calibration of sensors and are prone to producing minimally responsive sensors that have to be discarded. In contrast, our fabrication method enables industrial-scale and targeted nanomanufacturing of carbon-based electrodes that have desired sensitivity levels far beyond their predecessors. Materials and Methods Multilayer graphene films: CVD multilayer graphene samples were obtained from two com- mercial vendors: Graphene Supermarket and Graphene Platform, Inc. The CVD films were grown on nickel foils, which were then removed chemically during the layer transfer process using the commercial nickel etchant TFG (Transene Company Inc.). The free-standing CVD films were subsequently mounted on p+ silicon substrates capped with 285 nm thermally grown SiO2. The graphitized samples were produced using a custom-made system. Details of the metal-induced graphitization process is given in Supporting Information. 14 Raman measurements: To quantify the structural defects in multilayer graphene films, Raman measurements were performed using Horiba Xplora µ-Raman system with a 532 nm incident laser. The Raman spectra were fitted using Lorenztian functions, providing the FWHM and area of the G and D peaks. The curve fitting results were then used to extract the density of line and point defects according to the theoretical simulations described in (20). Details of the Raman analysis are given in Supporting Information. FSCV measurements: To examine the electrode sensitivity, FSCV measurements of dopamine and serotonin (Sigma Aldrich) were performed using a custom-made FSCV setup. The FSCV setup consists of a low-noise current amplifier (SR570, Stanford Research Systems), a data acquisition card (NI 6353 X series, National Instruments), and a MATLAB control interface. In FSCV sensing experiments, the biomolecules were dissolved in a 1x phosphate buffer saline (PBS) solution. PBS was prepared using the recipe in (56). References 1. W. Gao, S. Emaminejad, H. Y. Y. Nyein, S. Challa, K. Chen, A. Peck, H. M. Fahad, H. Ota, H. Shiraki, D. Kiriya, et al., Fully integrated wearable sensor arrays for multiplexed in situ perspiration analysis. Nature 529, 509 (2016). 2. X. Xuan, M. F. Hossain, J. Y. Park, A fully integrated and miniaturized heavy-metal- detection sensor based on micro-patterned reduced graphene oxide. Scientific reports 6, 33125 (2016). 3. B. Nasri, T. Wu, A. Alharbi, K.-D. You, M. Gupta, S. P. Sebastian, R. Kiani, D. Shahrjerdi, Hybrid cmos-graphene sensor array for subsecond dopamine detection. IEEE transactions on biomedical circuits and systems 11, 1192–1203 (2017). 15 4. X. Xuan, H. S. Yoon, J. Y. Park, A wearable electrochemical glucose sensor based on simple and low-cost fabrication supported micro-patterned reduced graphene oxide nanocomposite electrode on flexible substrate. Biosensors and Bioelectronics 109, 75–82 (2018). 5. Y. Shao, J. Wang, H. Wu, J. Liu, I. A. Aksay, Y. Lin, Graphene based electrochemical sensors and biosensors: a review. Electroanalysis 22, 1027–1036 (2010). 6. C. B. Jacobs, M. J. Peairs, B. J. Venton, Carbon nanotube based electrochemical sensors for biomolecules. Analytica chimica acta 662, 105–127 (2010). 7. Z. Zhu, An overview of carbon nanotubes and graphene for biosensing applications. Nano- Micro Letters 9, 25 (2017). 8. E. Sinkala, J. E. McCutcheon, M. J. Schuck, E. Schmidt, M. F. Roitman, D. T. Eddington, Electrode calibration with a microfluidic flow cell for fast-scan cyclic voltammetry. Lab on a Chip 12, 2403–2408 (2012). 9. A. C. Schmidt, X. Wang, Y. Zhu, L. A. Sombers, Carbon nanotube yarn electrodes for enhanced detection of neurotransmitter dynamics in live brain tissue. ACS Nano 7, 7864– 7873 (2013). 10. G. P. Keeley, N. McEvoy, H. Nolan, M. Holzinger, S. Cosnier, G. S. Duesberg, Electroana- lytical sensing properties of pristine and functionalized multilayer graphene. Chem. Mater. 26, 1807–1812 (2014). 11. G. Gao, M. Pan, C. D. Vecitis, Effect of the oxidation approach on carbon nanotube surface functional groups and electrooxidative filtration performance. J. Mater. Chem. A 3, 7575– 7582 (2015). 16 12. C. Chung, Y. K. Kim, D. Shin, S. R. Ryoo, B. H. Hong, D. H. Min, Biomedical applications of graphene and graphene oxide. Acc. Chem. Res. 46, 2211–2224 (2013). 13. C. E. Banks, R. G. Compton, New electrodes for old: From carbon nanotubes to edge plane pyrolytic graphite. Analyst 131, 15–21 (2006). 14. C. E. Banks, T. J. Davies, G. G. Wildgoose, R. G. Compton, Electrocatalysis at graphite and carbon nanotube modified electrodes: Edge-plane sites and tube ends are the reactive sites. Chem. Commun. pp. 829–841 (2005). 15. M. M. Ugeda, I. Brihuega, F. Guinea, J. M. G´omez-Rodr´ıguez, Missing atom as a source of carbon magnetism. Phys. Rev. Lett. 104, 096804 (2010). 16. F. Tuinstra, J. L. Koenig, Raman spectrum of graphite. J. Chem. Phys. 53, 1126–1130 (1970). 17. L. G. Canc¸ado, K. Takai, T. Enoki, M. Endo, Y. A. Kim, H. Mizusaki, A. Jorio, L. N. Coelho, R. Magalhaes-Paniago, M. A. Pimenta, General equation for the determination of the crystallite size la of nanographite by raman spectroscopy. Appl. Phys. Lett. 88, 163106 (2006). 18. M. M. Lucchese, F. Stavale, E. H. M. Ferreira, C. Vilani, M. V. O. Moutinho, R. B. Capaz, C. A. Achete, A. Jorio, Quantifying ion-induced defects and raman relaxation length in graphene. Carbon 48, 1592–1597 (2010). 19. L. G. Canc¸ado, A. Jorio, E. H. Martins Ferreira, F. Stavale, C. A. Achete, R. B. Capaz, M. V. O. Moutinho, A. Lombardo, T. S. Kulmala, A. C. Ferrari, Quantifying defects in graphene via raman spectroscopy at different excitation energies. Nano Lett. 11, 3190– 3196 (2011). 17 20. L. G. Canc¸ado, M. G. da Silva, E. H. M. Ferreira, F. Hof, K. Kampioti, K. Huang, A. P´enicaud, C. A. Achete, R. B. Capaz, A. Jorio, Disentangling contributions of point and line defects in the raman spectra of graphene-related materials. 2D Mater. 4, 025039 (2017). 21. R. M. Wightman, L. J. May, A. C. Michael, Detection of dopamine dynamics in the brain. Anal. Chem. 60, 769A–793A (1988). 22. R. M. Wightman, D. L. Robinson, Transient changes in mesolimbic dopamine and their association with "reward". Journal of neurochemistry 82, 721–735 (2002). 23. K. T. Kishida, I. Saez, T. Lohrenz, M. R. Witcher, A. W. Laxton, S. B. Tatter, J. P. White, T. L. Ellis, P. E. M. Phillips, P. R. Montague, Subsecond dopamine fluctuations in human striatum encode superposed error signals about actual and counterfactual reward. Proc. Natl. Acad. Sci. 113, 200–205 (2016). 24. W. G. Kuhr, R. M. Wightman, Real-time measurement of dopamine release in rat brain. Brain Res. 381, 168–171 (1986). 25. B. J. Venton, R. M. Wightman, Psychoanalytical electrochemistry: Dopamine and behavior. Anal. Chem. 75, 414 A-421 A (2003). 26. A. Reina, X. Jia, J. Ho, D. Nezich, H. Son, V. Bulovic, M. S. Dresselhaus, J. Kong, Large area, few-layer graphene films on arbitrary substrates by chemical vapor deposition. Nano Lett. 9, 30–35 (2008). 27. X. Li, C. W. Magnuson, A. Venugopal, R. M. Tromp, J. B. Hannon, E. M. Vogel, L. Colombo, R. S. Ruoff, Large-area graphene single crystals grown by low-pressure chem- ical vapor deposition of methane on copper. J. Am. Chem. Soc. 133, 2816–2819 (2011). 18 28. X. Li, Y. Zhu, W. Cai, M. Borysiak, B. Han, D. Chen, R. D. Piner, L. Colombo, R. S. Ruoff, Transfer of large-area graphene films for high-performance transparent conductive electrodes. Nano Lett. 9, 4359–4363 (2009). 29. K. S. Kim, Y. Zhao, H. Jang, S. Y. Lee, J. M. Kim, K. S. Kim, J.-H. Ahn, P. Kim, J.-Y. Choi, B. H. Hong, Large-scale pattern growth of graphene films for stretchable transparent electrodes. Nature 457, 706–710 (2009). 30. Y. Lee, S. Bae, H. Jang, S. Jang, S.-E. Zhu, S. H. Sim, Y. I. Song, B. H. Hong, J. H. Ahn, Wafer-scale synthesis and transfer of graphene films. Nano Lett. 10, 490–493 (2010). 31. E. S. Bucher, R. M. Wightman, Electrochemical analysis of neurotransmitters. Annu. Rev. Anal. Chem. 8, 239–261 (2015). 32. P. W. Glimcher, Understanding dopamine and reinforcement learning: the dopamine reward prediction error hypothesis. Proc. Natl. Acad. Sci. 108, 15647-15654 (2011). 33. S. B. Floresco, The nucleus accumbens: An interface between cognition, emotion, and action. Annu. Rev. Psychol. 66, 25-52 (2015). 34. R. M. Costa, A selectionist account of de novo action learning. Curr. Opin. Neurobiol. 21, 579 - 586 (2011). 35. G.-D. Lee, C. Wang, E. Yoon, N.-M. Hwang, D.-Y. Kim, K. M. Ho, Diffusion, coalescence, and reconstruction of vacancy defects in graphene layers. Phys. Rev. Lett. 95, 205501 (2005). 36. L. Vicarelli, S. J. Heerema, C. Dekker, H. W. Zandbergen, Controlling defects in graphene for optimizing the electrical properties of graphene nanodevices. ACS nano 9, 3428–3435 (2015). 19 37. F. Banhart, J. Kotakoski, A. V. Krasheninnikov, Structural defects in graphene. ACS nano 5, 26–41 (2010). 38. L. L. Zhang, X. Zhao, H. Ji, M. D. Stoller, L. Lai, S. Murali, S. Mcdonnell, B. Cleveger, R. M. Wallace, R. S. Ruoff, Nitrogen doping of graphene and its effect on quantum capac- itance, and a new insight on the enhanced capacitance of n-doped carbon. Energy Environ. Sco. 5, 9618–9625 (2012). 39. A. C. Ferrari, Raman spectroscopy of graphene and graphite: Disorder, electron–phonon coupling, doping and nonadiabatic effects. Solid State Commun. 143, 47–57 (2007). 40. A. C. Ferrari, D. M. Basko, Raman spectroscopy as a versatile tool for studying the prop- erties of graphene. Nat. Nanotechnol. 8, 235–246 (2013). 41. R. Beams, L. G. Canc¸ado, L. Novotny, Low temperature raman study of the electron coher- ence length near graphene edges. Nano Lett. 11, 1177–1181 (2011). 42. P. C. Rodriguez, D. B. Pereira, A. Borgkvist, M. Y. Wong, C. Barnard, M. S. Sonders, H. Zhang, D. Sames, D. Sulzer, Fluorescent dopamine tracer resolves individual dopamin- ergic synapses and their activity in the brain. Proc. Natl. Acad. Sci. 110, 870-875 (2013). 43. K. Wang, G. Tai, K. Wong, S. Lau, W. Guo, Ni induced few-layer graphene growth at low temperature by pulsed laser deposition. AIP Adv. 1, 022141 (2011). 44. J. Hass, W. A. De Heer, E. H. Conrad, The growth and morphology of epitaxial multilayer graphene. J. Phys.: Condens. Matter 20, 323202 (2008). 45. H. Cao, Q. Yu, R. Colby, D. Pandey, C. S. Park, J. Lian, D. Zemlyanov, I. Childres, V. Drachev, E. A. S. Stach, M. Hussain, H. Li, S. S. Pei, Y. P. Chen, Large-scale graphitic 20 thin films synthesized on ni and transferred to insulators: Structural and electronic proper- ties. J. Appl. Phys. 107, 044310 (2010). 46. J. A. Rodr´ıguez-Manzo, C. Pham-Huu, F. Banhart, Graphene growth by a metal-catalyzed solid-state transformation of amorphous carbon. ACS Nano 5, 1529–1534 (2011). 47. D. Berman, S. A. Deshmukh, B. Narayanan, S. K. Sankaranarayanan, Z. Yan, A. A. Ba- landin, A. Zinovev, D. Rosenmann, A. V. Sumant, Metal-induced rapid transformation of diamond into single and multilayer graphene on wafer scale. Nat. Commun. 7, 12099 (2016). 48. E. W. Schluter, A. R. Mitz, J. F. Cheer, B. B. Averbeck, Real-time dopamine measurement in awake monkeys. PLoS One 9, e98692 (2014). 49. M. K. Zachek, P. Takmakov, B. Moody, R. M. Wightman, G. S. McCarty, Simultaneous decoupled detection of dopamine and oxygen using pyrolyzed carbon microarrays and fast- scan cyclic voltammetry. Anal. Chem. 81, 6258–6265 (2009). 50. D. L. Robinson, B. J. Venton, M. L. A. V. Heien, R. M. Wightman, Detecting subsecond dopamine release with fast-scan cyclic voltammetry in Vivo. Clin. Chem. 49, 1763–1773 (2003). 51. A. Brajter-Toth, K. A. El-Nour, E. T. Cavalheiro, R. Bravo, Nanostructured carbon fiber disk electrodes for sensitive determinations of adenosine and uric acid. Anal. Chem. 72, 1576–1584 (2000). 52. C. Yang, Y. Wang, C. B. Jacobs, I. N. Ivanov, B. J. Venton, O2 plasma etching and antistatic gun surface modifications for cnt yarn microelectrode improve sensitivity and antifouling properties. Anal. Chem. 89, 5605–5611 (2017). 21 53. Z. Moktadir, S. Hang, H. Mizuta, Defect-induced fermi level pinning and suppression of ambipolar behaviour in graphene. Carbon 93, 325–334 (2015). 54. S. Massabeau, M. Baillergeau, T. Phuphachong, C. Berger, W. de Heer, S. Dhillon, J. Tignon, L. A. de Vaulchier, R. Ferreira, J. Mangeney, Evidence of fermi level pinning at the dirac point in epitaxial multilayer graphene. Phys. Rev. B 95, 085311 (2017). 55. V. M. Pereira, F. Guinea, J. M. B. L. Dos Santos, N. M. R. Peres, A. H. C. Neto, Disorder induced localized states in graphene. Phys. Rev. Lett. 96, 036801 (2006). 56. B. Chazotte, Labeling golgi with fluorescent ceramides. Cold Spring Harb. Protoc. 2012, 892–894 (2012). Acknowledgements: This research used resources of the Center for Functional Nanomateri- als, which is a U.S. DOE Office of Science Facility, at Brookhaven National Laboratory under Contract No. de-sc0012704. We also acknowledge the Surface Science Facility of CUNY Ad- vanced Science Research Center for the use of the XPS tool. DS acknowledges partial financial support by NSF-CMMI award 1728051. RK is supported by the National Institute of Men- tal Health grant R01MH109180, a Pew Scholarship in the Biomedical Sciences, and Simons Collaboration on the Global Brain. Author Contributions T.W., A.A., D.S., R.K. designed research; T.W., A.A., D.S. performed research; T.W., A.A., D.S., R.K. analyzed data; and T.W., A.A., D.S., R.K. wrote the paper. Competing Interests The authors acknowledge the following patent applications: U.S. Serial No. 62/599,303 and U.S. Serial No. 62/539,045. 22 Figure 1: FSCV sensors made of multilayer graphene films with different amounts of structural defects. (a) Schematics of a graphene-based electrode used for FSCV measure- ments. The multilayer graphene electrode was mounted on a SiO2/Si substrate and connected to a Cr/Au contact. A fluidic chamber filled with PBS solution of target biochemicals (dopamine or serotonin neuromodulators) was made around the sensor. To maintain neuromodulator concen- tration at a desired level, fresh solution was brought to the chamber by an inlet and old solution was taken out by an outlet. (b) The SEM image shows an example of our miniaturized graphene- based sensor array and fluidic chamber around the sensors. We used nanofabrication to build miniaturized sensors from our candidate CVD and graphitized films. Scale bar is 300 µm. (c) Topographic image of an example CVD multillayer graphene sensor and its thickness measured by atomic force microscopy. Scale bar is 5 µm. (d) In FSCV measurements of dopamine, the voltage is applied to the sensor electrode; it first quickly ramps up, which oxidizes dopamine to dopamine-o-quinone, and then ramps down, which reduces it back to dopamine. The resulting current is measured. (e) Area-normalized electrochemical current (IEC) as a function of time in one FSCV cycle for four sample electrodes made of CVD multilayer graphene films and carbon fibers. The noticeable variations of IEC for these sensors in response to the same dopamine concentration highlight the critical role of structural defects on sensitivity. 23 Figure 2: Quantifying structural defects in graphene electrodes. (a) Line and point de- fects are typically present at the same time in synthetic multilayer graphene films and can be characterized by the average crystallite size, La, and average distance between point defects, LD, respectively. (b) To evaluate the structural properties of the CVD films, we used Raman spectroscopy. The increase of the D peak intensity for samples from bottom to top indicates the higher density of sp2-hybridized defects. The gradual changes of the Raman peaks also high- light the transition from a graphitic structure in stage (i) into a fully disordered sp2 carbon in stage (ii). Conventional CFs are typically in stage (ii) as shown by the example electrode in the top row. Area-normalized sensitivity, SA, is indicated for each sample electrode. (c) Spatially resolved La and LD across the sensor surface for an example CVD electrode. To estimate the average density of point and line defects, we obtained similar spatial maps for all electrodes 2 a) and the average studied in this work. (d) The scatter plot of the average crystallite area (L −2 D ) shows that our candidate materials covered a broad range of defect point defect density (L densities. Numbers next to the CVD samples indicate example electrodes in panel b. The yel- low shading represents the stage (ii) of the amorphization trajectory, while the gray box marks the detection limit of Raman for estimating the point and line defect concentration. 24 a and L Figure 3: Revealing the link between structural defects and electrode sensitivity. (a) Con- −2 2 tour plot of SA versus L D , indicating that the sensitivity of the CVD sensors in stage (i) was largely independent of the average density of line defects and was amplified by increasing the average density of point defects. (b) We found that SA of the CVD sensors in stage (i) was amplified in linear proportion to the density of point defects, and dropped upon transition into stage (ii) (yellow shading). The dashed line represents the measurable limit of SA in our sensor readout system. The sensitivity of electrodes from pristine graphene was below the measurable limit. 25 Figure 4: Engineered multilayer graphene sensors with predictable sensitivity. (a) To con- firm the connection between the area-normalized sensitivity and the density of point defects, we produced engineered multilayer graphene through nickel-induced graphitization of diamond- like carbon (DLC). (b) XPS measurements confirmed the graphitic nature of the films through the presence of sp2-hybridized C-C bond peak highlighted in red. (c) We used Raman mea- surements to quantify the structural defects in our graphitized samples. (d) We created three groups of graphitized samples. A first group that had overlapping density of point defects with the CVD sensors; a second group which contained a higher density of point defects than the CVD samples without transitioning to stage (ii); and a third group that had even higher density of point defects and were in stage (ii) (yellow shading). Numbers next to the data points denote the example GR (I,II,III,IV) and CVD (1,2,3) electrodes in panel c and Fig. 1b, respectively. (e) From the FSCV measurements of dopamine, we confirmed that the electrode sensitivity of the GR samples was largely independent of the crystallite size and increased with the point defect concentrations. (f) We found that GR sensor samples with a similar density of point defects to CVD samples had the same SA. By increasing the density of point defects in stage (i), we achieved a maximum SA of about 177 pA.µm−2.µM−1, which is 20 times higher than conven- tional CFs. The electrode sensitivity decreased rapidly once the structure of the carbon lattice transitioned into stage (ii). The dashed line denotes the minimum measurable limit of SA in our experiments. 26 Figure 5: Role of oxygen-containing functional groups on electrode sensitivity. To in- vestigate this, we performed XPS on multiple multilayer graphene electrodes with markedly different area-normalized sensitivity. (a) The XPS measurements were performed immediately after FSCV neurochemical measurements in a UHV environment. (b) The zoomed-in plot of the region highlighted by the gray shading in panel (a), showing the fitted curves for the differ- ent carbon-oxygen peaks. For better illustration, the intensity was also increased by a factor of 5. Side-by-side comparison of these samples revealed that oxygen-containing functional groups have no direct effect on the electrode sensitivity of multilayer graphene electrodes in stage (i). 27
1910.07249	1	1910	2019-10-16T09:46:38	Stumbling Through the Research Wilderness, Standard Methods to Shine Light on Electrically Conductive Nanocomposites for Future Health-Care Monitoring	[ "physics.app-ph", "cond-mat.soft" ]	Electrically conductive nanocomposites are an exciting ever expanding area of research that has yielded many new technologies for wearable health devices. Acting as strain sensing materials, they have paved the way towards real time medical diagnostic tools that may very well lead to a golden age of healthcare. Currently, the goal in research is to create a material that simultaneously has both a large gauge factor G and sensing range. However, a weakness in the area of electromechanical research is the lack of standardisation in the reporting of the figure of merit, i.e. G, and the need for new metrics to give researchers a more complete view of the research landscape of resistive type sensors. A paradigm shift in the way in which data is reported is required, to push research in the right direction and to facilitate achieving research goals. Here, we report a standardised method for reporting strain sensing performance and the introduction of the working factor W and the Young's modulus Y of a material as two new material criteria. Using this new method, we can now for the first time define the benchmarks for an optimum sensing material, G > 7, W > 1, Y < 300 kPa, using limits set by standard commercial materials and the human body. Using extrapolated data from 200 publications normalised to this standard method, we can review what composite types meet these benchmark limits, what governs composite performances, the literary trends in composites and individual nanomaterial performance and the future prospects of research.	physics.app-ph	physics	Stumbling Through the Research Wilderness, Standard Methods to Shine Light on Electrically Conductive Nanocomposites for Future Health-Care Monitoring Conor S Boland School of Mathematical and Physical Sciences, University of Sussex, Brighton, BN1 9QH, United Kingdom [email protected] Abstract Electrically conductive nanocomposites are an exciting ever-expanding area of research that has yielded many new technologies for wearable health devices. Acting as strain sensing materials, they have paved the way towards real-time medical diagnostic tools that may very well lead to a golden age of healthcare. Currently, the goal in research is to create a material that simultaneously has both a large gauge factor (G) and sensing range. However, a weakness in the area of electromechanical research is the lack of standardisation in the reporting of the figure of merit (i.e. G) and the need for new metrics to give researchers a more complete view of the research landscape of resistive-type sensors. A paradigm shift in the way in which data is reported is required, to push research in the right direction and to facilitate achieving research goals. Here, we report a standardised method for reporting strain- sensing performance and the introduction of the working factor (W) and the Young's modulus (Y) of a material as two new material criteria. Using this new method, we can now for the first time define the benchmarks for an optimum sensing material (G > 7, W > 1, Y < 300 kPa) using limits set by standard commercial materials and the human body. Using extrapolated data from 200 publications normalised to this standard method, we can review what composite-types meet these benchmark limits, what governs composite performances, the literary trends in composites and individual nanomaterial performance and the future prospects of research. Introduction Over the past decade, nanoscience has dominated the landscape of research. From electronics[1-4], energy storage[5-8] to bioengineering[9-12], its profound influence in driving forward innovation has been felt in every corner of science. Recently, nanoscience and more specifically composites filled with nanomaterials, has drawn the attention of many researchers as tactile and health sensors due to them possessing sensitivities which far surpass that of current commercial sensing devices.[13-15] The preeminent goal of research is the creation of an 'all-in-one' sensing material that provides both high sensitivities and large working ranges.[16, 17] But, is such a material even possible? Due to a lack of performance criteria, current research in electromechanics is done so thintelligently, akin to wandering into a dark woods with no torch. With no way to see the paths taken by others in previous works due to a lack of well-defined standard procedures for reporting results, it is difficult for researchers to relate findings of others to their own or to even get an indication of how the field is progressing.[18] This leaves researchers trying to navigate the research wilderness by attempting to increase the most relatable yet least defined metric available, the gauge factor. But in doing so, we only lose sight of the purpose of our journey; the practicality of material applications. In this review, we seek to shine a light on the need for standardisation in which data is reported in the field of electromechanics and how through this method we can create a research compass to point tech-paths in the right direction. With the reporting of the method of measuring gauge factor (G) varying greatly from article to article, a standardised method must be introduced to establish consistency. In considering this, when looking at the functionality of materials in application, a clear definition between strain sensing ranges and mechanical strain tolerance must also be made. To overcome this, we use a quantitative numerical value for the functional sensing range of materials, the working factor (W), as a new metric for defining the limitations of the electromechanical response for a given composite. Finally, it must be noted that one of the most overlooked aspects of these nanocomposites is the importance of Young's modulus (Y). In order to be applied in future wearable health technologies, these materials are required to not only have large values for G and W but also very small values for Y. Using extrapolated literaturary values for G, W, and Y from 200 research articles normalised to this standard procedure, we can derive information and achieve a level of understanding of electromechanical research that was previously inaccessible. For the first time we are able to identify the trends and target values in electromechanics, guiding future research out of the wilderness. Results From Composites to Sensing Materials Composite materials, currently, are one of the most well studied areas in materials science research. The search for creating lighter, stronger materials is not a new revelation. In fact, it was a journey of discovery that begin thousands of years ago, with Neolithic people adding straw and wooden sticks to reinforce mud structures. With unparalleled mechanical strength, large aspect ratios and surface areas and the advances in their processability[19-22], it is to no surprise that researchers have turned towards nanomaterials, and the likes of carbon nanotubes (CNTs) and graphene, to use as filler materials to achieve such goals. With the addition of small amounts of material into a polymeric matrix we see just that, increases in the mechanical properties of the polymer.[23] As these two aforementioned materials are also electrical conductors, with the addition of further material, polymers are found to transition from insulators to conductors of electricity when a critical loading level is reached in accordance with percolation theory.[24] Most interestingly, if one were to use a soft pliable material, such as an elastomer, stretchy nanocomposite materials with similar increases in electrical conductivity can be made.[13] Upon the application of strain, these materials display a reversible change in electrical resistance. Essentially, these materials are strain sensors. This change in resistance occurs due to the deformation of the conductive network with the housing matrix (Fig. 1A-B). As strain is applied to the composite, the network of conductive fillers move out of contact with one another and back again upon release. Such behaviour has been extensively investigated for CNTs and graphene-based composites[23, 25, 26] and in using that knowledge, research has expanded to demonstrate such behaviour in composites using a wide range of other nanofillers, such MoS2[27], carbon black[28], carbon fibres[29], MXenes[30], AgNWs[31] and AuNWs.[32] Typically, nano-based composite materials present values of gauge factor (G) ≤ 40[15], much improved over commercial metal foil strain gauges, which generally possess values between 2-7. Not often noted, nanocomposites also have an added advantage mechanically, with metal foil sensors typically made using high stiffness polymer backings that only deform to strains up to ~5% before fracture and would thus not be suitable for wearable electronics. However, with such high sensitives and flexibility, nanocomposites on the other hand have been applied as wearable bodily sensors in many forms, measuring such biomechanical functions as joint bending[33], breathing[34], pulse[35] and blood pressure.[15] Though this implies that modulus maybe an important feature of nanocomposite based electromechanical sensors, no standard has yet been set. Currently, the gauge factor is the only accepted metric for quantifying the performance of these materials. But, is this enough to determine what is a good sensor or nanomaterial or polymer matrix for strain sensing and what is not? In fact, if we look at the use of nanomaterials in other applications, such as in Li-ion storage and clean energy harvesting, electromechanics falls flat in comparison to their regimented reporting. For battery electrode composite materials, there are a number of metrics that define performance; the conductivity, volumetric capacity, energy density, coulomb efficiency and charge/discharge times.[36] For catalyses, there is the conductivity, tafel slope, overpotential and rate of production.[37] However, in electromechanics, there is only one figure of merit but one metric is not quite enough to determine the quality of performance. By using the identification of key material properties and setting benchmarks as our light forward, we can interpret past results into a road map to help progress electromechanical research and outline what the attainable future goals are. Defining Performance As stated above, the current and only metric for which researchers can compare the performances of their strain sensing composites, is the gauge factor (G). Starting at low strain, resistance change can be described by ∆R 𝑅0 = G ∙ ε (Eq. 1) This equation describes a linear relationship between the fractional resistance change (∆𝑅 𝑅0⁄ ) and strain (ε), both dimensionless, during the deformation of the material, where G is the slope (Fig. 1C). This, however, in the past has been misinterpreted as the slope at any point along the response curve yielding a value for gauge factor. Many literary sources in fact quote values of G beginning at strains >>100%, where the matrix has yielded and the conductive network has undergone such extreme rarefication that it is artificially highly sensitive to strain due to being on the verge of failure. However, this is not the correct reporting of the metric for a material and does not allow for the reported figure of merit to be relatable to others. In terms of application, the reporting of such values is not practical, as to utilize such sensitivities it would imply that materials would need to be prestrained to 100s of percent strain. In terms of device development, having a polymeric component held at such high strain would also not be feasible due to mechanical creep.[38] Critically, Eq. 1 describes a relationship with an intercept of zero and only holds for the initial linear response of the material beginning at low strain. Points beyond this initial linear region, no longer satisfy this relationship, as fractional resistance change becomes non-linear. It is important to clarify this point, simple as it maybe, because in future applications of these composites, this proportionality will be required to calibrate electrical response to strain precisely, and vis versa. In application, these materials are envisioned as future wearable health technologies. To function as such, they would need to measure low strain stimuli such as ones associated with pulse (ε ~ 2%)[15] to strains as large as the ones associated with the bending of a joint (ε ≥ 100%).[39, 40] Thus, looking at our response curve and our linear fit for G in Fig. 1C, this would imply that a desired property of these materials would be to have a linear region or working range that extends up to ε > 100%, to facilitate the measurement of both types of stimuli (i.e. an all-in-one material). With such a measurement organically arising from the fit for gauge factor, we propose that the introduction of a new metric to allow researchers to quantitatively measure the working range, the working factor (W), be set as a minimum standard of reporting. We define W simply as the strain limit (in absolute strain) at which the fractional resistance is no longer linear with strain (Fig. 1C). The introduction and adoption of such a metric would allow researchers for the first time be able to compare what filler materials and polymer systems allow for a response measurement over an extended strain range. Stiff Competition Again, keeping in mind the end goal of these materials, certain stipulations are implied. To function as skin-on sensors, the nanocomposites will be required to be very thin to accommodate adhering to the epidermis.[41] Additionally, much like the human body, the materials also are required to be soft and compliant in nature, thus having a low elastic modulus.[42] However, the mechanics of sensing materials is something that goes largely unreported in literature (see SI). Take the measurement of pulse for example (Fig. 2), this non-invasive measurement generally consisting of the sensing material being attached to the body on the wrist or neck. The artery exerts a tensile strain on the skin, which can be detected by the sensor as it deforms with the skin. As the gauge factor of our material increases, we subsequently receive more clarity in the outputted signal, with more distinguishable features of the waveform apparent. However, if a material is too stiff, this will dampen the electromechanical signal as it inhibits the deformation, regardless of gauge factor.[43] As the materials becomes stiffer and stiffer, the signal is lost in the background noise all together. As arteries only exert a small finite pressure (~5kPa)[15], there a limit to the modulus a sensing material can have that has yet to be defined. The human body, however, has set the limits for us. Looking at the mechanical properties of skin and ligaments, respectively, they possess stiffness of ~300 kPa and ~700 kPa.[44-46] To measure unimpeded signals, the composite thus requires a stiffness comparable to that of the skin of the human body, i.e. Y < 300 kPa.[41, 42] In terms of functionality, as wearables, if the sensors are of a Young's modulus greater than that of the limit our bodies has set, it will plainly be too uncomfortable for the user.[47, 48] As the materials are envisioned to seamlessly be adhered to or integrated into worn clothing on the body, if they restrict motion or movement, it defeats the purpose of application. Looking back at our previous criteria, essentially, the idyllic all-in-one sensing material would be one which has G > 7, W > 1, Y < 300 kPa. Discussion Gauge Factor vs Working Factor: A Negative Correlation Using the standardised procedure described above, values for G, W, and Y were extrapolated from 200 published sources reporting the use of a nanocomposite as a sensing material, indiscriminate of filler or matrix type (see SI, Table S1). Firstly, we plot G as a function of W in Fig. 3A. On the same plot, the goal region, identified as a sensing material possessing G > 7 (the gauge factor limit of a commercial sensor) and W > 1 is overlaid. We set this limit for the working factor in accordance with biomechanical measurements. Joints and muscles can undergo inflections up to ε ≥ 100%, so a material with W > 1 would satisfy any measurement with an associated strain inclusive in that range. Currently, only two such materials fall within this zone, both of which utilise highly aligned CNT networks.[49, 50] For composite materials, the values for sensitivity overall fell with increasing sensing limit, from G ~ 3000 at W ~ 0.001 to G ~ 0.7 at W ~ 14. In what is an unexpected result, this decay in sensitivity in literature is found to follow a very well-defined power-law, with gauge factor universally scaling as G ∝ 1/√W. Unfortunately, the underlying reasons for this inverse root dependence cannot be fully understood without a systematic study performed. However, though this is the first time such a scaling has been reported, the relation may not come as a surprise. Yielding Behaviours To understand the relationship between G and W, we must first discuss the mechanisms which controls the characteristics of the response curve and which yields a regime change from linear to non- linear behaviour. As previously stated, all composite materials, including the ones whose datasets have been sampled from for this manuscript, follow the same basic principles as described previous in Fig. 1A; the changing of inter-particle connections resulting in a change of composite electrical resistance with strain. Beginning at low strain, this resistance change is linear and dominated by conductors moving out of direct contact with one another (i.e. interfacial overlapping decreases). At a critical point, the potential barrier between adjacent materials starts to increase. Tunnelling effects then start to play a dominant role upon reaching the critical strain, with tunnelling resistance increasing exponentially with continual deformation of the composite.[51] As the average distance between filler material grows, the resistance's response to strain becomes nonlinear.[51-53] Thus, at low loading levels, though a conductive network will be more sensitive to applied strain due to fewer connections in the network, it will also lead to a lower critical limit in the linear response (i.e. working factor).[51] At the critical point, it is suggested that the strain near the tunnel gaps will be large and non-uniform so the local response may exceed the elastic linear limit at the filler interface[54, 55], corresponding to the yielding of the polymer material and an increase in the gap.[56] This implies that the working factor is proportional to that of the yield strain for a given composite system. Dispersal Effects It is well known that for composite systems, material quality and loading level can give rise to aggregation effects that effect the network structure.[57, 58] Thus, dispersion factors can result in regions of high and low conductor density which have been reported to greatly affect the linearity of a composite system's electrical response to strain.[14, 59] Poor dispersal of conductive material can lead to a scenario where network effects now dominate the linearity of the response rather than the matrix, which would lead to a less than optimum working range for a particular system. Or in other words, a working range that is lower than the maximum value set by the polymer and it's yield strain. Aggregation and dispersion factors are a common problem in most mixed phase and surface deposition systems and when they occur it can lead to the evolution of "bottle-necking" between the areas of high and low density in the network with applied strain, leading to non-linear response.[59] This can commonly be overcome though through increasing the loading levels of the filler[51, 59]; however, this is a double-edged sword. As previously mentioned, at percolation, gauge factor is at it's highest and the addition of more material to a network, decreasing its sensitivity to deformation. Conversely, though increasing the loading level would also result in an increase in the working factor, it would also increase the Young's modulus. Orientation Dependence We can use this connection between yield strain and the working factor and apply it to the Kraus model[60, 61] and from this discuss how we can exert control over the gauge factor of a system. This model considers the effect of oscillatory shear strain on a network structure and relates connection changes between fillers with the applied strain and the yield strain. The number of inter-particle connections per volume, N, as a function of applied strain, , can be described as: N = N0 1+( ε εc ) 2m (Eq. 2) Where, N0 is the initial number of inter-particle connections per volume and m is a constant related to specific fractal dimensions of the fractal agglomerate structures and is equal to ~ 0.5. This constant is mainly a geometrical quantity of the filler network and agglomerates, independent of the specific filler or polymer type and the reasons for the universality remaining unclear.[61] c is also a constant related to the yield strain, or in this case the working factor, and is given by 𝑊 ≅ εc = (kr ⁄ ) kb 1 2m (Eq. 3) Where kr and kb are the rate constants for reformation and breaking of connections. Interestingly, the Kraus model also implies that the yield strain will be constant for all loading levels of a given composite material regardless of conductor size[60, 61]. This confirms the previous assumption that the polymer's yield strain controls the maximum attainable working factor. The resistivity of a system can be related to conductor interconnectivity through a percolation-like relationship 𝜌 ∝ (𝑁 − 𝑁0)𝑛𝜀 (Eq. 4) Where nε is a percolation exponent. From our previous work[15], using Eq. 2 and 4, an expression that describes a relationship between resistivity and the yield strain (working factor) can be given by 𝜌 = 𝜌0 2𝑚 ) −1 ) + 𝜀 𝜀𝑡 ] [(1+( 𝜀 𝑊 𝑛𝜀 (Eq. 5) Where t is a composite parameter descripted by εt = Λ𝑁0 ⁄ (Eq. 6) 𝑘2 And Λ is the strain rate and k2 is the rate constant for time dependent connections in the network. Eq. 5 describes both the destruction and reformation of conductor connections with strain within a composite material of very low matrix viscosity, which leads to conductor mobility. For composite materials with relatively high viscosities, leading to immobilised conductors with a reconnection constant approximately equalling zero, resistivity can be rewritten as[15] 𝜌 = 𝜌0 (1 + ( 𝜀 𝑊 2𝑚 ) 𝑛𝜀 ) (Eq. 7) This expression implies that the application of strain reduces the number of inter-particle connections, which results in an increase in resistivity. Furthermore, this equation can be used to show that gauge factor is related to the working factor through the following relationship[15] G = 2+ nε ( 1 W ) (Eq. 8) Looking at Eq. 8, it implies that for large values of G, the working factor is intrinsically low. This describes a similar behaviour between G and W that was observed in Fig. 3A. Showing the validity of Eq. 7, several datasets from literature are fitted and using Eq. 8, values for G derived (Fig. S1 & S2). Eq. 7 was found to fit the literary data well up to the yield strain (Fig. S1) and when comparing the calculated values using Eq. 8 to the previously measured values as a function of W in Fig. S2, both are found to be in close agreement, as well as the inverse root dependence with W reaffirmed. Interestingly, from Eq. 8, G shows a strong dependence on the percolation exponent from Eq. 4, nε, where an increasingly larger value of nε would yield an increase in a composite's gauge factor. If we require a sensing material with at least W ≥ 1 and G ≥ 7, this would mean that a value of nε ≥ 5 would be required. From the fitted datasets, the calculated magnitude of nε would seem to be dependent on the ordination of the filler. Lin et al.[62] describes a hybrid conductor system that lies in-plane with the direction of the applied strain and a calculated value of nε ~ 5. Whereas for a highly randomised system like what is reported in Wang et al[63], nε is ~ 0.95. An intermediate would be Arif et al.[64], which describes a semi-structured filler system and a value of nε ~ 1.7. Essentially, in order to maximise gauge factor in a random well-dispersed system where a maximum value of working factor is achieved, one would need to improve orientation of the conductive network. However, this could prove to be a delicate balancing act as alignment of filler material would negatively impact the mechanics of the composites . In CNT nanocomposites, alignment of the fillers in the polymer matrix was found to further increase the modules by a factor of five.[22] For the purpose of the nanocomposites described here, this could be problematic if a pristine polymer has a base Young's modulus approaching that of skin, particularly as the addition of nanofillers to the matrix would push those values even closer or beyond a workable range. For future works, researchers would need to choose base elastomeric polymer matrix materials which have values of Young's modulus much lower than that of skin to allow room for mechanical variations. Additionally, alignment of fillers in a matrix can increase the electrical percolation threshold, thus requiring a higher loading level of filler material to allow for conduction of electrical current.[65] This in itself poses a problem as the mechanical properties can increase with increased loading levels or alternatively the increased loading levels could lead to higher levels of filler aggregation[22], which would negatively affect the working factor. In general, composites from literature which control the orientation of conductive nanomaterials through lay deposition, fibre formation or microstructures, were found to have some of the higher values for W whist also having relatively high values for G.[49, 50, 66-69] This dependency on network structure in nε is not too dissimilar to that of the network dimensionality dependency seen for the percolation exponent in systems where percolation-like scaling in conductivity is described by the loading level of conductors in a network.[70-73] The unexpected effects network dispersion and orientation have on G and W explain some of the scatter in the literary data reported in Fig.3A. As the parameters are largely random throughout the data set, this would lead to values of G that would be higher than expected for particular values of W and vis versa. Gauge Factor vs Young's Modulus and the Lay of the Land The relationship between G and W presents quite a challenge for researchers to push back against and will require a refined approach to engineering a more complete sensing material that allows both values to be maximised. However, G's variation with W is only one side of the story. Plotting G as a function of Y in Fig. 3B, again it is seen that the vast majority of datasets lie outside of the goal region, defined as G > 7 and Y < 300 kPa (the Young's modulus of skin). Unlike the plot in Fig 3A, the data falls in a cloud pattern, with no discernible trend and no intuitive link between the two metrics found. Combining the three datasets (G, W, and Y) into a 3-D master plot in Fig 4 however, we can truly see the lay of the land of electromechanics. From the surveyed literature, to the best of the author's knowledge, no reported sensing material possesses values that match that of the proposed optimum material. However, it would be overly critical not to note that before this review, such a large volume of data had not been laid out in such a way to see these trends or the links between these proposed sensing material parameters shown. The closest any material comes to the desired benchmark values for the optimum material is again a highly aligned/orientated CNT composite with G = 4, W = 2.5, Y = 125 kPa.[74] Conclusions In summary, this review has created a standardisation in the reporting of sensing material results that facilitates the communication and comparing of data and a method to which researchers can use to gauge the practicality of a material's application. Through values derived from 200 literaturary sources using this method, we were able to identify the challenges for the research area; engineering materials that have both large values for G and W and with low values of Y. From this, we find that G universally scales as G ∝ 1/√W, inferring that sensing performances of composites maybe capped. Furthermore, we discuss how this relationship might controlled by the ordination, dispersive effects and/or the loading level of conductors in the conductive network, though further studies would be required. From this literary study, we find that to the best of the author's knowledge, no material to date meets the require benchmarks for an all-in-one material. Outlook Nevertheless, despite the previous lack of definitive performance criteria, is electromechanics stuck in a research wilderness? Well, uniquely, from using the standard method described here and the large volume of data extrapolated, interesting statistics that were previously inaccessible to researchers can be derived. Using the standardised method as a form of clarity, trends in sampled data over time in Fig. 5 from the years 2008 to 2019 show a steady overall increase in G and W values independently of one another (Fig. 5A & B respectively). G over time is found to increase from ~14 to ~100, while W on average goes from ~0.34 to ~1.2. Conversely, in Fig. 5C, an overall fall in values for Y from 800 MPa to 150 MPa was observed over time. Research is instinctively going in the right direction, however, these trends in the literary data truly shows the need for the introduction of W and Y as new standard metrics to tie everything together when considering the data trends reported here. Overall, the average nanocomposite strain sensor is found to have a gauge factor of G ~ 41 (very close to the previously assumed value), a Young's modulus of Y ~ 300 MPa and a working strain of W ~ 0.62. To compare how composites prepared using different nanomaterials perform in comparison to one another, in Fig 6, we calculate and plot the average values for G, W and Y for each nanomaterial type. In terms of materials used, graphene and graphene-like materials were the most common, followed by carbon nanotubes and multi-walled carbon nanotubes (Fig. 6A). It is to be noted, that ten materials had a count value ≤ 3, six of which only had one count. For the forthcoming ranking of average nanomaterials parameters in literature, they are deemed to not be statistically reliable due to their low count number and are denoted by an asterisk in the following figures. In Fig. 6B, on average, hybrid- based materials (those that which utilised two or more nanomaterials) had the highest average gauge factor (G ~ 200). Graphene and graphene-like materials (G ~ 50) reported the second highest literary value, followed by CNT-based composites (G ~ 35). In terms of working factor (Fig. 6C), hydrogels were found to have the largest sensing ranges, at W ~ 1.2, which was followed by conductive polymers (W ~ 0.6) and CNT-based materials (W ~ 0.55). Hydrogels, again, were found to also have the lowest average Young's modulus (Y ~ 200 kPa) in Fig 6D. Amorphous carbon materials were found to have the second lowest (Y ~ 3 MPa) and AgNP composites the third (Y ~ 8 MPa). In Fig. 7, we make a 3-D master plot using the values of the three metrics (G vs W vs Y) for the individual nanomaterials. Similar to Fig 4, no nanomaterial has an average value that lies within the current goal range of the all-in-one sensing material, though hybrid-based materials would come the closest. Future Challenges The core challenge for research remains the same for the foreseeable future, i.e. to identify materials which have G > 7, W > 1, Y < 300 kPa. However, though the reported negative correlation between G and W does cast doubt on whether such an all-in-one material is achievable the modified Kraus model predicts that G and W can be greatly improved through engineering of the conductive filler network structure. An example of such engineering is work done by Pu et al., which demonstrated CNT/ethylene-α-octene block copolymer (OBC) nanocomposites that have a very well dispersed conductive network structure in the polymer matrix.[65] Using the author's definition of working factor and applying it to this study, it can be seen that not only does the yield strain appear to match that of the linear limit of the electromechanical response but is also constant for the composite system. Additionally, when the orientation of the CNTs was altered, so that all the fillers were aligned in one direction, the gauge factor dramatically increased. The model here wholly predicts the electromechanical behaviour of the nanocomposites in this study and proves that utilising it may in fact be an intuitive way forward in improving and understanding performances. Interesting, this model also uniquely gives researchers the opportunity to possibly revisit nanocomposite materials that had exceptional properties in the past and further optimise their performance through the better sense of perspective provided by this review. Playing devil's advocate, perhaps the argument of all-in-one materials is a red herring, as plenty of sensing materials can measure small stimuli and others large. This problem of measuring both in the future could simply come down to using two types of sensing materials in conjunction with one another. A very much achievable goal for current research would be the study of signal hysteresis and conditioning, both largely unreported phenomena but equally important for commercialisation. A good sensing material would need to report a response that is undiminished with cycle number but any links between materials properties and low hysteresis has yet to be drawn. Regardless of goal, this review gives us collectively as researchers the ability to take previous studies and use them as a springboard to progress the field. Only through the understanding of the limitations of our past work can we move the field forward with perspective, towards real application potential, towards the forest edge. Methods Data from literaturary sources was extracted and fitted using the "Digitizer" function in Origin 2019 software. As calculated values are extrapolated from published and not raw data, all values are to be taken as an estimate. Only sources that reported or presented data that allowed for values of gauge factor, working factor and Young's modulus to be calculated we used in this review. For further information on the derivation of the modified Kraus model see Ref 15. Figures: Figure 1: A) Top, polymer matrix with embedded electrically conductive nanosheet network in its initial state. Bottom, matrix in its strained state resulting in rarefication of the network. B) Typical electromechanical response curve of nanocomposite material being deformed from its initial state (ε=0) to a strained state (ε=0.1). C) Log-log plot of Fig. 1B with linear region fitted with Eq. 1 to extrapolate gauge factor, G=100. The limit to the linear region, and thus its sensing range, is defined as 0.028%, the working factor (W). Figure 2: Effect of increasing gauge factor and Young's modulus on the response curve of a pulse measurement. Figure 3: Gauge factor (G) as a function of working factor (W) and Young's modulus (Y). (A) The target region is designated by the values of G > 7 and W > 1. The dashed line represents an apparent intrinsic decay in the gauge factor of nano-based strain sensors with the working factor described by G ∝ 1/√W. (B) The target region is designated by the values of G > 7 and Y < 300 kPa. Figure 4: Plot of extrapolated literary values for gauge factor (G) vs working factor (W) vs Young's modulus (Y). The challenge region is designated by values of G = 7, W = 1, and Y = 300 kPa. Figure 5: Evolution of gauge factor (G), Young's modulus (Y) and working factor (W) over time for sampled literature. Figure 6: (A) Distribution of nanomaterials used in sampled literary sources. (B-C) Average gauge factor (G), working factor (W) and Young's modulus (Y) of materials samples from literature. Green lines and arrows denote the optimum value for each parameter and to which direction values need to tread towards to achieve surpassing it. Asterisk denotes materials with a count number ≤ 3. Figure 7: Plot of extrapolated literary values for gauge factor (G) vs working factor (W) vs Young's modulus (Y) of each material type sampled. Asterisks denotes materials with a count number ≤ 3. References 1. 4. 5. 6. 7. 8. 2. 3. from graphene-BN-graphene nanosheet Kelly, A.G., et al., All-printed capacitors heterostructures. Applied Physics Letters, 2016. 109(2): p. 023107. Kelly, A.G., et al., All-printed thin-film transistors from networks of liquid-exfoliated nanosheets. Science, 2017. 356(6333): p. 69-73. McManus, D., et al., Water-based and biocompatible 2D crystal inks for all-inkjet-printed heterostructures. Nature Nanotechnology, 2017. 12(4): p. 343-350. Worsley, R., et al., All-2D Material Inkjet-Printed Capacitors: Toward Fully Printed Integrated Circuits. ACS Nano, 2019. 13(1): p. 54-60. Park, S.-H., et al., High areal capacity battery electrodes enabled by segregated nanotube networks. Nature Energy, 2019: p. 1. Zhang, C., et al., High capacity silicon anodes enabled by MXene viscous aqueous ink. Nature Communications, 2019. 10(1): p. 849. Wang, X., et al., Influences from solvents on charge storage in titanium carbide MXenes. Nature Energy, 2019. 4(3): p. 241. Yu, L., et al., MXene-Bonded Activated Carbon as a Flexible Electrode for High-Performance Supercapacitors. ACS Energy Letters, 2018. 3(7): p. 1597-1603. Boutry, C.M., et al., Biodegradable and flexible arterial-pulse sensor for the wireless monitoring of blood flow. Nature Biomedical Engineering, 2019. 3(1): p. 47. Liu, Y., et al., Soft and elastic hydrogel-based microelectronics for localized low-voltage neuromodulation. Nature Biomedical Engineering, 2019. 3(1): p. 58. Oh, J.Y. and Z. Bao, Second Skin Enabled by Advanced Electronics. Advanced Science, 2019. 6(11): p. 1900186. Ryan, A.J., et al., Electroconductive Biohybrid Collagen/Pristine Graphene Composite Biomaterials with Enhanced Biological Activity. Advanced Materials, 2018. 30(15): p. 1706442. Boland, C.S., et al., Sensitive, High-Strain, High-Rate Bodily Motion Sensors Based on Graphene -- Rubber Composites. ACS Nano, 2014. 8(9): p. 8819-8830. Boland, C.S., et al., Surface coatings of silver nanowires lead to effective, high conductivity, high-strain, ultrathin sensors. Nanoscale, 2017. 9(46): p. 18507-18515. Boland, C.S., et al., Sensitive electromechanical sensors using viscoelastic graphene-polymer nanocomposites. Science, 2016. 354(6317): p. 1257. Heikenfeld, J., et al., Wearable sensors: modalities, challenges, and prospects. Lab on a Chip, 2018. 18(2): p. 217-248. Amjadi, M., et al., Stretchable, Skin-Mountable, and Wearable Strain Sensors and Their Potential Applications: A Review. Advanced Functional Materials, 2016. 26(11): p. 1678-1698. 18. Millstone, J.E., et al., Redefining the Experimental and Methods Sections. ACS Nano, 2019. 13. 10. 15. 16. 14. 9. 11. 12. 17. 19. 20. 21. 22. 23. 24. 25. 26. 13(5): p. 4862-4864. Hernandez, Y., et al., High-yield production of graphene by liquid-phase exfoliation of graphite. Nature nanotechnology, 2008. 3(9): p. 563-568. Paton, K.R., et al., Scalable production of large quantities of defect-free few-layer graphene by shear exfoliation in liquids. Nature Materials, 2014. 13(6): p. 624-630. Nicolosi, V., et al., Liquid Exfoliation of Layered Materials. Science. 340(6139): p. 1226419- 1226419. Coleman, J.N., et al., Small but strong: A review of the mechanical properties of carbon nanotube -- polymer composites. Carbon, 2006. 44(9): p. 1624-1652. Kinloch, I.A., et al., Composites with carbon nanotubes and graphene: An outlook. Science, 2018. 362(6414): p. 547-553. Stankovich, S., et al., Graphene-based composite materials. Nature, 2006. 442(7100): p. 282- 286. Gong, S. and W. Cheng, One-Dimensional Nanomaterials for Soft Electronics. Advanced Electronic Materials, 2017. 3(3): p. n/a-n/a. Duan, L., et al., The resistivity -- strain behavior of conductive polymer composites: stability and sensitivity. Journal of Materials Chemistry A, 2014. 2(40): p. 17085-17098. 27. 28. 29. 30. 31. 32. 33. Biccai, S., et al., Negative Gauge Factor Piezoresistive Composites Based on Polymers Filled with MoS2 Nanosheets. ACS Nano, 2019. Narongthong, J., et al., Ionic liquid enabled electrical-strain tuning capability of carbon black based conductive polymer composites for small-strain sensors and stretchable conductors. Composites Science and Technology, 2019. 174: p. 202-211. Shui, X. and D.D.L. Chung, A piezoresistive carbon filament polymer-matrix composite strain sensor. 5(2): p. 243-243. Yang, Y., et al., Strain Sensors with a High Sensitivity and a Wide Sensing Range Based on a Ti3C2Tx (MXene) Nanoparticle -- Nanosheet Hybrid Network. Advanced Functional Materials, 2019. 29(14): p. 1807882. Kim, S.-R., J.-H. Kim, and J.-W. Park, Wearable and Transparent Capacitive Strain Sensor with High Sensitivity Based on Patterned Ag Nanowire Networks. ACS Applied Materials & Interfaces, 2017. 9(31): p. 26407-26416. Ho, M.D., et al., Percolating Network of Ultrathin Gold Nanowires and Silver Nanowires toward "Invisible" Wearable Sensors for Detecting Emotional Expression and Apexcardiogram. Advanced Functional Materials, 2017. 27(25): p. n/a-n/a. Zhang, H., W. Niu, and S. Zhang, Extremely Stretchable, Stable, and Durable Strain Sensors Based on Double-Network Organogels. ACS Applied Materials & Interfaces, 2018. 10(38): p. 32640-32648. 35. 34. Ma, J., et al., Highly Sensitive and Large-Range Strain Sensor with a Self-Compensated Two- Order Structure for Human Motion Detection. ACS Applied Materials & Interfaces, 2019. 11(8): p. 8527-8536. Larimi, S.R., et al., Low-cost ultra-stretchable strain sensors for monitoring human motion and bio-signals. Sensors and Actuators A: Physical, 2018. 271: p. 182-191. Zhang, C., et al., Enabling Flexible Heterostructures for Li-Ion Battery Anodes Based on Nanotube and Liquid-Phase Exfoliated 2D Gallium Chalcogenide Nanosheet Colloidal Solutions. Small, 2017. 13(34): p. 1701677. 36. 37. McAteer, D., et al., Liquid Exfoliated Co(OH)2 Nanosheets as Low-Cost, Yet High- Performance, Catalysts for the Oxygen Evolution Reaction. Advanced Energy Materials, 2018. 8(15): p. 1702965. 38. Ward, I.M. and D.W. Hadley, An introduction to the mechanical properties of solid polymers. 39. 1993: John Wiley & Sons Ltd . Barberio, C., et al., The effect of shoulder abduction and medial epicondylectomy on ulnar nerve strain: A preliminary study. Journal of Musculoskeletal Surgery and Research, 2019. 3(1): p. 134. 40. Marieswaran, M., et al., An extended OpenSim knee model for analysis of strains of connective 41. 42. 43. 44. 45. 46. 47. 48. tissues. BioMedical Engineering OnLine, 2018. 17(1): p. 42. Xu, S., et al., Soft Microfluidic Assemblies of Sensors, Circuits, and Radios for the Skin. Science, 2014. 344(6179): p. 70-74. Park, Y.-L., et al., Hyperelastic pressure sensing with a liquid-embedded elastomer. Journal of Micromechanics and Microengineering, 2010. 20(12): p. 125029. S. Boland, C., et al., Graphene-coated polymer foams as tuneable impact sensors. Nanoscale, 2018. 10(11): p. 5366-5375. Kim, D.-H., et al., Epidermal Electronics. Science, 2011. 333(6044): p. 838-843. Li, C., et al., Determining elastic properties of skin by measuring surface waves from an impulse mechanical stimulus using phase-sensitive optical coherence tomography. Journal of The Royal Society Interface, 2012. 9(70): p. 831-841. Agache, P.G., et al., Mechanical properties and Young's modulus of human skin in vivo. Archives of Dermatological Research, 1980. 269(3): p. 221-232. Paradiso, R., G. Loriga, and N. Taccini, A wearable health care system based on knitted integrated sensors. IEEE Transactions on Information Technology in Biomedicine, 2005. 9(3): p. 337-344. Park, Y., B. Chen, and R.J. Wood, Design and Fabrication of Soft Artificial Skin Using Embedded Microchannels and Liquid Conductors. IEEE Sensors Journal, 2012. 12(8): p. 2711- 2718. 49. Ma, L., et al., Multi-dimensional strain sensor based on carbon nanotube film with aligned 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60. 61. 62. conductive networks. Composites Science and Technology, 2018. 165: p. 190-197. Darabi, M.A., et al., Gum Sensor: A Stretchable, Wearable, and Foldable Sensor Based on Carbon Nanotube/Chewing Gum Membrane. ACS Applied Materials & Interfaces, 2015. 7(47): p. 26195-26205. Hu, N., et al., Tunneling effect in a polymer/carbon nanotube nanocomposite strain sensor. Acta Materialia, 2008. 56(13): p. 2929-2936. Zhang, R., M. Baxendale, and T. Peijs, Universal resistivity--strain dependence of carbon nanotube/polymer composites. Physical Review B, 2007. 76(19): p. 195433. Knite, M., et al., Polyisoprene-carbon black nanocomposites as tensile strain and pressure sensor materials. Sensors and Actuators A: Physical, 2004. 110(1): p. 142-149. Dawson, J.C. and C.J. Adkins, Conduction mechanisms in carbon-loaded composites. Journal of Physics: Condensed Matter, 1996. 8(43): p. 8321-8338. Chen, L., G.H. Chen, and L. Lu, Piezoresistive Behavior Study on Finger-Sensing Silicone Rubber/Graphite Nanosheet Nanocomposites. Advanced Functional Materials, 2007. 17(6): p. 898-904. Halary, J., et al., Chemically Engineered Carbon Nanotube-Polymer Composite Coatings for use as Remote Strain-Sensors. MRS Online Proceedings Library Archive, 2004. 858. Coleman, J.N., et al., Small but strong: A review of the mechanical properties of carbon nanotube-polymer composites. Carbon, 2006. 44(9): p. 1624-1652. Bauhofer, W. and J.Z. Kovacs, A review and analysis of electrical percolation in carbon nanotube polymer composites. Composites Science and Technology, 2009. 69(10): p. 1486- 1498. Amjadi, M., et al., Highly Stretchable and Sensitive Strain Sensor Based on Silver Nanowire -- Elastomer Nanocomposite. ACS Nano, 2014. 8(5): p. 5154-5163. Kraus, G., Mechanical losses in carbon-black-filled rubbers. 1984: J. Appl. Polym. Sci. Symp. 75-92. Heinrich, G. and M. Klüppel, Recent Advances in the Theory of Filler Networking in Elastomers, in Filled Elastomers Drug Delivery Systems. 2002, Springer Berlin Heidelberg: Berlin, Heidelberg. p. 1-44. Lin, L., et al., Towards Tunable Sensitivity of Electrical Property to Strain for Conductive Polymer Composites Based on Thermoplastic Elastomer. ACS Applied Materials & Interfaces, 2013. 5(12): p. 5815-5824. 63. Wang, Y., et al., Highly stretchable and sensitive piezoresistive carbon nanotube/elastomeric Journal of Materials 64. 65. 66. 67. 68. 69. triisocyanate-crosslinked polytetrahydrofuran nanocomposites. Chemistry C, 2016. 4(3): p. 460-467. Arif, M.F., et al., Strong linear-piezoresistive-response of carbon nanostructures reinforced hyperelastic polymer nanocomposites. Composites Part A: Applied Science and Manufacturing, 2018. 113: p. 141-149. Pu, J.-H., et al., 2D end-to-end carbon nanotube conductive networks in polymer nanocomposites: a conceptual design to dramatically enhance the sensitivities of strain sensors. Nanoscale, 2018. 10(5): p. 2191-2198. Seyedin, M.Z., et al., Strain-Responsive Polyurethane/PEDOT:PSS Elastomeric Composite Fibers with High Electrical Conductivity. Advanced Functional Materials, 2014. 24(20): p. 2957-2966. Losaria, P.M. and J.-H. Yim, A highly stretchable large strain sensor based on PEDOT -- thermoplastic polyurethane hybrid prepared via in situ vapor phase polymerization. Journal of Industrial and Engineering Chemistry, 2019. 74: p. 108-117. Kim, H.-J., A. Thukral, and C. Yu, Highly Sensitive and Very Stretchable Strain Sensor Based on a Rubbery Semiconductor. ACS Applied Materials & Interfaces, 2018. 10(5): p. 5000-5006. Liu, Q., et al., High-Performance Strain Sensors with Fish-Scale-Like Graphene-Sensing Layers for Full-Range Detection of Human Motions. ACS Nano, 2016. 10(8): p. 7901-7906. Sahimi, M., Applications Of Percolation Theory. CRC Press, 1994. 70. 71. McLachlan, D.S., M. Blaszkiewicz, and R.E. Newnham, Electrical Resistivity of Composites. Journal of the American Ceramic Society, 1990. 73(8): p. 2187-2203. 72. Weber, M. and M.R. Kamal, Estimation of the volume resistivity of electrically conductive composites. Polymer Composites, 1997. 18(6): p. 711-725. 73. White, S.I., et al., Simulations and electrical conductivity of percolated networks of finite rods with various degrees of axial alignment. Physical Review B, 2009. 79(2): p. 024301. Zhang, S., et al., Vertical CNT -- Ecoflex nanofins for highly linear broad-range-detection wearable strain sensors. Journal of Materials Chemistry C, 2018. 6(19): p. 5132-5139. 74.
1710.00068	3	1710	2017-12-16T19:02:11	Heptagonal Photonic Crystal Fiber for Dispersion Compensation with a Very Low Confinement Loss	[ "physics.app-ph", "physics.optics" ]	This paper presents a photonic crystal fiber (PCF) with heptagonal core and heptagonal cladding for dispersion compensation. Different optical properties of the suggested PCF are explored using the finite element method (FEM). The proposed dispersion compensating PCF (DC-PCF) exhibits a very large negative chromatic dispersion of $-940 ps/(nm-km)$ at $1550 nm$ wavelength. The overall dispersion of the DC-PCF is $-420.1$ to $-1160 ps/(nm-km)$ in the wavelength range of $1390$ to $1700 nm$ ($310 nm$ band). The relative dispersion slope is $0.0036 nm^{-1}$ which is a perfect match with the standard single mode fibers. Moreover, it exhibits a very low confinement loss of about $10^{-5} dB/km$ and low nonlinearity of $45 W^{-1}km^{-1}$ at $1550 nm$ wavelength. Since the suggested DC-PCF has very high negative dispersion and low nonlinearity, it can be a potential candidate for broadband dispersion compensation in fiber-optic communication.	physics.app-ph	physics	> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1 Heptagonal Photonic Crystal Fiber for Dispersion Compensation with a Very Low Confinement Loss Md Borhan Mia, Mohammad Faisal, Syeda Iffat Naz, Kanan Roy Chowdhury, and Animesh Bala Ani  Abstract- This paper presents a photonic crystal fiber (PCF) with heptagonal core and heptagonal cladding for dispersion compensation. Different optical properties of the suggested PCF are explored using the finite element method. The proposed dispersion compensating PCF (DC-PCF) exhibits a very large negative chromatic dispersion of −940 ps/(nm-km) at 1550 nm wavelength. The dispersion variation of the DC-PCF is −420.1 to −1160 ps/(nm-km) in the wavelength range of 1390 to 1700 nm (310 nm band). The relative dispersion slope is 0.0036 nm−1 which is a perfect match with the standard single mode fibers. Moreover, it exhibits a very low confinement loss of about 10−5 dB/km and low nonlinearity of 45 W−1km−1 at 1550 nm wavelength. Since the suggested DC-PCF has very high negative dispersion and low nonlinearity, it could be a potential candidate for broadband dispersion compensation in fiber-optic communication. Index Terms-Confinement loss; Chromatic dispersion; Finite element method; Photonic crystal fiber I. INTRODUCTION W avelength division multiplexed (WDM) and dense wavelength division multiplexed (DWDM) based fiber- optic transmission networks are widely used to transport high bit rate and ultra-high bit rate data [1]. Due to growing demand of various services, e.g., wireless digital cameras, video conferencing, online video streaming, Manuscript received October 9, 2001. (Write the date on which you submitted your paper for review.) Paper titles should be written in uppercase and lowercase letters, not all uppercase. Write "[Invited]" in the title if your paper was invited to appear in a feature or focus issue. Full names of authors are preferred, not required, in the author field. Put a space between authors' initials. F. A. Author is with the National Institute of Standards and Technology, Boulder, Colorado 80305, USA (corresponding author to provide e-mail: author@ boulder.nist.gov). S. B. Author, Jr., was with Rice University, Houston, Texas 77005, USA. He is now with the Department of Physics, Colorado State University, Fort Collins, Colorado 80523, USA. T. C. Author is with the Electrical Engineering Department, University of Colorado, Boulder, Colorado 80309, USA, on leave from the National Research Institute for Metals, Tsukuba, Japan. things (IoT) etc., the data over gaming on demand, IPTV, telemetry, cloud computing, Internet of the telecommunication links are increasing day by day. In WDM systems, the bit rate of 40 Gb/s has been widely used [2]. Moreover, data rate per channel of 100 Gb/s [3] and 400 Gb/s [4] have been realized in order to transport massive data over WDM and DWDM systems for long-haul transmission. The optical fiber has mainly two linear impairments, namely, attenuation and dispersion which require to be mitigated. Between the two, the chromatic dispersion imposes a considerable limitation as it broadens the optical pulses propagating through the fiber. With the increase of demand for capacity, the bit rate increases and the pulse width decreases which consequently further enhances the pulse broadening and drastically reduces the performance of the systems. is indispensable in fiber-optic communication. is more detrimental and compensation it Therefore, dispersion Standard single mode fibers (SMFs) are mostly deployed in fiber-optic transmission links which have a typical dispersion of 15 to 20 ps/(nm-km) at 1550 nm wavelength. This will cause a huge accumulated dispersion along the transmission line. To compensate this dispersion of the SMFs, different strategies have been adopted such as conventional dispersion compensating fibers (DCFs) [5], electronic dispersion compensation [6], Bragg grating fibers [7] and optical phase conjugation [8] etc. Among these techniques, DCFs are widely used due to their negative dispersion, low loss and low nonlinearity [9]. Furthermore, compensation is done in optical domain with ease of coupling with SMF, if we use DCFs. Dispersion compensating PCF (DC-PCF) offers additional benefit by tailoring the optical properties like dispersion, loss and birefringence etc. We use DC-PCF to achieve desirable negative dispersion with suitable slope, low or high birefringence, and in losses. Therefore, DC-PCFs are preferred over DCFs. further reduction the In recent years, photonic crystal fibers (PCFs) or micro-structure optical fibers (MOFs) consisting of air hole channel running down their length have earned a lot of interest due to their sui-generis properties and potential applications. Their optical properties are easily engineered since they have various tuneable design parameters such as pitch, diameter, shape and number of air holes and air-hole rings around the core as well as the cladding region [5]– [10]. These flexibilities of the PCFs make it suitable and potential to supersede SMFs and DCFs. There are some > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2 other attractive properties that can be tailored from the PCFs such as improved birefringence, nonlinearity, large effective area, splice loss, bending loss etc. Physical shapes of the PCFs are quite feasible today because of the invention of fabrication techniques like stack and draw [11], drilling and extrusion [12], sol-gel casting [13] etc. from 1460 The idea to use PCF in dispersion compensation was first introduced by Birks et al. [14] who proposed a PCF showing chromatic dispersion of −2×10−3 ps/(nm-km) at 1550 nm wavelength. Kim et al. [15] has proposed a PCF displaying dispersion of −156 ± 0.5 ps/(nm-km) over the wavelength range of C+L bands. However, the core and cladding consist of elliptical air holes which are difficult to fabricate, difficult to coupling and dispersion value is very low. Habib et al. [16] proposed a PCF of dispersion −588 ps/(nm-km) at 1550 nm. However, the confinement loss is 10-1 dB/m which is very high. In [17], Haque et al. proposed another design with chromatic dispersion of −790.12 ps/(nm-km) and the confinement loss of 10−4 dB/km at 1550 nm wavelength. However, dispersion coefficient varies from −248 to −1069 ps/(nm-km) over E to L bands. In another design, Habib et al. [18] proposed an MOF for which dispersion coefficient is −130 to −360 ps/(nm-km) in the wavelength range of 1400 nm to 1500 nm (100 nm band). Moreover, the background material used here is not silica but organic materials like propanol, butanol and ethanol. In [19], Hasan et al. proposed a PCF of dispersion −555.93 ps/(nm-km) at the wavelength of 1550 nm. Besides, dispersion co-efficient varies from −388.72 to −723.1 ps/(nm- km) ranging to 1625 nm wavelength. Additionally, confinement loss is also high, of the order of 10−2 dB/km at 1550 nm wavelength. In [20], a structure using 5 rings of air holes is proposed whose dispersion varies from −386.57 to −971.44 ps/(nm-km) over the wavelength ranging from 1400 nm to 1610 nm. The chromatic dispersion at 1550 nm is −790.12 ps/(nm-km). However, the design is much complex because of air-holes with different diameters which are angularly rotated and different pitches. All these make the design a fabrication challenge. In [21], the obtained average dispersion is −138 ps/(nm-km) from 980 nm to 1580 nm, however, again the design is complex since the core region consists of pentagonal shaped air holes. Besides, the confinement loss is also high. In [22], Habib et al. suggested a single-mode fiber of dispersion −712 ps/(nm-km). Additionally, dispersion varies from −300 to −1000 ps/(nm-km) in the wavelength range of 1340 nm to 1640 nm. Monfared et al. [23] investigated a PCF and obtained dispersion of −2108 ps/(nm-km) at 1550 nm wavelength. However, necessary information regarding relative dispersion slope (RDS) is missing. Furthermore, core is elliptical shape doped with GeO2 which makes it costlier and suffered fabrication difficulty. In [24], the chromatic dispersion at 1550 nm wavelength is −672 ps/(nm-km). However, there are two elliptical air holes in the core and the confinement loss is very high. Additionally, dispersion varies from −470 to −850 ps/(nm-km) over S to L bands. The PCF in [25] exhibits chromatic dispersion of −1054.4 ps/(nm-km) at 1550 nm wavelength and RDS value match with SMFs. Notwithstanding, the structure is hybrid which increases the complexity and there is no information of confinement loss. Besides, the dispersion varies from −270 to −1100 ps/(nm-km) in the wavelength range of 1340 to 1640 nm (300 nm bands). The large negative dispersion is shown in [26], [27], however, information regarding either RDS or confinement loss is missing and the core is doped in [26] and dispersion is fluctuating in [27]. literatures is of −420.1 to −1160 ps/(nm-km) In this paper, we propose a PCF with heptagonal core and cladding for dispersion compensation. The proposed DC- PCF shows a negative dispersion of −940 ps/(nm-km) at the wavelength of 1550 nm which is higher than that of recently reported [16-20]. The dispersion variation in the wavelength range of 1390 to 1700 nm (310 nm bands). The RDS value of the proposed DC-PCF is 0.0036 nm−1, which is perfectly matched with the SMFs. Besides, the confinement loss is very low in the order of 10−5 dB/km. The proposed DC-PCF operates on single mode since Veff parameter is less than π from wavelength of 1390 nm to 1700 nm (310 nm bands). II. GEOMETRY OF THE DESIGN The cross sectional view of the proposed DC-PCF is demonstrated in Fig. 1. Design is simple since the core and cladding consist of circular air holes in a shape of heptagon. For the background material silica is used, which is industrially available. The proposed design consists of total nine air hole rings divided into two sectors, i.e. inner core region and outer cladding. Inner and outer cladding both have the heptagonal shape. The core region is comprised of three air-hole rings with the pitch value of Ʌ1 = 0.781 μm and the diameter of each hole is d1 = 0.631 μm for the optimum case. It is evident that change of diameter of air hole or pitch near the core region strongly affects the dispersion property of the fiber. We check the effect of both in core as well as cladding for our work. The angle between two adjacent holes in core is 51.43°. For the outer cladding, there are 6 air-hole rings with the pitch value of Ʌ2 = 0.867 μm and the hole diameter of d2 = 0.80 μm for the optimum results. As seen from the Fig. 1, the core has identical air- hole with diameter of d1 and the cladding with diameter of d2. The air filling fraction in core and the cladding are chosen to be d1/Ʌ1=0.81 and d2/Ʌ2=0.92, respectively to achieve optimum result. III. MODEL AND SIMULATION METHOD To investigate the optical properties of the designed fiber, finite element method is used. COMSOL MULTIPHYSICS software, version 5.0 is used as the design simulator. A circular perfectly matched layer (PML) is positioned outside the outermost ring to model the leakage and it produces no reflection. (FEM) Maxwell's equations are solved in layers comprising a finite number of homogeneous isotropic regions ended by PML. By solving the eigenvalue problem of the Maxwell's curl equation, the effective refractive index can be obtained. Maxwell's equation [28] is expressed as (1) 01220EskEs> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3 where E is the electric field vector, k0 = is the free space wave number, η is the refractive index, [s] is the matrix of the PML circular layers and [s]−1 is the inverse matrix. Once the modal refractive index ηeff is obtained, other parameters like chromatic dispersion D(λ), nonlinear co- efficient γ, confinement loss Lc, effective area Aeff, birefringence B, and V parameter etc. can be obtained from their respective equation [29]–[32]. (2) (a) (3) (4) (5) where, and are the real and the imaginary part of the effective refractive indices, respectively; and are the refractive indices of the x and y polarization mode, respectively, E is the electric field vector, k0 is the free space wave number , λ is the operating wavelength and c is the velocity of the light in vacuum. The material dispersion is considered by applying Sellmeier formula in simulation The change of dispersion with a small change in wavelength is known as dispersion slope and it is necessary to calculate the slope mismatch with other fibers. Dispersion slope is calculated by the following equation [33]. (6) The ratio of dispersion slope to dispersion is defined as relative dispersion slope (RDS). The RDS value of the proposed fiber is calculated by using the following equation [34]. (7) where, SSMF(λ) and DSMF(λ) are the dispersion slope and dispersion for the SMF, respectively. Similarly, SDC-PCF(λ) and DDC-PCF(λ) are for the dispersion compensating fibers. Once we obtain the RDS of the DC-PCFs close to that of the (b) (c) Fig. 1. Transverse cross section of the suggested DC-PCF with the optimum parameters d1, d2, Ʌ 1 and Ʌ 2. (a) Full cross sectional view of DC-PCF, Cross sectional view of (b) magnified core (c) Cladding A. Figures and Tables 22ReλλeffdDcdyeffxeffB224effEdxdyAEdxdy602010Imln10ceffLkeffReeffImxeffyeffλλλdDSd(λ)(λ)(λ)(λ)SMFDCPCFSMFDCPCFSSRDSDD> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4 standard single mode fibers (SMFs), our target will be accomplished for the broadband dispersion compensation in fiber-optic communication. The mode property of the proposed DC-PCF is scrutinized carefully. There is a cut off frequency for each mode below which the mode cannot propagate. The normalized frequency or V-parameter of the optical fiber is a very important quantity which is used to determine the propagating modes. If V-parameter of the fiber is less than 𝜋, fiber will allow only one mode and effectively acts as a single-mode fiber. The V-parameter, Veff is expressed by the following equation [35]. (8) where, ηeff and ηFSM are the refractive indices of the fundamental mode and fundamental space filling mode, respectively. IV. RESULTS AND DISCUSSION Fundamental mode field distribution and the effective refractive index of the proposed DC-PCF are presented in Fig. 2 (a) and (b), respectively. Light is well confined in the core region since higher refractive index is achieved in the core region than the cladding which results in high negative value of dispersion. It is seen form the Fig. 2 (b) that the effective refractive index is 1.28 at 1550 nm wavelength. With the increase of wavelength, refractive index decreases indicating that power is well confined in core region at lower wavelength. (a) (a) (b) Fig. 2. With the parameter d1 = 0.631 µm, Ʌ1 = 0.781 µm, d2 = 0.80 µm and Ʌ2 = 0.867 µm for the proposed DC-PCF, (a) Poynting vector profile and (b) Refractive index vs. wavelength (b) Fig. 3. Optimum parameter d1 = 0.631 µm, Ʌ1 = 0.781 µm, d2 = 0.80 µm and Ʌ2 = 0.867 (a) Chromatic dispersion (b) dispersion slope as a function of wavelength. Firstly, we have changed core diameter d1 keeping other parameters constant, i.e., Ʌ 1 = 0.781 µm, Ʌ 2 = 0.867 µm and d2 = 0.80 µm. Taking the value of d1 as 0.621, 0.631, 0.641 µm, the corresponding chromatic dispersions are found −890, −940, −1000 ps/(nm-km), respectively. Calculated Fig. 1. Sample graph with blue (dotted), green (solid) b 222πλeffeffFSMV> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5 RDS values are 0.00416, 0.0036 and 0.0049 nm−1 at d1 = 0.621, 0.631 and 0.641 µm, respectively at 1550 nm wavelength. The effect of variation of diameter d1 on chromatic dispersion is depicted in Fig. 4. Moreover, the confinement loss of the proposed PCF to the corresponding diameters, d1 = 0.621, 0.631, 0.641 µm are 4.64×10−6, 1×10−5 and 1.46×10−5 dB/km, respectively and it is seen in Fig. 5. pitch, Ʌ1 is demonstrated in the Fig. 7. It is seen from the Fig. 7 the confinement loss is 6.14×10−6, 1×10−5 and 1.28×10−5 dB/km for the pitch of Ʌ1 = 0.779, 0.781 and 0.782 µm, respectively. Fig. 4. Chromatic dispersion of the signified DC-PCF as function of wavelength with the parameter d1 taken as 0.621 µm, 0.631 µm and 0.641 µm keeping d2, Ʌ2 and Ʌ1 unaltered. Fig. 6. Chromatic dispersion of the denoted DC-PCF as function of wavelength with the parameter Ʌ1 taken as 0.779 µm, 0.781 µm and 0.782 µm keeping d2, Ʌ2 and d1 unchanged. Fig. 5. Confinement loss vs. wavelength with parameter d1 taken as 0.621 µm, 0.631 µm and 0.641 µm keeping d2, Ʌ2 and Ʌ1 unchanged. Later on, we keep d1 = 0.631 µm, d2 = 0.80 µm and Ʌ2 = 0.867 µm unchanged and vary the value of Ʌ1 in core. The chromatic dispersion of the proposed PCF with the changing values of Ʌ1 is shown in the Fig. 6. Ʌ1 is chosen as 0.779, 0.781 and 0.782 µm while the calculated dispersion is −840, −940 and −1010 ps/(nm-km), respectively. The calculated RDS values are 0.0040, 0.0036, 0.0043 nm−1, respectively at 1550 nm wavelength. The confinement loss for varying core Fig. 7. Confinement Loss of the proposed DC-PCF as function of wavelength with the parameter Ʌ1 taken as 0.779 µm, 0.781 µm and 0.782 µm keeping d2, Ʌ2 and d1 untouched. The above discussion reveals that whether d1 value is increased or decreased, either way, RDS value moves away from 0.0036nm−1 though the chromatic dispersion increases with the increment of d1 at the wavelength of 1550 nm. On the other hand, the increment or decrement of Λ1 has the same effect as d1. So the optimum core diameter and pitch are chosen to be d1 = 0.631 µm and Ʌ1 = 0.781 µm, respectively. The confinement loss from the Fig. 5 reveals that with the increment of d1 loss will increase since optical field moves towards the cladding region for large d1. From > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6 the Fig. 7, the confinement loss increases for the large pitch Ʌ1, due to mode field travels to cladding region. 5.50×10−6, 1×10−5 and 1.50×10−5 dB/km, respectively at 1550 nm wavelength. The confinement loss is decreased below the value of d2/Ʌ2 = 0.92 and increased above the below of d2/Λ2 = 0.92. This is because the optical field is more confine when the ratio of d2/Λ2 = 0.92. Therefore, the optimum parameters of the proposed fiber are taken to be d2/Ʌ2 = 0.92, d1 = 0.631 µm and Ʌ1 = 0.781 µm. Fig. 8. The proposed DC-PCF's chromatic dispersion as a function of wavelength changing d2/Ʌ 2 from 0.91 µm to 0.93 µm keeping other parameters unaltered. Fig. 10. Nonlinear co-efficient and effective area as a function of wavelength for the optimum parameter of the Proposed DC-PCF. The effective area and the nonlinearity of the proposed fiber are depicted in Fig. 10 for the optimum values of parameters d1, d2, Ʌ1 and Ʌ2. It is clear from the Fig. 10 that at 1550 nm wavelength the effective area is 2.5 µm2 and the nonlinear co-efficient is 45 W−1km−1 which is low. Therefore, the proposed fiber won't be much affected by the non- linearity. Fig. 9. Confinement loss vs. wavelength for the changing d2/Ʌ2 from 0.91 µm to 0.93 µm keeping other parameters unaltered. To explore the effect of the outer cladding, we have changed the d2/Ʌ2 ratio keeping d1 = 0.631 µm and Ʌ1 = 0.781 µm unaltered. The value of d2/Ʌ2 is taken as 0.91, 0.92 and 0.93 µm and the corresponding dispersion curves are shown in the Fig. 8. Since the parameter d2/Ʌ2 is changing from 0.91 µm to 0.93 µm, the chromatic dispersion has increased. The chromatic dispersions are −900, −940 and −980 for d2/Λ2 = 0.91, 0.92 and 0.93, respectively at communication band. However, RDS values at 1550 nm are obtained as 0.004, 0.0036, and 0.0048 nm−1, respectively. The confinement loss is demonstrated in the Fig. 9 and found Fig. 11. RDS of the proposed DC-PCF vs. wavelength for the optimum parameters d1, Ʌ1, d2 and Ʌ2. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7 The RDS of the proposed DC-PCF is shown in the Fig. 11 and it is realized the RDS value of recommended DC-PCF is close to that of SMFs; nominally it is 0.0036 nm−1 at 1550 nm wavelength. Since RDS is used to judge the dispersion compensation (DC), the proposed fiber is a perfect candidate for broadband dispersion compensation. Fig. 12. V-parameter of the proposed DC-PCF as a function of wavelength for the inner and outer cladding with the optimum geometric entities. The DC-PCF can be operated as a SMF if V-parameter, Veff ≤ π. If Veff is greater than the π, higher order modes get associated with it. The V-parameter of the proposed DC- PCF is presented in Fig. 12. From the Fig. 12, the obtained value for Veff is less than π from 1390 nm to 1760 nm wavelength (310 nm bands). So, it is ensured that the proposed DC-PCF will operate as single-mode over the O+E+S+C+L bands. Fabrication viability is the major issue in realization of PCFs. Heterogeneous air hole diameters at the cladding region already has been fabricated using the stack-and- draw technique [36]. Bise et al. [13] used sol-gel technology to fabricate microstructure optical fibers with missing air holes. Even PCFs with high air filling fraction have been reported fabrication viability [37], [38]. All these methods could be used to fabricate our proposed DC-PCF, however, stack-and-draw technique will be more convenient. Finally in Table 1, we compare our work with some recently published works for compensating dispersion. It is seen from the table, our proposed DC-PCF has offered much better results with simpler design. TABLE I COMPARISON OF PROPOSED DC-PCF WITH RECENT ARTICLES D at Lc at Dispersion variation 1550 nm (ps/(nm- 1550 nm (dB/km) Ref (ps/(nm-km)) km)) 16 −588 103 −400 to −725 (165 nm bands) 17 −790.12 10−4 −248.65 to −1069 (270 nm bands) 19 −555.93 × −388.72 to−723.1(165 20 −790.12 10−2 −386.57 to −971.44 nm bands) −712 × −200 to −1000 (300 nm (210 nm bands) −672 −940 10−1 10−5 bands) −470 to −850 (165 nm band) −420.1 to −1160 (310 nm band) 21 22 Prop osed DC- PCF V. CONCLUSION Heptagonal lattice structure DC-PCF is presented in this paper which exhibits a very high chromatic dispersion of −940 ps/(nm-km) at 1550 nm wavelength. The proposed DC- PCF shows dispersion variation from −420.1 to −1160 ps/(nm-km) covering the wavelength from 1440 nm to 1700 nm. To our best knowledge, this is highest result compared to recently published articles [16-17], [19-20], [22], [24]. The RDS of the suggested fiber is perfectly matched with standard SMF. The proposed fiber demonstrates a very low confinement loss of 10−5 dB/km at the wavelength of 1550 nm. In addition, the proposed DC-PCF operates on single mode over a wide telecom band. So, our suggested DC-PCF can be a suitable choice for broadband dispersion compensation. ACKNOWLEDGMENT This article is based on our own research work and we have not received any fund from any organization. REFERENCES M. L. Loeb and G. R. Stilwell, "High-speed data transmission on an optical fiber using a byte-wide WDM system," J. Light. Technol., vol. 6, no. 8, pp. 1306–1311, 1988. K. Fukuchi et al., "10.92-Tb/s (273 x 40-Gb/s) triple- band/ultra-dense WDM optical-repeatered transmission experiment," in Optical Fiber Communication Conference, 2001, p. PD24. L. W. D. M. Transmission et al., "No-Guard-Interval Coherent Optical OFDM for," J. Light. Technol. Vol. 27, Issue 16, pp. 3705-3713, vol. 27, no. 16, pp. 3705–3713, Aug. 2009. X. Zhou et al., "High Spectral Efficiency 400 Gb/s Transmission and Training-Assisted Carrier Recovery," J. Light. Technol. Vol. 31, Issue 7, pp. 999-1005, vol. 31, no. 7, pp. 999–1005, Apr. 2013. B. Zsigri, J. Laegsgaard, and A. Bjarklev, "A novel photonic [1] [2] [3] [4] [5] > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8 [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] highly negative birefringent, crystal fibre design for dispersion compensation," J. Opt. A Pure Appl. Opt., vol. 6, no. 7, pp. 717–720, Jul. 2004. N. M. Litchinitser, B. J. Eggleton, and D. B. Patterson, "Fiber Bragg gratings for dispersion compensation in transmission: Theoretical model and design criteria for nearly ideal pulse recompression," J. Light. Technol., vol. 15, no. 8, pp. 1303–1313, 1997. H. Bülow, F. Buchali, and A. Klekamp, "Electronic dispersion compensation," J. Light. Technol., vol. 26, no. 1, pp. 158–167, 2008. S. Watanabe, T. Naito, and T. Chikama, "Compensation of chromatic dispersion in a single-mode fiber by\noptical phase conjugation," IEEE Photonics Technol. Lett., vol. 5, no. 1, pp. 92–95, Jan. 1993. A. Bala, K. R. Chowdhury, M. B. Mia, and M. Faisal, "Highly dispersion compensating photonic crystal fiber," Appl. Opt., vol. 56, no. 25, p. 7256, Sep. 2017. J. Broeng, D. Mogilevstev, S. E. Barkou, and A. Bjarklev, "Photonic Crystal Fibers: A New Class of Optical Waveguides," Opt. Fiber Technol., vol. 5, no. 3, pp. 305– 330, Jul. 1999. P. S. J. Russell, "Photonic-Crystal Fibers," J. Light. Technol. Vol. 24, Issue 12, pp. 4729-4749, vol. 24, no. 12, pp. 4729–4749, Dec. 2006. M. C. J. Large et al., "Microstructured polymer optical fibres: New opportunities and challenges," Proc. Mol. Cryst. Liq. Cryst., vol. 446, no. 1, pp. 219–31, Apr. 2006. R. T. Bise and D. J. Trevor, "Sol-gel derived microstructured fiber: fabrication and characterization," OFC/NFOEC Tech. Dig. Opt. Fiber Commun. Conf. 2005., vol. 3, pp. 11–13, 2005. T. A. Birks, D. Mogilevtsev, J. C. Knight, and P. S. J. Russell, "Dispersion compensation using single-material fibers," IEEE Photonics Technol. Lett., vol. 11, no. 6, pp. 674–676, Jun. 1999. S. E. Kim, B. H. Kim, C. G. Lee, S. Lee, K. Oh, and C.-S. Kee, "Elliptical defected core photonic crystal fiber with high birefringence and negative flattened dispersion," Opt. Express, vol. 20, no. 2, p. 1385, Jan. 2012. M. Selim Habib, M. Samiul Habib, S. M. Abdur Razzak, and M. Anwar Hossain, "Proposal for highly bi-refringent broadband dispersion compensating octagonal photonic crystal fiber," Opt. Fiber Technol., vol. 19, no. 5, pp. 461– 467, Oct. 2013. M. M. Haque, M. S. Rahman, M. Samiul Habib, M. Selim Habib, and S. M. A. Razzak, "A new circular photonic crystal fiber for effective dispersion compensation over e to L wavelength bands," J. Microwaves, Optoelectron. Electromagn. Appl., vol. 12, no. 2, pp. 281–291, Dec. 2013. M. Selim Habib, M. Samiul Habib, S. M. Abdur Razzak, Y. Namihira, M. A. Hossain, and M. A. Goffar Khan, "Broadband dispersion compensation of conventional single mode fibers using microstructure optical fibers," Optik (Stuttg)., vol. 124, no. 19, pp. 3851–3855, Oct. 2013. M. I. Hasan, M. Selim Habib, M. Samiul Habib, and S. M. Abdur Razzak, "Highly nonlinear and highly birefringent dispersion compensating photonic crystal fiber," Opt. Fiber Technol., vol. 20, no. 1, pp. 32–38, Jan. 2014. M. M. Haque, M. S. Rahman, M. S. Habib, and S. M. A. Razzak, "Design and characterization of single mode circular photonic crystal fiber for broadband dispersion compensation," Opt. - Int. J. Light Electron Opt., vol. 125, no. 11, pp. 2608–2611, Jun. 2014. M. Samiul Habib, M. Selim Habib, M. I. Hasan, and S. M. A. Razzak, "A single mode ultra flat high negative residual dispersion compensating photonic crystal fiber," Opt. Fiber Technol., vol. 20, no. 4, pp. 328–332, Aug. 2014. M. Samiul Habib, R. Ahmad, M. Selim Habib, and S. M. A. Razzak, "Design of single polarization single mode dispersion compensating photonic crystal fiber," Opt. - Int. J. Light Electron Opt., vol. 125, no. 16, pp. 4313–4318, [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] fibers low extremely confinement "Design of hybrid photonic crystal Aug. 2014. Y. E. Monfared and S. A. Mojtahedinia, "Highly birefrigent photonic crystal fiber with high negative dispersion for broadband dispersion compensation," Optik (Stuttg)., vol. 125, no. 20, pp. 5969–5972, Oct. 2014. M. Selim Habib, M. Mejbaul Haque, M. Samiul Habib, M. I. Hasan, M. Shaifur Rahman, and S. M. A. Razzak, "Polarization maintaining holey for residual dispersion compensation over S + C + L wavelength bands," Optik (Stuttg)., vol. 125, no. 3, pp. 911–915, Feb. 2014. M. I. Hasan, M. S. Habib, M. S. Habib, and S. M. A. Razzak, fiber: Polarization and dispersion properties," Photonics Nanostructures - Fundam. Appl., vol. 12, no. 2, pp. 205– 211, Apr. 2014. J. M. Hsu, W. H. Zheng, C. L. Lee, and J. S. Horng, "Theoretical investigation of a dispersion compensating photonic crystal fiber with ultra-high dispersion coefficient and loss," Photonics Nanostructures - Fundam. Appl., vol. 16, pp. 1–8, Aug. 2015. R. Hao and G. Sun, "Design of photonic crystal fiber with large negative dispersion andhigh nonlinearity," Optik (Stuttg)., vol. 126, no. 22, pp. 3353–3356, Nov. 2015. K. Saitoh and M. Koshiba, "Full-vectorial imaginary- distance beam propagation method based on a finite element scheme: application to photonic crystal fibers," IEEE J. Quantum Electron., vol. 38, no. 7, pp. 927–933, Jul. 2002. A. Agrawal, N. Kejalakshmy, J. Chen, B. M. Rahman, and K. T. Grattan, "Golden spiral photonic crystal fiber: polarization and dispersion properties," Opt. Lett., vol. 33, no. 22, p. 2716, Nov. 2008. K. Saitoh and M. Koshiba, "Leakage loss and group velocity dispersion in air-core photonic bandgap fibers," Opt. Express, vol. 11, no. 23, p. 3100, Nov. 2003. K. Saitoh, M. Koshiba, T. Hasegawa, and E. Sasaoka, "Chromatic dispersion control in photonic crystal fibers: application to ultra-flattened dispersion.," Opt. Express, vol. 11, no. 8, pp. 843–852, Apr. 2003. K. Suzuki, H. Kubota, S. Kawanishi, M. Tanaka, and M. Fujita, "Optical properties of a low-loss polarization- maintaining photonic crystal fiber.," Opt. Express, vol. 9, no. 13, pp. 676–680, Dec. 2001. M. Samiul Habib, R. Ahmad, M. Selim Habib, and M. Imran Hasan, "Residual dispersion compensation over the S + C + L + U wavelength bands using highly birefringent octagonal photonic crystal fiber," Appl. Opt., vol. 53, no. 14, p. 3057, May 2014. L. Grüner-Nielsen et al., "Dispersion Compensating Fibers," Opt. Fiber Technol., vol. 6, no. 2, pp. 164–180, Apr. 2000. N. A. Mortensen, J. R. Folkenberg, M. D. Nielsen, and K. P. Hansen, "Modal cutoff and the V parameter in photonic crystal fibers," Opt. Lett., vol. 28, no. 20, p. 1879, Oct. 2003. Z. Liu, C. Wu, M.-L. Vincent Tse, C. Lu, and H.-Y. Tam, "Ultrahigh birefringence index-guiding photonic crystal fiber and its application for pressure and temperature discrimination," Opt. Lett., vol. 38, no. 9, pp. 1385–1387, May 2013. D. Ghosh, S. Bose, S. Roy, and S. K. Bhadra, "Design and Fabrication of Microstructured Optical Fibers with Optimized Core Suspension for Enhanced Supercontinuum Generation," J. Light. Technol., vol. 33, no. 19, pp. 4156– 4162, Oct. 2015. S. O. Leonov, V. A. Lazarev, M. K. Tarabrin, D. A. Dvoretskiy, A. S. Pasishnik, and A. D. Pryamikov, "Visible supercontinuum generation in large-core photonic crystal fiber with high air-filling fraction," J. Phys. Conf. Ser., vol. 584, no. 1, p. 12015, Jan. 2015. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9 Md Borhan Mia was with the Electrical and Electronic Engineering Department, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh. Dr. Mohammad Faisal is with the Electrical and Electronic Engineering Department, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh. Syeda Iffat Naz was with the Electrical and Electronic Engineering Department, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh. Kanan Roy Chowdhury was with the Electrical and Electronic Engineering Department, Chittagong University of Engineering and Technology, Chittagong, Bangladesh. Animesh Bala was with the Electrical and Electronic Engineering Department, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh.
1810.08801	1	1810	2018-10-20T13:23:32	Performance analysis of a direct-absorption parabolic trough solar collector	[ "physics.app-ph" ]	A parabolic trough solar collector is a dominant technology for high-temperature industrial applications, but efficient use of a conventional surface-based parabolic trough solar collector (SBPTSC) is limited by its high radiation loss due to the high surface temperature. Recently, direct-absorption parabolic trough solar collector (DAPTSC) using nanofluids has been proposed, and its thermal efficiency has been reported to be 5-10$\%$ higher than the conventional SBPTSC for inlet temperature up to 250$^\circ$C. However, the inner tubes of the receivers of the existing DAPTSCs are all transparent, so the sun rays entering the inner tube can only travel once through the nanofluids. As a result, the optical path length for the sun rays is limited by the inner tube size, which in turn requires high value of the absorption coefficient of nanofluids. Due to the approximately linear relation between the absorption coefficient and the particle concentration, higher absorption coefficient is likely to cause particle agglomeration, leading to detrimental effects on maintaining stable collector performance. In the current study, the transparent DAPTSC is improved by applying a reflective coating on the upper half of the inner tube outer surface, such that the optical path length is doubled compared to the transparent DAPTSC; thus, the absorption coefficient of the nanofluids can be reduced accordingly. The coated DAPTSC is found to have obvious advantage compared to the transparent DAPTSC at absorption coefficient below 0.5 cm$^{-1}$ for a receiver with inner tube diameter of 7 cm. In addition, performance of the transparent DAPTSC, the coated DAPTSC and the SBPTSC with black chrome coating have been compared to explore their advantageous operation conditions, such as inner tube diameter, flow rate, and inlet temperature, with or without a glass envelope for vacuum evacuation.	physics.app-ph	physics	Performance analysis of a direct-absorption parabolic trough solar collector Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, South Korea Caiyan Qin, Joong Bae Kim, Bong Jae Lee∗ 8 1 0 2 t c O 0 2 ] h p - p p a . s c i s y h p [ 1 v 1 0 8 8 0 . 0 1 8 1 : v i X r a Abstract A parabolic trough solar collector is a dominant technology for high-temperature industrial applications, but efficient use of a conventional surface-based parabolic trough solar collector (SBPTSC) is limited by its high radiation loss due to the high surface temperature. Recently, direct-absorption parabolic trough solar collector (DAPTSC) using nanofluids has been proposed, and its thermal efficiency has been reported to be 5 -- 10% higher than the conventional SBPTSC for inlet temperature up to 250◦C. However, the inner tubes of the receivers of the existing DAPTSCs are all transparent, so the sun rays entering the inner tube can only travel once through the nanofluids. As a result, the optical path length for the sun rays is limited by the inner tube size, which in turn requires high value of the absorption coefficient of nanofluids. Due to the approximately linear relation between the absorption coefficient and the particle concentration, higher absorption coefficient is likely to cause particle agglomeration, leading to detrimental effects on maintaining stable collector performance. In the current study, the transparent DAPTSC is improved by applying a reflective coating on the upper half of the inner tube outer surface, such that the optical path length is doubled compared to the transparent DAPTSC; thus, the absorption coefficient of the nanofluids can be reduced accordingly. The coated DAPTSC is found to have obvious advantage compared to the transparent DAPTSC at absorption coefficient below 0.5 cm−1 for a receiver with inner tube diameter of 7 cm. In addition, performance of the transparent DAPTSC, the coated DAPTSC and the SBPTSC with black chrome coating have been compared to explore their advantageous operation conditions, such as inner tube diameter, flow rate, and inlet temperature, with or without a glass envelope for vacuum evacuation. The findings in this study will facilitate the use of nanofluids in a parabolic trough solar collector with more stability and provide a guide for choosing suitable type of parabolic trough solar collectors for specific working conditions. Keywords: Direct-absorption solar collector, Parabolic trough solar collector, Blended plasmonic nanofluids ∗Corresponding author Email address: [email protected] (Bong Jae Lee) Preprint submitted to Elsevier October 23, 2018 1. Introduction Solar energy is a promising renewable energy in both aspects of sustaining the economic growth and reducing pollutants caused by energy consumption. Among the several available technologies for solar energy harvesting, the concentrated solar power (CSP) is an emerging and significant technology with the advantages, such as built-in storage capability and low green house gas emissions [1]. The parabolic trough solar collector is a dominant CSP technology among the medium-high-temperature solar collectors working for the operating temperature up to 800 K [2]. In the parabolic trough solar collector, a mirror or a reflector with parabola shape concentrates the solar radiation to its focal line where a tubular receiver containing working fluid is mounted to generate high-temperature heat. So far, the receiver is usually composed of a coated metal absorber (e.g., black chrome coating) and a glass envelope, and the annulus between them is evacuated to suppress the convection heat loss. The desirable absorber surface should have high absorptivity in the solar spectrum but low emissivity in the mid-infrared spectrum to minimize the radiation loss. The major disadvantages of the surface-based parabolic trough solar collector (SBPTSC) are the low thermal efficiency at high temperature [3] and the instability of selective coating at high temperatures [4]. To remedy those issues associated with the SBPTSC, a direct-absorption parabolic trough solar collector (DAPTSC) using nanofluids has been proposed recently, and its thermal efficiency has been found 5 -- 10% higher than the SBPTSC [5]. Therefore, the DAPTSC can be an efficient alternative for harvesting solar energy at high temperature. Nanofluids is a dispersion of nanoparticles in a base fluid, such as water or oil [6]. Due to the capability of nanoparticles that can directly interact with the solar radiation, nanofluids can absorb the solar radiation uniformly inside. As a result, the temperature distribution inside the solar collector becomes relatively uniform, leading to less heat loss from the collector surface and improved thermal efficiency [6, 7]. Nanofluids has been studied intensively for direct absorption solar collectors (DASCs) in low-temperature applications [8 -- 18]. In addition, the optical property of various nanofluids have also been studied intensively to improve its photothermal conversion efficiency [19 -- 24]. On the basis of the study on DASCs using nanofluids, a direct-absorption parabolic trough solar collector (DAPTSC) for high temperature applications has been studied recently. Khullar et al. [5] first proposed the idea of utilizing aluminum nanofluids with a volume fraction of 5% for concentrating solar collector, and their DAPTSC was reported to have about 5 -- 10% higher thermal efficiency than the conventional SBPTSC. Later, several numerical works have been performed to investigate the effect of nanofluid concentrations, mass flow rate and inlet temperature for the DAPTSC [4, 25 -- 27]. However, all these studies employed the transparent glass inner tube; thus, the sun rays can only travel once through the nanofluids after entering the inner tube. Consequently, the optical path length for sun rays is limited by the inner diameter of the inner tube of the receiver, which in turn requires high value of the absorption coefficient of nanofluids to gain sufficiently 2 high energy absorption [14]. As low absorption coefficient indicates low particle concentration [28], which is beneficial for avoiding particle agglomeration, a DAPTSC should be designed to have a larger diameter of inner tube to reduce the particle concentration of nanofluids. Besides, all these studies considered nanofluids with a spectrally varying absorption coefficient simply depending on nanoparticle material (i.e., uncontrolled manner). However, Qin et al. [17] recently showed that the blended plasmonic nanofluid with either a uniform absorption coefficient (when the nanoparticle concentration is sufficiently high) or a spectrally varying absorption coefficient following the shape of solar spectrum (when the nanoparticle concentration is below a certain value) is more beneficial for broadband solar energy absorption. In addition, a uniform solar irradiance is assumed in recent works [3, 5, 29], which is not practical in case of parabolic trough solar collectors where the solar radiation is highly dependent on the angular direction of the receiver. Lastly, in Refs. [5, 26], the radiation from the inner tube surface is assumed to be dissipated to the environment directly, but this assumption is not valid because the glass envelope is not transparent to the long-wavelength thermal emission in the mid-infrared spectral region [30]. The present work carries out the performance analysis of a new type of direct-absorption parabolic trough solar collector. Inspired by the flat-plate DASCs where a mirror is employed at the bottom of the channel to enhance the optical path length of sun rays inside the collector [9, 31], we propose to apply a highly reflective metallic coating on the upper half of the inner tube outer surface such that the optical path length for the concentrated rays can be doubled due to the reflection. It is worth mention that the radiation heat loss from the inner tube outer surface can also be reduced greatly due to the low emissivity of a metallic coating compared to that of bare glass. Besides, the blended plasmonic nanofluid with uniform absorption coefficient will be applied to have broadband absorption of solar energy. In addition, the ray tracing method will also be employed in this study to obtain the nonuniform solar irradiance on the inner tube so that the corresponding nonuniform heat generation by the blended plasmonic nanofluid can be obtained. 2. Theoretical model of a direct-absorption parabolic trough solar collector In order to simulate the DAPTSC, two models were developed in COMSOL Multiphysics R(cid:13). First, a two-dimension (2-D) optical model was built to simulate the solar incidence on the collector based on the ray tracing method as well as the corresponding heat source generation based on the Beer-Lambert law. Second, a three-dimensional (3-D) thermal model that couples the turbulent fluid flow with the heat transfer inside the inner tube of the receiver was built. The heat source obtained from the 2-D optical model was incorporated to the 3-D thermal model for calculating the velocity and the temperature distributions inside the inner tube of the receiver. Because there is no difference along the tube length in the volumetric heat generation inside the nanofluid, the 2-D optical model is sufficient for calculating the heat source for the 3-D thermal model, leading to substantially reduced calculation time. 3 tvac = 1.65 cm Vacuum layer Outer glass envelop tenv = 3 mm Inner glass tube ttube = 3 mm Reflective parabola z = 0.1462y2 m 1 7 . 1 = H z Highly reflective coating r Blended nanofluids R e c e i v e r p a r t o W = 5.76 m y Figure 1: Schematic of the coated DAPTSC. The schematic of the 2-D model is shown in Fig. 1 with geometrical sizes annotated. Sizes of the parabola and the receiver are chosen according to the commercialized LS-3 trough and the LS-3 receiver, respectively [32], and its geometrical concentration ratio is W/D ≈ 82.3. In this study, only the normally incident solar radiation is considered for simplicity. The parabolic trough solar collector system consists of a receiver part (two concentric glass tubes, i.e., an inner glass tube and an outer glass envelope) and one aluminum parabola with a high reflectivity of 0.94 [3, 32]. The parabola is used to reflect and concentrate the incident sun rays to the receiver tubes. The glass envelope is transparent in the solar spectrum with the transmittance of τgla = 0.94 [3, 5, 32] but has emissivity of ǫgla = 0.86 in the mid-infrared spectral region [3, 33]. The property of the inner tube is the same with the envelope glass, except that a highly reflective metallic coating is applied on its upper half of the outer surface in order to double the optical path length of the concentrated rays. Notice that among all the rays incident onto the coated DAPTSC, only a small portion (D/W ≈ 1.2%) are blocked by the metallic coating on the outer surface of the inner tube while most other rays are intercepted by the parabola reflector and then concentrated to the receiver. Therefore, compared to the transparent DAPTSC, about 1.2% of the solar radiation is blocked in order to gain the doubled optical path length in the coated DAPTSC. Another advantage of the reflective coating is that due to the metal's much lower infrared emissivity than that of glass [30], it can significantly reduce the radiation loss from the inner tube outer surface. In the evaluation of the extinction of the incident rays, the scattering effect is ignored by assuming the small size of suspended nanoparticles, usually several tens of nanometers, as also done in other studies related to DASC with nanofluids [6, 7, 10, 12, 19, 34]. In the current study, the absorption coefficient α of the blended plasmonic nanofluid is assumed to be constant in the solar spectrum for achieving broadband solar energy 4 absorption. This assumption is valid when nanoparticles with different absorption peaks are well mixed together [14, 15]. Note also that the magnitude of α is approximately proportional to the concentration of the nanoparticles and thus, can be tuned by varying the nanoparticle concentration [9, 17]. Therefore, the spectral uniformity and the magnitude of the absorption coefficient of nanofluid can be individually tuned. The solar irradiance reaching the inner tube surface is numerically calculated from the 2-D optical model using the ray tracing method. The attenuation of the solar radiation in the working fluid is taken into consideration according to the Beer-Lamber law, resulting in the locally varying volumetric heat generation, Qgen(r, φ). The solar weighted absorption coefficient, Am, can be estimated from Am = 1 W GT Z D/2 0 Z 2π 0 Qgen(r, φ)rdrdφ (1) where GT is the solar irradiance assumed as 992 W/m2. In the 3-D thermal model, the fluid flow and heat transfer are coupled in the inner tube region. The thermophysical property of the nanofluids are taken the same as the base fluids, i.e., Therminol VP-1 [35], because the nanoparticle concentrations is far below 1% in volume fraction, as was also done in other related works [4, 5, 27, 36, 37]. The flow rate in this study is around 0.1 kg/s and the corresponding Reynold number is above ReD = 2300; thus, the flow is in the turbulent regime and the standard k − ε model was applied with corresponding parameters Cµ = 0.09, cε1 = 1.44, cε2 = 1.92, σk = 1 and σε = 1.3 [38]. The wall function method was used in the vicinity of the solid wall because the viscosity effects predominate over the the turbulent effects [38]. For thermal analysis, the following energy equation was solved [39]: ∂ ∂xi (ρuiT ) = ∂ ∂xi (cid:20)(k + kt) T xi(cid:21) + Qgen (2) where the subscript 'i' represents the coordinate x, y or z, and ui and xi indicates the velocity and dimension in the corresponding directions respectively. In the above equation, ρ is the fluid density, k is the fluid thermal conductivity, kt is the turbulent thermal conductivity defined as µtCp/P rt with turbulent viscosity µt from the turbulent flow model and turbulent Prandtl number from the Kays-Crawford model [40], and Qgen is transformed to the Cartesian coordinate. The boundary conditions are: T (x = 0) = T0 (3) (4) + hin(cid:2) ¯Tei − T (r = Rto)(cid:3) where T0 is the inlet temperature and L is the length of the receiver tube. And ∂T = 0 ∂x (cid:12)(cid:12)(cid:12)(cid:12)x=L ei − T 4(r = Rto)(cid:3) ǫei (cid:16) Dto Dei(cid:17) + 1−ǫei σ(cid:2) ¯T 4 2 ǫto,u+ǫto,l −(k + kt) = ∂T ∂r (cid:12)(cid:12)(cid:12)(cid:12)r=Rto where the subscripts 'ei' denotes the envelope inner surface, 'to, u' and 'to, l' denotes the upper half and lower half of the inner tube outer surface, respectively. In the above equation, the surface emissivity 5 values are ǫro,u = 0.1, ǫro,l = 0.86 and ǫei = 0.86, ¯Tei denotes the average of the envelope inner surface temperature, and hin is the convective heat transfer coefficient in the evacuated annulus which is assumed to be hin = 0.000174 W/m2-K [41]. The radiation term in Eq. (4) is derived based on the radiation network approach [30] assuming that the whole inner tube surface is at its average temperature. It is also recognized that for using the radiation network approach, the surface is assumed as diffusive and gray. Even though the glass surface is not real diffusive, it is an easy and good approximation by assuming diffusive as also done in other studies [1, 42]. Note that some previous works [5, 26] assumed that the radiation heat loss from the inner tube is dissipated directly to the environment through the glass envelope. Unfortunately, this simple assumption is not valid because that glass envelope is not transparent to the infrared thermal radiation [30]. In the current study, the heat loss from the inner tube is treated to be transferred to the envelope inner surface first and then this heat dissipates to the ambient via heat convection as well as to the space via thermal radiation. In order to obtain ¯Tei, the energy balance equation is applied again to the glass envelope; that is, the heat transfer from the outer surface of the inner tube to the inner surface of the envelope should be the same as the heat transfer from the outer surface of the envelope to the surroundings: σ(cid:2)T 4(r = Rei) − ¯T 4 to(cid:3) ǫei (cid:16) Dto Dei(cid:17) Dto(cid:19) (cid:18) Dei  (cid:18) Deo Dei(cid:19) hout [Ta − T (r = Reo)] + ǫeoσ(cid:2)T 4  + hin(cid:2)T (r = Rei) − ¯Tto)(cid:3)  = sky − T 4(r = Reo)(cid:3) + 1−ǫei 2 ǫto,u+ǫto,l (5) where the subscripts 'eo' denotes the envelope outer surface and ¯Tto is the average of inner tube outer surface temperature, hout is natural heat transfer coefficient at the glass envelope outer surface which is estimated as 10 W/m2-K [43]. Ta is ambient temperature with 293.15 K and Tsky is sky temperature assumed as 278 K [44]. Note that in Eq. (5) the term ( Dei Dto ) are used to guarantee the heat flux balance. Finally, ) and ( Deo Dei the collector thermal efficiency is defined as [45]: η = mcp( ¯Tout − T0) LW GT (6) where m is the flow rate, ¯Tout is the average temperature at the outlet, and cp is the heat capacity at the average temperature, i.e., Tm = (T0 + ¯Tout)/2. In addition, the temperature gain of the working fluid through the collector is estimated from ∆T = ¯Tout − T0. In order to validate the thermal modeling, the calculation results of the SBPTSC are compared to the experimental results from Sandia National Laboratory [46] and the predicted results by Xu et al. [42] for several selected cases, using the same parameters for configuration (LS-2 trough and collector) and the same base fluids (Syltherm 800). Note that in case of the SBPTSC, the inner tube outer surface absorbs the solar radiation directly, and thus, the surface heat source term is obtained from the 2-D optical model and is directly assigned to the 3-D thermal model. The test results are listed in Table 1. It is seen that the current 6 Table 1: Comparison of experimental and numerical data with the current simulation results. The unit for ambient temperature Ta, inlet temperature T0 and temperature gain ∆T are all in K, the unit for solar irradiance GT is W/m2, and that for volumetric flow rate Vin is L/min. The subscripts of 'exp', 'num' and 'cur' indicates the experimental results [46], numerical results [42] and the current simulation results, respectively. GT Ta T0 Vin ∆Texp ∆Tnum ∆Tcur ηexp ηnum ηcur Case 1 933.7 294.35 375.35 47.7 Case 2 968.2 295.55 424.25 47.8 Case 3 982.3 297.45 470.65 49.1 21.8 22.3 22.0 22.40 23.21 22.55 0.7251 0.7424 0.7440 23.01 0.7090 0.7394 0.7279 22.94 22.68 0.7017 0.7310 0.7154 simulation results are generally in between of the experimental results and the numerical simulation results. The reason that the current thermal efficiency is slightly higher than the experimental data is probably because the current model did not account for the heat loss through the brackets and tube ends. Because the relative difference of the current results with the experimental data and other numerical results are within 5%, the current thermal model can be considered to be reliable. In this study, the other four collectors, including the coated DAPTSC without a glass envelope, the transparent DAPTSC, the transparent DAPTSC without a glass envelope, and the conventional SBPTSC with black chrome coating, are investigated and compared with each others (refer to Fig. 2). Because their modeling approach are generally similar with that for the coated DAPTSC, only the difference are briefly described as below. For the transparent DAPTSC, the only difference is that the the term 2 ǫto,u+ǫto,l is replaced by 1 ǫto in Eqs. (4) and (5) as the whole inner tube outer surface has the same emissivity of 0.86. As to the SBPTSC, besides the above mentioned difference, the inner tube surface absorbs the solar energy directly (with an absorptance of 0.94) to obtain the boundary heat source term for the 3-D model. As for the coated DAPTSC as well as the transparent DAPTSC without a glass envelope, the modeling approach is similar as their counterparts with a glass envelope, except that the calculation for the 3-D model is much simpler since the inner tube surface is directly facing the environment. Note also that for SBPTSC, its inner tube (or absorber) surface has a temperature-dependent thermal emissivity ǫm in the range of 0.2 ∼ 0.4 assuming the surface as black chrome coated copper [27, 33]. (a) (b) (c) (d) (e) Figure 2: Schematics of the receiver part of the five solar collectors: (a) coated DAPTSC with a glass envelope; (b) coated DAPTSC without a glass envelope; (c) transparent DAPTSC with a glass envelope; (d) transparent DAPTSC without a glass envelope; (e) conventional SBPTSC with a glass envelope. 7 3. Results and discussion 3.1. Effects of the absorption coefficient of the blended plasmonic nanofluid Absorption coefficient α of the blended plasmonic nanofluid is an important parameter for the thermal performance of DAPTSCs by influencing the local heat generation inside the inner tube. In this section, effects of the absorption coefficient are investigated for the transparent DAPTSC and the coated DAPTSC, both with a glass envelope. The configuration of the two collectors are all the same except that a highly reflective metallic coating is applied on the upper half of the inner tube outer surface for the coated DAPTSC [see Fig. 2(a)] while no such a coating exists for the transparent DAPTSC [see Fig. 2(c)]. The inlet tem- perature T0 is set to be 550 K, which is a typical working temperature for parabolic trough solar collector [4, 5], the flow rate m is chosen as 0.1 kg/s, which is comparable to other studies [33, 37], the inner tube inner diameter D is chosen as 7 cm (according to the LS-3 receiver), and the collector length L is chosen as 5 m for all the evaluation. It can be seen from Fig. 3(a) that the solar weighted absorption coefficient Am keeps increasing with increasing α for both coated and transparent DAPTSCs. The increasing rate gets slower with increasing α and becomes zero near α = 0.4 cm−1 for the coated DAPTSC and near α = 0.8 cm−1 for the transparent DAPTSC. Note that 1−e−0.4×2D (for the coated DAPTSC) = 1−e−0.8×D (for the transparent DAPTSC) = 0.996, indicating that 99.6% of the solar energy entering the inner tube is absorbed by the nanofluid, which is also reflected in Fig. 3(b). For instance, when α = 0.8 cm−1 for the transparent DAPTSC, the heat source in the upper half of the inner tube is quite low because almost all the energy has been absorbed at the lower part. It is interesting to see that the maximum value of Am for the coated DAPTSC (i.e., 0.783) is slightly lower than that for the transparent DAPTSC (i.e., 0.793). This is because about 1% (i.e., D/W ≈ 0.01) of the solar radiation is blocked by the reflective metal coating for the coated DAPTSC. Similarly, the collector thermal efficiency η also increases with the increase of α at first and then starts to decrease when exceeding certain α value [e.g, about 0.4 cm−1 for the coated DAPTSC or about 0.8 cm−1 for transparent DAPTSC; refer to Fig. 3(a)]. The reason is that, at lower value of α, the increase of α causes more solar energy to be absorbed in the nanofluid as represented by Am in Fig. 3(a). However, when α > 0.4 cm −1 for the coated DAPTSC or α > 0.8 cm −1 for the transparent DAPTSC, the contribution by increasing α to Am becomes limited while the heat loss keeps increasing, as shown in Fig. 3(c). Here, Qloss is defined as the total heat loss (i.e., both heat convection and thermal radiation from the envelope outer surface) divided by the interception area of the parabola (i.e., W × L). The increased heat loss to the surrounding eventually leads to the decrease in the collector efficiency at those conditions. It can be also seen from Fig. 3(a) that at the same absorption coefficient α, the coated DAPTSC performs better than the transparent DAPTSC, especially for lower values of α region. At α = 0.2 cm−1, for instance, Am and η of the coated DAPTSC are much higher than those for the transparent DAPTSC. In 8 0.8 0.7 0.6 0.5 e 0.4 0.3 0.2 0.8 0.7 0.6 m A 0.5 0.4 0.3 0.2 0.1 130 120 110 100 9 0 8 0 ) 2 m / W ( s s o l q S olid curve: coated D A P T S C D ot curve: transparent D A P T S C 0.3 0.5 0.7 0.9 a (cm-1) transparent D A P T S C coated D A P T S C 70 0.1 0.3 0.5 a (cm-1) 0.7 0.9 Figure 3: Effect of the absorption coefficient on: (a) solar weighted absorption coefficient Am and collector thermal efficiency η; (b) volumetric heat generation in the nanofluid; (c) heat loss from the receiver; and (d) temperature distribution at the the tube outlet. In the calculation, we set T0 = 550 K, D = 7 cm, L = 5 m, and m = 0.1 kg/s. other words, the coated DAPTSC requires lower value of α, thus lower nanoparticle concentration [28] than the transparent DAPTSC for the same level of the thermal performance, which is beneficial for avoiding particle agglomeration and maintaining stable operation. If α > 0.5 cm−1, then the advantage of the coated DAPTSC becomes minimal because the absorption of the solar radiation is already high enough (i.e., 1 − e−0.5×D = 0.969) for the transparent DAPTSC. Nevertheless, due to the low emissivity of the metallic reflective coating, the coated DAPTSC still exhibits the thermal efficiency of about 0.551, which is slightly higher than η = 0.520 of the transparent DAPTSC. It is important to note that the transparent DAPTSC with doubled α values of the coated DAPTSC (i.e., both collectors have the same value of 'α × optical path length') cannot perform the same as the coated DAPTSC. For instance, η of the transparent DAPTSC with α = 0.4 and 0.8 cm−1 is lower than η of the coated DAPTSC with α = 0.2 and 0.4 cm−1, respectively [refer to Fig. 3(a)]. The main reason is that the emissivity of the inner tube surface for the coated DAPTSC (i.e., half glass surface and half metallic 9 coating) is lower than that of the transparent DAPTSC (whole glass surface), leading to lower radiation loss. Another reason is that when α is doubled for the transparent DAPTSC, the heat source as well as the resulting temperature distribution will be not as uniform as the coated DAPTSC at the corresponding α values (i.e., half of the transparent DAPTSC), leading to locally high temperature on the inner tube surface thus more heat loss [refer to Figs. 3(b) and 3(d)]. 3.2. Effects of the inner diameter of the inner tube In this section, effects of the inner diameter of the inner tube, D, is investigated, and comparisons are made for the coated DAPTSC, the transparent DAPTSC and the SBPTSC with black chrome coating, all with a glass envelope. For fair comparison, we set αD to be constant for different values of D for nanofluid- based DAPTSCs. Here, when D is varied, the envelope diameter also changes accordingly to make the annulus spacing unchanged. 0.6 0.55 0.5 e 0.45 coated D A P T S C transparent D A P T S C S BP T S C 0.4 0.35 0.6 0.55 0.5 0.45 0.4 0.35 e 3 4 5 6 7 8 9 10 11 D (cm) coated D A P T S C transparent D A P T S C S BP T S C 3 4 5 6 7 8 9 10 11 D (cm) ) 2 m / W ( s s o l q 350 300 250 200 150 100 50 0 3 350 300 250 ) 2 m / 200 W ( s s o l q 150 100 50 0 3 coated D A P T S C transparent D A P T S C S BP T S C 4 5 6 7 8 9 10 11 D (cm) coated D A P T S C transparent D A P T S C S BP T S C 4 5 6 7 8 9 10 11 D (cm) Figure 4: Effect of the tube inner diameter on: (a) and (c) collector thermal efficiency; and (b) and (d) heat loss from the receiver. In the calculation, we set T0 = 550 K, m = 0.1 kg/s, and L = 5 m. Note that in (a) and (b), 1 − e−αD = 0.753, and in (c) and (d), 1 − e−αD = 0.969. 10 Figures 4(a) and 4(b) show that η of all the three collectors decreases with increasing D due to nearly constant Am value (i.e., 0.735, 0.597 and 0.849 for the coated DAPTSC, the transparent DAPTSC and the SBPTSC, respectively) as well as increasing heat loss. Besides, the decrease rate of η for DAPTSCs is much slower than the SBPTSC owing to their much lower heat loss than the SBPTSC, which arises from the fact that the inner tube surface temperature of the SBPTSC is much higher than the nanofluid-based DAPTSCs (not shown here). Figure 4(a) suggests that the DAPTSCs are more suitable for collectors with large diameter than the SBPTSC. For instance, the coated DAPTSC performs better than the SBPTSC when D > 7 cm. It should also be noted that for the given level of αD value, the DAPTSCs require smaller α (i.e., lower particle concentrations) with increasing D, which is beneficial for avoiding particle agglomeration and maintaining stable collector performance. Because α is inversely proportional to D when αD = const, dα dD becomes lager when D is smaller. In principle, larger η and smaller α are desirable; thus, there exists an optimal value of D where the DAPTSCs can gain a reasonably high η at a sufficiently low α for stable operation. For instance, the coated DAPTSC with α = 0.39 cm−1 in Fig. 4(a) has η = 0.538, which is more advantageous than the one obtaining slightly higher η = 0.543 but with much higher value α = 0.5 cm−1. Now the value of αD increases from 1.4 to 3.5 in Figs. 4(c) and 4(d), along with the same SBPTSC as in Figs. 4(a) and 4(b) for comparison. We can see that the DAPTSCs with αD = 3.5 show higher η values. This is mainly due to the increased Am (i.e., 0.783 and 0.769 for the coated DAPTSC and the transparent DAPTSC, respectively). Comparison of Fig. 4(a) with Fig. 4(c) reveals that for the nanofluid of the same α, DAPTSCs with different D values may achieve the same level of the performance η. For instance, when α = 0.35 cm−1, the coated DAPTSC with D = 4 cm or D = 10 cm can both result in η ≈ 0.53. Even though Am for D = 4 cm case in Fig. 4(a) is 0.74 and smaller than 0.78 for D = 10 cm in Fig. 4(c), the corresponding heat loss is about 56 W/m2 in Fig. 4(b) and much less than 102 W/m2 in Fig. 4(d). Therefore, to obtain the same η (or similar level of performance) with nanofluid of the same α, there could be multiple possible D values to choose. 3.3. Effects of the flow rate The effect of flow rate on the collector performance is investigated here in the range from 0.02 to 0.3 kg/s with the corresponding ReD = 600 ∼ 9000. Performance of three collectors including the coated DAPTSC, the transparent DAPTSC and the SBPTSC are compared in Fig. 5. It can be seen from Fig. 5(a) that as the flow rate m increases, the collector thermal efficiency η increases but its increasing rate slows down gradually. This can be explained based on the following two perspectives. First, recall that η is determined by Eq. (6) and any change in η is due to the variation of the three parameters, i.e., m, ∆T (or ¯Tout − T0) and cp. With the increase of m, ∆T first decreases significantly and then the decrease rate becomes slower, as shown in Fig. 5(b). In addition, cp also decreases with slower decreasing rate as m increases because it has approximately linear relation with the average temperature 11 0.6 0.55 0.5 0.45 0.4 0.35 0.3 e 0.25 0.02 3500 ) K - g k / J ( p c 3000 2500 2000 coated D A P T S C transparent D A P T S C S BP T S C 0.06 0.1 0.14 0.18 0.22 0.26 0.3 m (kg/s) coated D A P T S C transparent D A P T S C S BP T S C 0.06 0.1 0.14 0.18 0.22 0.26 0.3 m (kg/s) coated D A P T S C transparent D A P T S C S BP T S C 300 250 200 150 100 50 ) K ( T l e d 0 0.02 600 500 400 ) 2 m / W ( 300 s s o l q 200 100 1500 300 400 500 T (K) m 600 700 8 00 0 0.02 0.06 0.1 0.14 0.18 0.22 0.26 0.3 Figure 5: Effects of the flow rate on: (a) thermal efficiency; (b) temperature gain; (c) temperature-dependent heat capacity of Therminol VP-1; and (d) heat loss. In the calculation, we set α = 0.5 cm−1, T0 = 550 K, D = 7 cm, and L = 5 m. Tm (i.e., T0+ ¯Tout 2 = T0 + ∆T /2), as noted from Fig. 5(c). Although both ∆T and cp decrease with increasing m, increase in m leads to the increasing η at initial stage. To be more specific, when m < 0.06 kg/s, the decreasing effects of ∆T and cp are less than the increasing effect of m, so there is an obvious increase for η. For instance, for the coated DAPTSC, when m changes from 0.02 to 0.05 kg/s (increased to 250%), ∆T decreases from 266.46 to 125.79 K (decreased to 47%) and the corresponding decrease of cp is decreased from 2594 to 2361 J/kg-K (decreased to 91%), resulting in η increased to 107% ≈ 2.5 × 0.47 × 0.91. For both DAPTSCs, when m > 0.1 kg/s, the decreasing effect of ∆T and cp become nearly offset to the increasing effect of m, leading to almost constant η. The second perspective for explaining the change of η with m is through the energy balance analysis. Since m does not affect Am, the solar energy absorbed in each collector is independent of m. As a result, the change of η depends mainly on the change of heat loss Qloss; that is, the higher Qloss is, the lower η will be. This is easily recognized in Figs. 5(a) and 5(d). Furthermore, the effect of m on Qloss can be understood as follows. At very low values of m, the plasmonic nanofluid spends much time in the receiver tube; thus, 12 higher ∆T can be gained as in Fig. 5(b). The higher ∆T in turn leads to the higher temperature on the receiver surface and consequently more Qloss. When m increases, the corresponding ∆T decreases, and so does the Qloss. The trend of Qloss is actually very similar to that of ∆T for the SBPTSC. While for the DAPTSCs, Qloss is much lower than that of the SBPTSC as the temperature on the inner tube surface is much lower than that of the SBPTSC owing to the volumetric absorption of the solar radiation. It is also noted that η of the DAPTSCs is higher than the SBPTSC until m = 0.1 kg/s for the transparent DAPTSC and until m = 0.16 kg/s for the coated DAPTSC. This is because when m > 0.2 kg/s, the temperature gain ∆T for all the collectors are as low as about 34 K, leading to much less heat loss, especially for the SBPTSC. On the other hand, Am for the SBPTSC is 0.849; that is, much higher than 0.783 and 0.770 for the coated and transparent DAPTSC, respectively. Therefore, the SBPTSC gradually surpasses the transparent and coated DAPTSCs. Hence, the DAPTSCs are more desirable for low flow rate operation conditions (e.g., m < 0.16 kg/s for the coated DAPTSC) while the SBPTSC works better for higher flow rate conditions (e.g., m > 0.18 kg/s). 3.4. Effects of the inlet temperature In this section, effects of the inlet temperature on the thermal efficiency of five PTSCs (refer to Fig. 2) are studied and compared. The effect of the outer glass envelope for preventing the convection heat loss is also investigated for the two DAPTSCs (i.e., coated and transparent DAPTSCs). Note that the outer glass envelope is necessary for the SBPTSC for fair comparison due to its high surface temperature and thus high heat loss. First of all, it is seen from Fig. 6(a) that the collector thermal efficiency η of all configurations decreases with increasing inlet temperature due to the increasing heat loss [see Fig. 6(b)]. Secondly, Fig. 6(a) also suggests that the best-performance configuration among the five collectors largely depends on the inlet temperature T0. When T0 = 350 K, for instance, the η for the DAPTSCs are higher than the SBPTSC due to the high heat loss of the SBPTSC. Interestingly, the DAPTSCs without the glass envelop performs better than the corresponding DAPTSCs with the glass envelope. This indicates that the DAPTSCs without the glass envelope become more advantageous than the DAPTSCs with the glass envelope at low temperature owing to their higher Am. In other words, at lower temperature applications, improving the solar weighted absorption coefficient by removing the glass envelope (∵ τgla = 0.94 < 1.0) is more crucial than decreasing the convection heat loss. When T0 > 500 K, however, the DAPTSCs with the glass envelope outperform the corresponding DAPTSC without the glass envelope. It is also confirmed again that the coated DAPTSCs are better than the transparent DAPTSC both with or without the envelope because the metallic coating doubles the optical path length and reduces the emissivity. As to the SBPTSC, its η is lower than the nanofluid-based DAPTSCs at low temperature, but surpasses the configurations of (b), (c) and (d) sequentially at 510, 550 and 700 K. It is worth mention that η of the transparent DAPTSCs with the envelope is very close to that of the SBPTSC when 350 < T0 < 550 K. On the other hand, the 13 0.8 0.7 0.6 0.5 0.4 0.3 0.2 e (a) (b) (c) (d) (e) (a) (b) (c) (d) (e) 0.1 350 400 450 500 550 T (K) 0 (a) (b) (c) (d) (e) 600 500 400 300 200 100 ) 2 m / W ( s s o l q 0 350 400 450 500 550 (K) 0 T 600 650 700 750 600 650 700 750 Figure 6: Effects of the inlet temperature on (a) collector thermal efficiency and (b) heat loss. In the calculation, we set α = 0.5 cm−1, D = 7 cm, L = 5 m, and m = 0.1 kg/s. coated DAPTSC performs better than the SBPTSC until up to T0 = 750 K mainly owing to the lower temperature on the inner tube surface than that of the SBPTSC. 4. Conclusion In the present study, a partially coated direct-absorption parabolic trough solar collector utilizing the blended plasmonic nanofluid with uniform absorption coefficient has been proposed and investigated. In the proposed coated DAPTSC, the upper half of the inner tube's outer surface is coated by a reflective metallic layer so that the optical path length of the concentrated sun rays can be doubled as well as its infrared emissivity can be reduced significantly. Key factors influencing the collector performance have also been investigated. The following conclusions have been drawn. • For a given inner tube's diameter of D = 7 cm, the absorption coefficient of the blended plasmonic 14 nanofluid directly affects the thermal performance of DAPTSCs, and higher α leads to higher η until a certain value (i.e., α = 0.4 cm−1 for the coated DAPTSC and α = 0.8 cm−1 for the transparent DAPTSC). The proposed coated DAPTSC was found to be advantageous as compared to the trans- parent DAPTSC, especially at low α values (e.g., α < 0.5 cm−1). Therefore, to achieve the same level of η, the coated DAPTSC requires less nanoparticle concentration, which is beneficial for stable operation. • The collector thermal efficiency was found to be decreasing with increasing the diameter of the inner tube. We showed that there exists an optimal diameter D for the DAPTSCs at which a reasonably high η can be obtained with relatively low α if αD is fixed. • The DAPTSCs (both coated and transparent) were found to be more desirable when m ≤ 0.1 kg/s, while the SBPTSC works better for the higher flow rate (e.g., m > 0.18 kg/s). This is because when m > 0.2 kg/s, the temperature gain ∆T for all the collectors are as low as about 34 K, resulting in much less heat loss, especially for the SBPTSC. • The DAPTSCs (both coated and transparent) with the glass envelope outperform the corresponding DAPTSC without the glass envelope until up to T0 = 500 K for the transparent DAPTSC and up to T0 = 550 K for the coated DAPTSC. It was found that for the low-temperature inlet conditions (e.g., 350 < T0 < 430 K), the transparent DAPTSC without the glass envelope is the best configuration. On the other hand, the coated DAPTSC with the glass envelope performs better than the SBPTSC until up to T0 = 750 K. The results obtained in this work will facilitate the use of nanofluids in a parabolic trough solar collector with more stability and provide a guide for choosing suitable type of parabolic trough solar collectors for specific working conditions. Acknowledgments This research was supported by Basic Science Research Program (NRF-2015R1A2A1A10055060) as well as Pioneer Research Center Program (NRF-2013M3C1A3063046) through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT. References References [1] I. H. Yılmaz, A. Mwesigye, Modeling, simulation and performance analysis of parabolic trough solar collectors: A com- prehensive review, Appl Energy 225 (2018) 135 -- 174. 15 [2] S. A. Kalogirou, Solar energy engineering: processes and systems, Academic Press, 2013. [3] M. Fan, H. Liang, S. You, H. Zhang, W. Zheng, J. Xia, Heat transfer analysis of a new volumetric based receiver for parabolic trough solar collector, Energy 142 (2018) 920 -- 931. [4] J. P. Freedman, H. Wang, R. S. Prasher, Analysis of nanofluid-based parabolic trough collectors for solar thermal appli- cations, J Sol Energ Eng 140 (5) (2018) 051008. [5] V. Khullar, H. Tyagi, P. E. Phelan, T. P. Otanicar, H. Singh, R. A. Taylor, Solar energy harvesting using nanofluids-based concentrating solar collector, J Nanotechnol Eng Med 3 (3) (2012) 031003. [6] H. Tyagi, P. Phelan, R. Prasher, Predicted efficiency of a low-temperature nanofluid-based direct absorption solar collector, J Sol Energ Eng 131 (4) (2009) 041004. [7] T. Otanicar, P. E. Phelan, R. S. Prasher, G. Rosengarten, R. A. Taylor, Nanofluid-based direct absorption solar collector, J Renew Sustain Ener 2 (3) (2010) 033102. [8] E. Sani, L. Mercatelli, S. Barison, C. Pagura, F. Agresti, L. Colla, P. Sansoni, Potential of carbon nanohorn-based suspensions for solar thermal collectors, Sol Energ Mat Sol C 95 (11) (2011) 2994 -- 3000. [9] B. J. Lee, K. Park, T. Walsh, L. Xu, Radiative heat transfer analysis in plasmonic nanofluids for direct solar thermal absorption, J Sol Energ Eng 134 (2) (2012) 021009. [10] A. Veeraragavan, A. Lenert, B. Yilbas, S. Al-Dini, E. N. Wang, Analytical model for the design of volumetric solar flow receivers, Int J Heat Mass Tran 55 (4) (2012) 556 -- 564. [11] H. K. Gupta, G. D. Agrawal, J. Mathur, Investigations for effect of Al2O3-H2O nanofluid flow rate on the efficiency of direct absorption solar collector, Case Studies in Thermal Eng 5 (2015) 70 -- 78. [12] V. Cregan, T. Myers, Modelling the efficiency of a nanofluid direct absorption solar collector, Int J Heat Mass Tran 90 (2015) 505 -- 514. [13] T. B. Gorji, A. A. Ranjbar, Geometry optimization of a nanofluid-based direct absorption solar collector using response surface methodology, Sol Energy 122 (2015) 314 -- 325. [14] J. Jeon, S. Park, B. J. Lee, Analysis on the performance of a flat-plate volumetric solar collector using blended plasmonic nanofluids, Sol Energy 132 (2016) 247 -- 256. [15] C. Qin, K. Kang, I. Lee, B. J. Lee, Optimization of a direct absorption solar collector with blended plasmonic nanofluids, Sol Energy 150 (2017) 512 -- 520. [16] K. H. Won, B. J. Lee, Effect of light scattering on the performance of a direct absorption solar collector, Front Energy 12 (1) (2018) 169 -- 177. [17] C. Qin, K. Kang, I. Lee, B. J. Lee, Optimization of the spectral absorption coefficient of a plasmonic nanofluid for a direct absorption solar collector, Sol Energy 169 (2018) 231 -- 236. [18] A. R. Mallah, S. Kazi, M. N. M. Zubir, A. Badarudin, Blended morphologies of plasmonic nanofluids for direct absorption applications, Appl Energy 229 (2018) 505 -- 521. [19] J. Jeon, S. Park, B. J. Lee, Optical property of blended plasmonic nanofluid based on gold nanorods, Opt Express 22 (104) (2014) A1101 -- A1111. [20] M. Chen, Y. He, J. Zhu, B. Jiang, Y. Huang, An experimental investigation on sunlight absorption characteristics of silver nanofluids, Sol Energy 115 (2015) 85 -- 94. [21] M. Chen, Y. He, J. Zhu, D. R. Kim, Enhancement of photo-thermal conversion using gold nanofluids with different particle sizes, Energ Convers Manage 112 (2016) 21 -- 30. [22] M. Du, G. H. Tang, Plasmonic nanofluids based on gold nanorods/nanoellipsoids/nanosheets for solar energy harvesting, Sol Energy 137 (2016) 393 -- 400. [23] V. Bhalla, H. Tyagi, Parameters influencing the performance of nanoparticles-laden fluid-based solar thermal collectors: A review on optical properties, Renew Sustain Energ Rev 84 (2018) 12 -- 42. 16 [24] Z. Wang, Z. M. Zhang, X. Quan, P. Cheng, A numerical study on effects of surrounding medium, material, and geometry of nanoparticles on solar absorption efficiencies, Int J Heat Mass Tran 116 (2018) 825 -- 832. [25] A. Menbari, A. A. Alemrajabi, A. Rezaei, Heat transfer analysis and the effect of CuO/water nanofluid on direct absorption concentrating solar collector, Appl Therm Eng 104 (2016) 176 -- 183. [26] G. J. O'Keeffe, S. L. Mitchell, T. G. Myers, V. Cregan, Modelling the efficiency of a low-profile nanofluid-based direct absorption parabolic trough solar collector, Int J Heat Mass Tran 126 (2018) 613 -- 624. [27] G. J. O'Keeffe, S. L. Mitchell, T. G. Myers, V. Cregan, Modelling the efficiency of a nanofluid-based direct absorption parabolic trough solar collector, Sol Energy 159 (2018) 44 -- 54. [28] C. F. Bohren, D. R. Huffman, Absorption and Scattering of Light by Small Particles, Wiley, 1983. [29] A. De Risi, M. Milanese, D. Laforgia, Modelling and optimization of transparent parabolic trough collector based on gas-phase nanofluids, Renew Energy 58 (2013) 134 -- 139. [30] F. P. Incropera, D. P. Dewitt, T. L. Bergman, A. S. Lavine, Principles of heat and mass transfer, John Wiley & Sons, 2013. [31] T. B. Gorji, A. A. Ranjbar, A numerical and experimental investigation on the performance of a low-flux direct absorption solar collector (DASC) using graphite, magnetite and silver nanofluids, Sol Energy 135 (2016) 493 -- 505. [32] Z. D. Cheng, Y. L. He, F. Q. Cui, B. C. Du, Z. J. Zheng, Y. Xu, Comparative and sensitive analysis for parabolic trough solar collectors with a detailed Monte Carlo ray-tracing optical model, Appl Energy 115 (2014) 559 -- 572. [33] Q. Li, C. Zheng, S. Mesgari, Y. L. Hewkuruppu, N. Hjerrild, F. Crisostomo, G. Rosengarten, J. A. Scott, R. A. Taylor, Experimental and numerical investigation of volumetric versus surface solar absorbers for a concentrated solar thermal collector, Sol Energy 136 (2016) 349 -- 364. [34] D. Rativa, L. A. G´omez-Malag´on, Solar radiation absorption of nanofluids containing metallic nanoellipsoids, Sol Energy 118 (2015) 419 -- 425. [35] Solutia, Therminol VP-1 Vapor Phase, Liquid Phase Heat Transfer Fluid 12◦C to 400◦C, http://twt.mpei.ac.ru/tthb/hedh/htf-vp1.pdf (2013). [36] A. Lenert, E. N. Wang, Optimization of nanofluid volumetric receivers for solar thermal energy conversion, Sol Energy 86 (1) (2012) 253 -- 265. [37] S. Dugaria, M. Bortolato, D. Del Col, Modelling of a direct absorption solar receiver using carbon based nanouids under concentrated solar radiation, Renew Energy 128 (2018) 495 -- 508. [38] B. E. Launder, D. B. Spalding, The numerical computation of turbulent flows, in: Numerical Prediction of Flow, Heat Transfer, Turbulence and Combustion, Elsevier, 1983, pp. 96 -- 116. [39] T. L. Bergman, F. P. Incropera, D. P. DeWitt, A. S. Lavine, Fundamentals of heat and mass transfer, John Wiley & Sons, 2011. [40] W. M. Kays, Turbulent prandtl number -- where are we?, J Heat Transf 116 (2) (1994) 284 -- 295. [41] R. Forristall, Heat transfer analysis and modeling of a parabolic trough solar receiver implemented in engineering equation solverdoi:10.2172/15004820. [42] G. Xu, W. Chen, S. Deng, X. Zhang, S. Zhao, Performance evaluation of a nanofluid-based direct absorption solar collector with parabolic trough concentrator, Nanomaterials 5 (4) (2015) 2131 -- 2147. [43] F. Wang, Q. Lai, H. Han, J. Tan, Parabolic trough receiver with corrugated tube for improving heat transfer and thermal deformation characteristics, Appl Energy 164 (2016) 411 -- 424. [44] W. C. Swinbank, Long-wave radiation from clear skies, Q J Roy Meteor Soc 89 (381) (1963) 339 -- 348. [45] J. A. Duffie, W. A. Beckman, Solar Engineering of Thermal Processes, John Wiley & Sons, 2013. [46] V. E. Dudley, G. J. Kolb, A. R. Mahoney, T. R. Mancini, C. W. Matthews, M. Sloan, D. Kearney, Test results: SEGS LS-2 solar collector (1994). 17
1907.10751	1	1907	2019-05-24T05:24:48	Thermal characterization of convective heat transfer in microwires based on modified steady state "hot wire" method	[ "physics.app-ph" ]	The convection plays a very important role in heat transfer when MEMS work under air environment. However, traditional measurements of convection heat transfer coefficient require the knowledge of thermal conductivity, which makes measurements complex. In this work, a modified steady state "hot wire" (MSSHW) method is proposed, which can measure the heat transfer coefficient of microwires' convection without the knowledge of thermal conductivity. To verify MSSHW method, the convection heat transfer coefficient of platinum microwires was measured in the atmosphere, whose value is in good agreement with values by both traditional measurement methods and empirical equations. Then, the convection heat transfer coefficient of microwires with different materials and diameters were measured by MSSHW. It is found that the convection heat transfer coefficient of microwire is not sensitive on materials, while it increases from 86 W/(m$^2$K) to 427 W/(m$^2$K) with the diameter of microwires decreasing from 120 ${\mu}$m to 20 ${\mu}$m. Without knowing thermal conductivity of microwires, the MSSHW method provides a more convenient way to measure the convective effect.	physics.app-ph	physics	Thermal characterization of convective heat transfer in microwires based on modified steady state "hot wire" method Xiaoman Wang1,2,#, Rulei Guo1,2,#, Qinping Jian1,2,#, Guilong Peng1,2, Yanan Yue3,Nuo Yang1,2 1State Key Laboratory of Coal Combustion, Huazhong University of Science and Technology (HUST), Wuhan 430074, People's Republic of China 2Nano Interface Center for Energy (NICE), School of Energy and Power Engineering, Huazhong University of Science and Technology (HUST), Wuhan 430074, People's Republic of China 3School of Power and Mechanical Engineering, Wuhan University, Wuhan 430072, People's Republic of China #XW, RG, and QJ contribute equally on this work. Corresponding author E-mail: [email protected] (NY) ABSTRACT The convection plays a very important role in heat transfer when MEMS work under air environment. However, traditional measurements of convection heat transfer coefficient require the knowledge of thermal conductivity, which makes measurements complex. In this work, a modified steady state "hot wire" (MSSHW) method is proposed, which can measure the heat transfer coefficient of microwires' convection without the knowledge of thermal conductivity. To verify MSSHW method, the convection heat transfer coefficient of platinum microwires was measured in the atmosphere, whose value is in good agreement with values by both traditional measurement methods and empirical equations. Then, the convection heat transfer coefficient of microwires with different materials and diameters were measured by MSSHW. It is found that the convection heat transfer coefficient of microwire is not sensitive on materials, while it increases from 86 W/(m2 ∙ K) to 427 W/(m2 ∙ K) with the diameter of microwires decreasing from 120 μm to 20 μm. Without knowing thermal conductivity of microwires, the MSSHW method provides a more convenient way to measure the convective effect. Keywords: hot wire method, micro wire, convection heat transfer coefficient, size effect, thermal conductivity INTRODUCTION With the prosperous development of micro-electro-mechanical systems (MEMS), heat dissipation within highly integrated circuits has drawn wide attention[1]. The chip- level heat generated by the increasing power density is pushing the demand of better cooling methods[2]. If the trend of integration and miniaturization keeps following the International Technology Roadmap for Semiconductors (ITRS), thermal management will become the bottleneck for further development of the electronic devices[2-4]. In the past two decades, the heat transfer in one-dimensional (1-D) structures has been investigated, which are widely employed in MEMS. It is important for understanding the thermal properties of materials[5,6], enhancing heat transfer in MEMS theoretically and experimentally [7-9], and enriching applications in energy conversion, such as nanowire solar cell[10,11] and thermoelectric application[12,13]. According to previous works, there are several methods to measure the thermal conduction of 1-D microwires including the steady-state hot wire method [14], the 3ω method [15-17] and the Raman spectroscopy method [18-20]. Among these methods, the steady-state hot wire method is one of the most convenient and simplest method to measure the thermal conduction of 1-D structure. To exclude convective heat transfer effect, these approaches are normally carried out in a vacuum environment. Besides the thermal conduction, convective heat transfer has also been investigated. It has been proved that the convection shows different performances when the size of sample shrinks to micro/nanoscale. For example, convection heat transfer coefficient continuously increases from 5~10 W/(m2 ∙ K) (natural convection on bulk material) [21] to over 4000 W/(m2 ∙ K) (micro carbon fiber with a diameter of 4.3 μm) [22].Moreover, convection heat transfer coefficient of carbon nanotubes with a diameter of 1.47 nm can reach 8.9 × 104 W/(m2 ∙ K) in an atmosphere environment [23]. Such high convection heat transfer coefficient under micro/nanoscale indicates that the convection plays a significant role in heat dissipation in MEMS devices. There are three main methods to measure the convection heat transfer coefficient of 1-D structure. The first one is the modified 3ω method[24]. By adding convection part into the original model, the convection heat transfer coefficient can be measured when other thermal properties are known, including thermal conductivity. The second one is Raman mapping measurement. By comparing Raman temperature measurement result with the value from resistance change by joule heating, Zhang's group searched for the optical absorption coefficient and then use the best fitting result to get the convection heat transfer coefficient during iteration[22]. The third one is taking advantage of the steady-state method to analyze the air side[25,26]. However, this method ignores the heat conduction within the sample. Theoretical calculation has shown that the heat conduction can be overlooked only when the length of the tested sample has reached a certain value[23]. Since conductive and convective effects are coupled in heat transfer of MEMS and traditional methods ignore thermal conduction or get it by other ways, a method for characterizing convective heat transfer effect without the knowledge of thermal conduction is convenient and important. Yue et al. proposed a method to study conductive and convective heat transfer of microwires simultaneously by using steady- state Joule-heating and Raman mapping together[7]. However, the expensive Raman equipment and the relatively large signal error limit its broader application. In this paper, we proposed a modified steady state hot wire (MSSHW) method which can characterize convection heat transfer coefficient of microwires without the knowledge of thermal conductivity. Firstly, we theoretically derived the mathematical model of MSSHW method. Secondly, we measured convection heat transfer coefficient of a platinum microwire to verify MSSHW method. Then, using MSSHW method, we studied the material dependence of convection heat transfer coefficient by measuring microwires of platinum, stainless steel and tungsten. Lastly, microwires with different diameters ranging from 20 to 120 μm were measured to study the size effect on convection heat transfer coefficient. This work presents a convenient experimental method, MSSHW method, to study the heat transfer at microscale by convection. MODEL & METHOD The schematic illustration of experimental setup is shown in Fig. 1a. The sample is connected to a multimeter (Keithley 2700) and a current Source (Keithley 6221) in a four-electrodes configuration. A direct current is applied on the two outside electrodes and the corresponding voltage is measured on the two inside electrodes, which diminishes the contact electric resistance. The photo of experimental setup is shown in Fig. 1b. When a direct current is applied to an electroconductive microwire, the microwire will heat itself up due to Joule heating. The temperature increase is closely related to both the conduction and the convection of the microwire, when the radiation can be neglected. When a series of microwires with different length has been measured, the thermal conductivity can be eliminated from the calculation of convection heat transfer coefficient. Thus, it provides a possibility to measure the convection heat transfer coefficient without knowing thermal conductivity. Fig.1 (a) The sketch schematic illustration of experimental setup with a four-electrodes configuration (not to scale). (b) Photo of experimental setup. The heat transfer in a microwire can be described by 1-D heat transfer equation as: 𝜅𝐴 𝑑2𝑇(𝑥) 𝑑𝑥2 − ℎ𝑃(𝑇(𝑥) − 𝑇𝐸) − 𝜀𝜎𝑃(𝑇4(𝑥) − 𝑇𝐸 4) + 𝑞𝐴 = 0 (1) where κ is the thermal conductivity, A is the cross area, h is the convection heat transfer coefficient, 𝑇𝐸 is the environment temperature, P is the perimeter, ε is the emissivity, σ is the Stefan-Boltzmann constant as 5.67× 10−8 𝑊/(𝑚2 ∙ 𝐾4)，and q is the heat generated per unit volume. The four terms in the left side represent the thermal conduction, the natural convection, the radiation and the heat generated inside the thin wire, respectively. The contribution from radiation is much smaller than that from conduction and convection, which can be therefore neglected. To confirm this assumption, the temperature profiles were calculated with/without radiation according to Eq. 1 by MATLAB. The calculated sample is a platinum microwire, whose length and diameter are 16.3 mm and 32.6 μm, respectively [24]. The size of the calculated sample is a typical situation in the real experiments. By applying 𝜅 = 66.5 𝑊/(𝑚 ∙ 𝐾) , ℎ = 410 𝑊/(𝑚2 ∙ 𝐾) [24] and 𝜀 = 0.075 into Eq. 1 [27], we can get the temperature profile of the wire as shown in Fig. 2a. We calculated the average temperature of four microwires with different lengths as shown in Fig. 2b, and radiation shows negligible influence on the temperature profile of those microwires. Fig.2 (a) The temperature distribution for platinum wire with d=32.6 μm, l=16.3 mm under the current of 60 mA with and without radiation effect considered. (b). Average temperatures for platinum wires with different lengths (d=32.6 μm). Blue down triangles represent values when radiation effect is considered. Red up triangles represent the values when radiation effect is ignored Thus, the 1-D heat transfer equation can be simplified as 𝜅𝐴 𝑑2𝑇(𝑥) 𝑑𝑥2 − ℎ𝑃(𝑇(𝑥) − 𝑇𝐸) + 𝑞𝐴 = 0 (2) Since the heat sinks have much larger heat capacities compared to the thin wire, the boundary condition is thus expressed as 𝑇(0) = 𝑇𝐸 { 𝑇(𝐿) = 𝑇𝐸 where L is the length of the microwire. The solution of the 1-D heat transfer equation is 𝑞 (− 𝑘𝑚2𝑒2𝑚𝐿+ 𝑘𝑚2𝑒𝑚𝐿)𝑒−𝑚𝑥+(− 𝑘𝑚2𝑒𝑚𝐿+ 𝑞 𝑞 𝑞 𝑘𝑚2)𝑒𝑚𝑥 𝑒2𝑚𝐿−1 𝑇(𝑥) = 𝑇𝐸 + 𝑞 𝑘𝑚2 + (3) (4) 𝑚 = √ℎ𝑃 𝜅𝐴 Then, 𝑇(𝑥) is integrated over the length and the average temperature increase is calculated as ∆𝑇 = 1 𝐿 𝐿 ∫ 𝑇(𝑥)𝑑𝑥 − 𝑇𝐸 = 0 𝑞 𝜅𝑚2 (1 − 2(𝑒𝑚𝐿−1) 𝑚𝐿(𝑒𝑚𝐿+1) ) (5) It is noticed that, when the characteristic length of sample is in microscale, 𝑒𝑚𝐿 ≫ 1. Then, the right side of Eq. 5 can be simplified as ∆𝑇 = 𝑞 𝜅𝑚2 (1 − 2 𝑚𝐿 ) (6) The error made from this simplification is within 0.001%. Moreover, the resistance change caused by the average temperature change can be described with resistance change rate with temperature 𝑑𝑅 𝑑𝑇 [28] ∆𝑅 = 𝑑𝑅 𝑑𝑇 ∆𝑇 = 𝑑𝑅 𝑑𝑇 𝑞 𝜅𝑚2 (1 − 2 𝑚𝐿 ) (7) The resistance change rate with temperature 𝑑𝑅 𝑑𝑇 can be calculated using temperature coefficient of resistance 𝛽 shown as Eq. 8[28]. 𝑑𝑅 𝑑𝑇 = 𝛽𝑅0 (8) Where 𝑅0 is the resistance of the microwire at the environment temperature, and it can be calculated as[28] 𝑅0 = 𝜌 𝐿 𝐴 Where ρ is the electrical resistivity. The heat generated per unit volume q is calculated using Joule's first law 𝑞 = 𝐼2𝑅 𝑉 = 𝐼2𝑅 𝐴𝐿 ≈ 𝐼2𝑅0 𝐴𝐿 = 𝐼2𝜌 𝐿 𝐴 𝐴𝐿 Where 𝑉 is the volume of the wire sample. Then Eq. 7 can be expressed as: (9) (10) ∆𝑅 = 𝑎𝐿 − 𝑏 { 𝑎 = 𝑏 = 𝛽𝐼2𝜌2 𝐴3𝜅𝑚2 2𝛽𝐼2𝜌2 𝐴3𝜅𝑚3 (11) It shows that there is a linear relationship between the increase of resistance and the length of samples. Based on Eq. 11, the main equations of MSSHW method are derived out as ℎ = 16𝛽𝐼2𝜌2 𝜋3𝑑5∙𝑎 (12) where 𝑑 is the diameter of the thin wire and 𝐼 is the given current. That is, convection heat transfer coefficient (h) can be obtained by measuring linear relationship between ∆R and L without knowing thermal conductivity. Based on Eq. 11, the thermal conductivity can also be calculated as 𝜅 = 16𝛽𝐼2𝜌2∙𝑏2 𝜋3𝑑6∙𝑎3 (13) However, the equation of κ includes the intercept b which is a very small value compared with ∆R. The value of b is so small that it is highly uncertain. So that the thermal conductivity 𝜅 calculated by this equation has a high uncertainty. Therefore, we don't discuss the thermal conductivity here. To proof the validity of MSSHW method, we measured the natural convection heat transfer coefficient of platinum microwires. Firstly, the diameters of the platinum wire were measured by optical microscope and the picture is shown in Fig. 3b. By averaging the nine measured values at different positions of the microwire, the diameter of the platinum wire is obtained as 41 μm. Secondly, by measuring their resistances, the lengths of the samples are obtained based on Eq. 9 [28]. Thirdly, the increase of electric resistance (∆𝑅) of seven samples, whose lengths range from 20 mm to 130 mm, were measured under the current of 60 mA (shown in Fig. 4a). Finally, in accordance with Eq. 11, the parameters of a and b can be obtained by linear fitting. When the current was applied as 60 mA, a and b were obtained as 2.41±0.02 Ω/m and 0.015±0.002 Ω, respectively. Therefore, convection coefficient can be calculated out at a given direct current by Eq. 12 as 246±6 W/m2-K. Fig.3 The microscope photo of microwires (the diameter is measured from at least nine different positions of those microwires): (a) platinum: 20± 1 μm in diameter; (b) platinum: 41± 3 μm in diameter; (c) Tungsten: 40±1 μm in diameter; (d) stainless steel: 42±2 μm in diameter. (e) stainless steel: 120±6 μm in diameter. The convection heat transfer coefficient, measured under different direct current, are shown in Fig. 4b. As the results show, the convection coefficient is current independent. Of course, the current can't be too high or too low. The error of measurement will be too high due to the too low increase of average temperature or the radiation heat loss can not be ignored due to the too high increase of average temperature. Our results of convection coefficient agree well with values calculated by the empirical equations of Churchill, Morgan and Fujii. However, our results have a slight difference between Jan's empirical equation[29]. The difference can be attributed to the following factors: thermal conduction to the supports and the temperature measurement locations; distortion of the temperature and velocity fields by bulk fluid movements; the use of undersized containing chambers or the presence of the temperature system and supports; and temperature loading effects[32]. Fig.4 (a) The increase of electric resistance with length. A 60 mA current is applied in measurement. Red line represents the fitting line. (b) Experimental results of convection heat transfer coefficient with variable current. The black open circles represent values of convection coefficient results from our work (MSSHW method). The red solid line, blue dash line, magenta dot line, green dash-dot line and navy dash-dot-dot line are convection heat transfer coefficient calculated from the empirical equations of Fujii[30], Morgan[31], Churchill[32], Churchill*[32] and Jan[33], respectively. To study the material effect on the convection heat transfer coefficient, microwires (~40 μm in diameter) of platinum, tungsten, and stainless steel (Fig.3 b, c, d) are measured with MSSHW method. As shown in Fig. 5a, the values of convection heat transfer coefficient are independent on materials. The values of three materials with 40 μm diameter are close to each other, which is also close to the prediction values in Ref.[30- 32] by empirical equations. According to the optical microscope images, the slight difference in convection coefficients may be attributed to the surface roughness which has influence on heat convection[34]. Fig.5 (a) The natural convection heat transfer coefficients of the three microwires with different materials measured by MSSHW method. The diameter of platinum, stainless steel and tungsten microwires is 41 μm, 42 μm and 40 μm, respectively. (b) Experimental convection heat transfer coefficients under different diameters measured by MSSHW method. (black open circles were measured results by MSSHW method, the orange up triangles were measured by Hou[25], and the purple down triangles were measured by Tang[24].) Moreover, microwires with different diameters were measured to show the size effect of convection heat transfer coefficient (Fig. 5b). As we can see, the diameter largely affects convection heat transfer coefficient. With diameter decreasing from 120 μm to 20 μm, the convection heat transfer coefficient increases from 86 W/(m2 ∙ K) to 427 W/(m2 ∙ K) . The measured convection heat transfer coefficients decrease sharply when the diameter increases, and converge to a constant for bulk. This phenomenon might attribute to two physical mechanisms[35]. Firstly, when the diameter of the wire is comparable to the mean free path of the air molecular, the continuum theory breaks. Thus, the theory for the macroscopic phenomena is no longer applicable. Secondly, the surface volume ratio increases with the decreases of wire diameter, making the surface-related influences more significant [35]. In comparison, both experimental results[24,25] and empirical equations results[30-33] are also shown in Fig. 5b. Fig.6 (a) The percentage of convection heat transfer under different diameters. The length of those microwires is ~126mm (dots are measured results by MSSHW method and calculated by Eq. 14; lines are calculated by Eq.15 with parameters from Ref.[31]). (b) The percentage of convection heat transfer under different lengths. To get a better understanding of heat transfer in microscale, the percentage of heat transferred by convection (Qcv/Qtotal) has been investigated. Based on Eq.2 and Eq.10, Qcv/Qtotal can be obtained by measurements as 𝑄𝑐𝑣 𝑄𝑡𝑜𝑡𝑎𝑙 ⁄ = ℎ𝑃𝐿∆𝑇 𝐼2𝑅 Besides, from Eq. 5, the value of ratio can also be estimated as 𝑄𝑐𝑣 𝑄𝑡𝑜𝑡𝑎𝑙 ⁄ = ℎ𝑃𝐿∆𝑇 𝑞𝐴𝐿 = 1 − 2 𝑚𝐿 𝑒𝑚𝐿−1 ( 𝑒𝑚𝐿+1 ) (14) (15) Firstly, we studied the dependence of Qcv/Qtotal on the diameter of microwires. The data from measurements (Eq. 14) are matched well with the prediction curves by theory (Eq.15). As shown in Fig. 6a, with the diameter decreasing, Qcv/Qtotal increases. For example, for a steel microwire with a diameter of 120 μm and a length of 126mm, the percentage of heat flux carried by convection is more than 95%. So, convection plays a predominant role in heat transfer of microwires. Moreover, the ratio depends on the materials, as shown in Fig. 6a. The thermal convections of microwires are not sensitive on materials. However, the thermal conduction depends on materials. The thermal conductivity of steel, platinum and tungsten microwires measured by traditional steady state hot wire method are 18±3, 79±3 and 224±9 W/mK, respectively. Because the steel microwire has a lower value of thermal conductivity than platinum and tungsten. it has a higher value of Qcv/Qtotal. Secondly, we studied the dependence of Qcv/Qtotal on the length of microwires. Fig. 6b shows that, with the increasing of length, the ratio of Qcv/Qtotal increases. It means that the longer a wire is, the more important the convective heat transfer is. Because the area of convection increases with the length. CONCLUSION In conclusion, a modified steady-state hot wire (MSSHW) method is proposed to characterize the conductive and convective heat transfer of microwires simultaneously. The method is verified by measuring the convection heat transfer coefficient of microwires. The convective heat transfer between micro metal wires and their surrounding air environment is found to be irrelevant to material composition, but is strongly connected with the diameter of the wires. When the diameter of microwires decreases from 120 μm to 20 μm, the natural convection coefficient increases from 86 W/(m2 ∙ K) to 427 W/(m2 ∙ K) . The convection coefficient is in reasonable range compared with those from the references. This MSSHW method provides a convenient way for measuring convective heat transfer of microwires without knowing thermal conduction, which is beneficial to studying the comprehensive heat transfer at microscale ACKNOWLEDGEMENTS This work is financially supported by the National Natural Science Foundation of China (No. 51576076, No. 51711540031 and No. 51622303), the Natural Science Foundation of Hubei Province (No. 2017CFA046), and the Fundamental Research Funds for the Central Universities (No. 2016YXZD006). The authors thank the National Supercomputing Center in Tianjin (NSCC-TJ) and China Scientific Computing Grid (ScGrid) for providing assistance in computations. We are grateful to Xiaoxiang Yu for useful discussions. We also acknowledge the precious suggestions about this paper from Wei Liu, Zhichun Liu, Haisheng Fang and Run Hu. REFERENCES [1] M. Gad-el-Hak, The MEMS handbook (CRC press, 2001). [2] I. Sauciuc, R. Prasher, J.-Y. Chang, H. Erturk, G. Chrysler, C.-P. Chiu, and R. Mahajan, in ASME Conference Proceedings2005), pp. 353. [3] E. Pop, S. Sinha, and K. E. Goodson, Proceedings of the IEEE 94, 1587 (2006). [4] C. Silvestri, M. Riccio, R. H. Poelma, B. Morana, S. Vollebregt, F. Santagata, A. Irace, G. Q. Zhang, and P. M. Sarro, Nanoscale 8, 8266 (2016). [5] S. Lepri, Physics Reports 377, 1 (2003). [6] P. Kim, L. Shi, A. Majumdar, and P. L. McEuen, Phys Rev Lett 87, 215502 (2001). [7] M. Li, C. Li, J. Wang, X. Xiao, and Y. Yue, Appl. Phys. Lett. 106, 253108 (2015). [8] D. G. Cahill, K. Goodson, and A. Majumdar, J. Heat Transf. 124, 223 (2002). [9] J. Hone, M. Whitney, C. Piskoti, and A. Zettl, Physical Review B 59, R2514 (1999). [10] E. C. Garnett, M. L. Brongersma, Y. Cui, and M. D. McGehee, Annu. Rev. Mater. Res. 41, 269 (2011). [11] A. I. Hochbaum and P. Yang, Chem. Rev. 110, 527 (2010). [12] A. I. Boukai, Y. Bunimovich, J. Tahir-Kheli, J. K. Yu, W. A. Goddard, 3rd, and J. R. Heath, Nature 451, 168 (2008). [13] A. I. Hochbaum, R. Chen, R. D. Delgado, W. Liang, E. C. Garnett, M. Najarian, A. Majumdar, and P. Yang, Nature 451, 163 (2008). [14] Z. Cheng, L. Liu, S. Xu, M. Lu, and X. Wang, Sci Rep 5, 10718 (2015). [15] L. Lu, W. Yi, and D. Zhang, Rev. Sci. Instrum. 72, 2996 (2001). [16] D. G. Cahill, Review of Scientific Instruments 61, 802 (1990). [17] D. G. Cahill and R. O. Pohl, Phys. Rev. B. 35, 4067 (1987). [18] M. Soini, I. Zardo, E. Uccelli, S. Funk, G. Koblmüller, A. Fontcuberta i Morral, and G. Abstreiter, Appl. Phys. Lett. 97, 263107 (2010). [19] A. A. Balandin, Nature materials 10, 569 (2011). [20] Z. Yan, C. Jiang, T. R. Pope, C. F. Tsang, J. L. Stickney, P. Goli, J. Renteria, T. T. Salguero, and A. A. Balandin, J. Appl. Phys. 114, 204301 (2013). [21] X. J. Hu, A. Jain, and K. E. Goodson, Int. J. Therm. Sci. 47, 820 (2008). [22] H.-D. Wang, J.-H. Liu, X. Zhang, and Y. Song, Int. J. Heat Mass Transfer 70, 40 (2014). [23] H.-D. Wang, J.-H. Liu, Z.-Y. Guo, X. Zhang, R.-F. Zhang, F. Wei, and T.-Y. Li, Nanoscale Microscale Thermophys. Eng. 17, 349 (2013). [24] Z. L. Wang and D. W. Tang, Int. J. Therm. Sci. 64, 145 (2013). [25] W. X. C. Hou Y.L., ZhangC.W., Liu Z.G. , Journal of Engineering Thermophysics 28, 460 (2007). [26] L. Z. G. Guan N., Liang S.Q., Zhang C.W., Journal of Beijing University of Technology 35, 977 (2009). [27] J. P. Holman, Heat Transfer (McGraw-Hill, New York, NY 10020, 2010), 10th edn. [28] S. O. Kasap, Principles of Electronic Materials and Devices (McGraw-Hill Higher Education, New York, NY10020, 2002), p.^pp. 58-59. [29] J. Peirs, D. Reynaerts, and H. Van Brussel, in Proceedings of the IEEE International Conference on Robotics and Automation (IEEE, 1998)Leuven, Belgium., 1998), pp. 1516. [30] T. Fujii, in Proc. 7th Int. Heat Transf. Conf. Munich, Germany, (1982), pp. 311. [31] V. T. Morgan, in Advances in Heat Transfer, edited by T. F. Irvine, and J. P. Hartnett (Elsevier, 1975), pp. 199. [32] S. W. Churchill and H. H. S. Chu, International Journal of Heat and Mass Transfer 18, 1049 (1975). [33] A. S. L. Theodore L.Bergman, Frank P. Incropera, David P. Dewitt, Fundamentals of Heat and Mass Transfer (John Wiley & Sons, 2011), pp. 613. [34] V. T. Morgan, International Journal of Heat and Mass Transfer 16, 2023 (1973). [35] Z.-Y. Guo and Z.-X. Li, Int. J. Heat Mass Transfer 46, 149 (2003).
1903.03639	1	1903	2019-03-08T19:31:37	Enhancing the Electrical and Thermal Conductivities of Polymer Composites via Curvilinear Fibers: An Analytical Study	[ "physics.app-ph" ]	The new generation of manufacturing technologies such as e.g. additive manufacturing and automated fiber placement has enabled the development of material systems with desired functional and mechanical properties via particular designs of inhomogeneities and their mesostructural arrangement. Among these systems, particularly interesting are materials exhibiting \textbf{Curvilinear Transverse Isotropy} (CTI) in which the inhomogeneities take the form of continuous fibers following curvilinear paths designed to e.g. optimize the electric and thermal conductivity, and the mechanical performance of the system. In this context, the present work proposes a general framework for the exact, closed-form solution of electrostatic problems in materials featuring curvilinear transverse isotropy. First, the general equations for the fiber paths that optimize the electric conductivity are derived leveraging a proper conformal coordinate system. Then, the continuity equation for the curvilinear, transversely isotropic system is derived in terms of electrostatic potential. A general exact, closed-form expression for the electrostatic potential and electric field is derived and validated by Finite Element Analysis. Finally, potential avenues for the development of materials with superior electric conductivity and damage sensing capabilities are discussed.	physics.app-ph	physics	A&A Program in Structures William E. Boeing Department of Aeronautics and Astronautics University of Washington Seattle, Washington 98195, USA Enhancing the Electrical and Thermal Conductivities of Polymer Composites via Curvilinear Fibers: An Analytical Study Marco Salviato, Sean E. Phenisee INTERNAL REPORT No. 19-02/01E Submitted to Mathematics and Mechanics of Solids February 2019 Contents 1 Introduction 2 Preliminary remarks 2.1 Continuity equation in isotropic and homogeneous media . . . . . . . . . . . 2.2 General solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 General field line equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Electrostatics in media featuring curvilinear transverse isotropy 3.1 Constitutive behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 General solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Optimal fiber path for conductivity and calculation of the electrical resistance 4 Results and discussion 4.2 Deep hyperbolic notch in an infinite plate 4.1 Semicircular notch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Optimal fiber paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Electric potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.3 Current density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Optimal fiber paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Electric potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Current density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Sharp V-notch of finite depth . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Optimal fiber paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Electric potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.3 Current density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Rectangular notch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Optimal fiber paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2 Electric potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.3 Current density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Electrical resistance of CTI systems vs. traditional composites . . . . . . . . 4.6 Comments on the applicability of the theoretical framework . . . . . . . . . 5 Conclusion 2 3 3 5 7 9 9 10 11 13 16 17 18 19 19 20 20 21 21 22 24 24 25 25 26 26 27 28 29 30 1 Abstract The new generation of manufacturing technologies such as e.g. additive manu- facturing and automated fiber placement has enabled the development of material systems with desired functional and mechanical properties via particular designs of inhomogeneities and their mesostructural arrangement. Among these systems, partic- ularly interesting are materials exhibiting Curvilinear Transverse Isotropy (CTI) in which the inhomogeneities take the form of continuous fibers following curvilinear paths designed to e.g. optimize the electric and thermal conductivity, and the mechan- ical performance of the system. In this context, the present work proposes a general framework for the exact, closed-form solution of electrostatic problems in materials fea- turing curvilinear transverse isotropy. First, the general equations for the fiber paths that optimize the electric conductivity are derived leveraging a proper conformal coor- dinate system. Then, the continuity equation for the curvilinear, transversely isotropic system is derived in terms of electrostatic potential. A general exact, closed-form ex- pression for the electrostatic potential and electric field is derived and validated by Finite Element Analysis. Finally, potential avenues for the development of materials with superior electric conductivity and damage sensing capabilities are discussed. Keywords: Electrostatics, Curvilinear Transverse Isotropy, Additive Manufacturing, Conductive Composites 1 Introduction Emerging manufacturing technologies have provided engineers with an unprecedentedly broad design space. Laser lithography has enabled the development of lattice materials with sub-micron resolution [1, 2, 3, 4, 5][e.g.]. Multi-material additive manufacturing has allowed the design of multi-phase systems with novel physical and mechanical properties [6, 7, 8][e.g.]. Automated fiber deposition along curvilinear paths has introduced new ways to achieve damage tolerance of composite structures [9, 10, 11]. A common trait of all these technologies is that they have enabled the development of material systems with desired functional and mechanical properties via particular designs of inhomogeneities and their micro- and meso-structural arrangements [12]. Among these systems, particularly interesting are materials exhibiting Curvilinear Transverse Isotropy (CTI) in which the inhomogeneities take the form of continuous fibers following curvilinear paths. The morphology of the fiber curves is designed to e.g. optimize the electric and thermal conductivity, and the mechanical performance of the system [13, 14]. In this context, the present work investigates the use of curvilinear conductive fibers to increase the conductivity of insulating materials such as e.g. polymer or concrete. Achieving 2 Fig. 1: (a) a notched 2D body Ω ∪ ∂Ω subjected to an electric potential and (b) example of a conformal mapping used in this work (equation (49)). a sufficient electrical conductivity is quintessential to enable the use of resistivity measure- ments for the detection of damage initiation and evolution close to stress raisers such as sharp notches, corners and cracks. Towards this goal, several researchers investigated the use of e.g. nanoparticles or graphene to increase the conductivity [15, 16, 17][e.g.] but the application of curvilinear fibers has never been explored. This manuscript takes a first step in this direction by proposing a general framework for the exact, closed-form solution of elec- trostatic problems in materials featuring curvilinear transverse isotropy. First, the general equations for the fiber paths optimizing the electrical conductivity are derived leveraging a proper conformal coordinate system. Then, the continuity equation for the curvilinear, transversely isotropic system is derived in terms of electrostatic potential. A general ex- act, closed-form expression for the electrostatic potential and electric field is derived and validated by Finite Element Analysis. 2 Preliminary remarks 2.1 Continuity equation in isotropic and homogeneous media Let us consider a 2D body Ω ∪ ∂Ω made of a homogenous and isotropic material obeying the theory of the linear electrostatics and suppose that the body is subjected to a potential difference as shown in Figure 1a. With reference to the Cartesian coordinate system (x,y,z) represented in Figure 1, the equation of continuity for the electrostatic field, assuming the 3 absence of sources, states: ∂Jx ∂x ∂Jy ∂y = −∂ρ ∂t + (1) where Ji (i = x, y) = components of current density in x− and y− directions, and ρ = electric charge density. The current density is related to the electric field, E, through Ohm's law [18] Ei = k−1Ji (i = x, y) where k = kx = ky represents the electrical conductivity of the material, treated as homogeneous and isotropic. Further, the conservativeness of the electric field leads to the following link to the electric potential ϕ: Ex = −∂ϕ/∂x and Ey = −∂ϕ/∂y. Substituting the foregoing expressions into equation 1 and focusing on the steady-state solution of the problem, one gets the following two-dimensional Laplace equation for the electric potential: (2) with ∇2 = ∂2/∂x2 + ∂2/∂y2 being the laplacian operator. The solution of equation (2) in a given two-dimensional domain Ω ∪ ∂Ω subjected to Dirichlet or von Neumann conditions on the boundary ∂Ω, allows the full characterization of the electric potential and the related current density. Towards this goal, it is often convenient to employ a transformation of coor- dinates to simplify the application of the boundary conditions when the shape of the domain Ω is complex. In this work, only conformal mappings are considered, i.e. transformations that can be described by means of a complex analytic function [19, 20, e.g.] (an example of conformal mapping used in the present work is provided in Figure 1b). In this case, the change of coordinates z = z (ξ) with z = x + iy and ξ = u + iv is always bijective, satisfies the Cauchy-Riemann (C -- R) conditions ∂u/∂x = ∂v/∂y, ∂u/∂y = −∂v/∂x in Ω, and trans- forms the governing equation to an expression that is formally identical to the Cartesian case [21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, e.g.]: ∇2ϕ = 0 ∇2ϕ = ∂2ϕ ∂u2 + ∂2ϕ ∂v2 = 0 (3) with an easier description of the boundary of the problem. In this work, the following conformal mappings are considered in which the condition u = u0 or v = v0 (with u0, v0 being real constants) describes the boundary ∂Ωn,0 [28, 29]. Here ∂Ωn,0 refers to the portion of the boundary where Jn = J · n = 0. In the introduced coordinate system, the electric field can be calculated as follows: E = −∇ϕ = − 1 hu ∂ϕ ∂u eu − 1 hv ∂ϕ ∂v ev (4) 4 Fig. 2: Schematic representation of the field lines and the family of curves γ (s, uc, vc) in the (a) transformed domain Ω(cid:48) and (b) Cartesian domain Ω. The angle ρ0 between the tangent unit vector of the field lines, in this case eu, and of γ (s, uc, vc), t, is preserved by the conformal transformation. hu =(cid:112)(∂x/∂u)2 + (∂y/∂u)2 and hv =(cid:112)(∂x/∂v)2 + (∂y/∂v)2 are the related metric coeffi- where eu, ev represent the orthonormal basis of the curvilinear system. cients. 2.2 General solution Let us consider the domain Ω ∪ ∂Ω and a conformal map z = z (ξ) with z = x + iy and ξ = u + iv such that ∂Ωn,0 is described by the condition u = u0. The constant u0 is taken as a positive number, so that the domain of integration belongs to the right half plane of the (u, v) space. Under this condition, u is always positive whilst v can take any value. The body is subjected to a uniform current density or potential in the remaining portion of the domain, Ωd as shown in Figure 1. Assuming separation of variables in curvilinear coordinates, ϕ (u, v) = f (u) g (v), and substituting into equation (3), one gets: or, equivalently: f(cid:48)(cid:48) (u) g (v) + f (u) g(cid:48)(cid:48) (v) = 0 f(cid:48)(cid:48) (u) f (u) = −g(cid:48)(cid:48) (v) g (v) = λ2 (5) (6) with λ being a real constant. Introducing the boundary conditions, one obtains the following 5 system of differential equations: g(cid:48)(cid:48) (v) + λ2g (v) = 0 f(cid:48)(cid:48) (u) − λ2f (u) = 0 f(cid:48) (u0) = 0 f(cid:48) (u) g(cid:48) (u) hu < ∞ for u, v → ∞ < ∞ for u, v → ∞ hv (7a) (7b) (7c) (7d) (7e) where equation (7c) states that the component of the current density normal to the surface is zero whereas equations (7d, 7e) express the condition for bounded, remote current. The case λ2 (cid:54)= 0 can be disregarded since it only provides trivial solutions. In contrast, under the condition λ2 = 0, the general solution is: ϕ (u, v) = (A + Bu) (C + Dv) which simplifies further after applying the boundary conditions: ϕ (u, v) = C1 + ζv (8) (9) where C1 represents a reference potential and can be taken as zero whereas ζ is a real constant that can be found imposing the remote potential or current density. Then, recalling equation (4), the curvilinear components of the current density can be written as follows: Ju = 0, Jv = ψ (cid:107)z(cid:48) (ξ)(cid:107) (10) where (cid:107)z(cid:48) (ξ)(cid:107) = h is the magnitude of the first derivative of the conformal map and ψ = −ζk is a constant to determine. Leveraging the relation between the electric field, the electric potential, and the current density along with the C -- R conditions on the curvilinear coordinates one can write the Cartesian components of the current density as Jx = ψ∂v/∂x It is worth noting that, since ξ(cid:48) = dξ/dz = ∂u/∂x + i∂v/∂x, the and Jy = ψ∂u/∂x. Cartesian components of the current density represent the real and imaginary parts of the first derivative of the conformal map used to describe the domain, ξ = ξ (z) [28, 29, 33]. Accordingly, the current density can be written as: Jy + iJx = ψ dξ (z) dz 6 (11) This result provides an explicit link between the domain geometry and the current density field. Once the proper conformal map is determined, the problem of calculating the electric potential and the related current density can be solved directly. It is worth mentioning that in case ∂Ωn,0 is described by the condition v = v0, a similar approach can be used to show that ϕ (u, v) = C1 + C2u. In such a case, the current density in curvilinear components can be calculated as follows: Ju = Jv = 0 ψ (cid:107)z(cid:48) (ξ)(cid:107), (12) (13) whereas the Cartesian components take the following form [28, 29, 33]: Jx − iJy = ψ dξ (z) dz 2.3 General field line equations The theoretical framework described in the foregoing sections can be used to calculate the general equations of the field lines. This is a particularly important information since, as it will be shown in the following sections, the field lines also represent the fiber paths optimizing the electrical conductivity of the system. Leveraging the curvilinear, orthogonal coordinate system and considering the case in which u = u0 describes ∂Ωn,0, this task is particularly easy as the paths are described by the condition u = const. In fact, being the gradient of the electric potential, E is orthogonal to the equipotential lines which, as shown in equation (9), are described by the condition v = const. For the sake of generality, however, let us consider the family of paths defined by the condition t · ev = cos ρ0 with t = tangent unit vector of the path, ev = tangent unit vector of the field lines when u = u0 describes the insulated portion of the boundary and ρ0 being a constant angle as represented in Figure 2. Then, the fiber paths can be calculated as the integral curve, γ (s), of the tangent vector t = cos ρ0ev + sin ρ0eu:  ∂x ∂u ∂y ∂u u(cid:48) + u(cid:48) + ∂x ∂v ∂y ∂v v(cid:48) = sin ρ0 v(cid:48) = sin ρ0 ∂x ∂u ∂y ∂u + cos ρ0 + cos ρ0 ∂x ∂v ∂y ∂v (14a) (14b) where u(cid:48) = du/ds and v(cid:48) = dv/ds respectively. The foregoing system was obtained noting that (a) eu = 1/hu (∂x/∂ui + ∂y/∂uj) and ev = 1/hv (∂x/∂vi + ∂y/∂vj) with i and j being the Cartesian unit vectors in x− and y−direction, (b) hu = hv = h thanks to the C -- R conditions on the conformal transformation and (c) hev is parallel to ev. The solution of the foregoing system can be obtained noting that u(cid:48) = sin ρ0 and v(cid:48) = cos ρ0, which leads to the 7 Fig. 3: (a) Infinitesimal material element featuring curvilinear orthotropy in (eu, ev, k) and the related current densities; (b) schematic representation of the microstructure as cut by the plane π : π ⊥ t. As can be noted, an infinite number of symmetry planes Hi exist supporting the assumption of curvilinear transverse isotropy in (t, m, k). following solution: γ (s, uc, vc) : u (s) = uc + sin ρ0s, v (s) = vc + cos ρ0s (15) where s ∈ [0,∞), and uc = u (0), vc = v (0) respectively. Then, substitution of equation (15) into the conformal map, leads to the description of the curves in Cartesian coordinates: z (s) = (z ◦ ξ) (s). A comparison between the field lines and the family of curves γ (s, uc, vc) can be found in Figure 2. It is worth noting here that a similar procedure can be used when v = v0 describes ∂Ωn,0, which is common when the conformal mapping is constructed leveraging a Schwarz- Christoffel transformation [34]. In such a case, the field lines are described by the condition v = const and eu represents the unit tangent vector. The integral curves characterized by the unit vector t t · eu = cos ρ0 can be calculated as follows: γ (s, uc, vc) : u (s) = uc + cos ρ0s, v (s) = vc + sin ρ0s (16) Finally, it should be noted that when ρ0 = 0, both equations (15) and (16) reduce to the equations for the field lines. 8 3 Electrostatics in media featuring curvilinear trans- verse isotropy 3.1 Constitutive behavior Let us consider a material reinforced by fibers following a family of curvilinear paths γ (s, uc, vc) as shown in Figure 2. As discussed in Section 2, the paths are constructed such that their tangent, t is rotated of a constant angle ρ0 with respect to field lines. In the particular case in which ρ0 = 0, the fiber paths coincide with the field lines of the problem. For each material point, a curvilinear orthonormal basis (t, m, k) can be defined such that m × t = k with k being the unit vector of the z−direction. Now, considering the material microstructure, it is easy to note that any plane H : t ∈ H is a symmetry plane (Figure 3b). Accordingly, in each material point the transverse plane π : π ⊥ t is a plane of isotropy and the material exhibits curvilinear transverse isotropy [35, 36]. Assuming that each material point is reinforced by a sufficiently large number of curvilinear fibers, the material can be treated as homogeneous. Then, the anisotropic Ohm's equation can be written in matrix form as follows: (cid:40) (cid:41) (cid:34) (cid:35)(cid:40) (cid:41) Jt Jm = ktt 0 0 kmm Et Em (17) where Jt, Jm, Et, Em represent the components of the current density and electric field re- spectively in the orthonormal basis (t, m, k). The electric conductivities ktt and kmm in the curvilinear system can be measured experimentally or estimated by means of micromechan- ics [37, e.g.] and depend on the electric conductivities of the fibers and the matrix as well as the volume fraction of fibers. Assuming that the electrical conductivity of the fibers is significantly larger than the matrix and considering that fibers and matrix are in parallel coupling along t and in series coupling along m, one can conclude that generally ktt ≥ kmm. Although the orthonormal basis (t, m, k) guarantees the simplest description of the con- stitutive behavior, it is convenient to recast equation (17) in the curvilinear coordinate system (eu, ev, k) which enables the simplest application of the boundary conditions without affect- ing the continuity equation significantly. Considering the case in which u = u0 provides the description of ∂Ωn,0 and recalling that t = cos ρ0ev + sin ρ0eu and m = − sin ρ0ev + cos ρ0eu, the constitutive behavior can be written as follows: (cid:40) (cid:41) (cid:34) Jv Ju = kvv (ρ0, ktt, kmm) kuv (ρ0, ktt, kmm) kuv (ρ0, ktt, kmm) kuu (ρ0, ktt, kmm) Ev Eu (18) (cid:35)(cid:40) (cid:41) 9 where:  kvv = cos2 (ρ0) ktt + sin2 (ρ0) kmm kuu = sin2 (ρ0) ktt + cos2 (ρ0) kmm kuv = (ktt − kmm) cos ρ0 sin ρ0 (19a) (19b) (19c) It is worth mentioning that equations (19a), (19b) and (19c) can be obtained noting that K(u,v) = T(cid:62)K(t,m)T with T being the rotation matrix linking (t, m, k) to (eu, ev, k) and K(t,m), K(u,v) being the conductivity matrices in the two orthonormal bases respectively. In case that v = v0 provides the description of ∂Ωn,0, eu becomes the unit tangent vector of the field lines and ρ0 represents the angle between t and eu. It can be shown that the constitutive behavior can be obtained from equation (18) simply by switching the diagonal terms in K(u,v). 3.2 Continuity equation To solve for the electric potential, it is convenient to write the continuity equation for the material system schematically represented in Figure 3 in curvilinear coordinates. Let us consider an infinitesimal curvilinear element and the current densities represented in Figure 3. The energy balance for a infinitesimal time increment dt, excluding the presence of sources, reads: (cid:18) Ju + (cid:18) ∂Ju ∂u du Jv + ∂Jv ∂v (cid:19)(cid:18) (cid:19)(cid:18) Juhudvdt − + Jvhvdudt − (cid:19) hv + ∂hv ∂u du ∂hu ∂v dvdt+ (cid:19) (20) dv hu + dv dudt = 0 which, neglecting the infinitesimal terms of higher order and recalling that hu = hv = h, dt > 0, dudv > 0, can be simplified as follows: Ju ∂h ∂u + h ∂Ju ∂u + Jv ∂h ∂v + h ∂Jv ∂v = 0 (21) Finally, introducing the constitutive equation, equation (18), and the relation between the electric field and electric potential in curvilinear coordinates, equation (4), the continuity equation can be written as follows: kvv ∂2ϕ ∂v2 + 2kuv ∂2ϕ ∂u∂v + kuu ∂2ϕ ∂u2 = 0 (22) 10 3.3 General solution First, let us consider the case for which u = u0 provides the description of ∂Ωn,0. We seek solutions of equation (22) in the form: ϕ (u, v) = f (u + λv) (23) where λ is a constant and can be complex. Substituting equation (23) into equation (22), one gets: (cid:0)λ2kvv + 2λkuv + kuu (cid:1) f(cid:48)(cid:48) (u + λv) = 0 (24) (cid:54)= 0, then equation (24) provides a characteristic equation for λ: If one assumes that f(cid:48)(cid:48) λ2kvv + 2λkuv + kuu = 0, which is generally satisfied by two complex conjugate solutions. However, thanks to the curvilinear system adopted in this work and the particular arrange- ment of the fibers, it suffices to consider the case in which f(cid:48)(cid:48) = 0 and the electric potential is written as ϕ (u, v) = (u + λv) ζ with ζ = real constant. This function satisfies the governing equation and the boundary condition at u = u0. In fact, the condition J · n = Ju = 0 in ∂Ωn,0 reads: (25) which is satisfied imposing λ = −kuu/kuv. Accordingly, the following general expression holds for the electric potential: λζ = 0 for u = u0 ζ + kuu h kuv h ϕ (u, v) = ζ v (26) (cid:18) ψ which, following equation (4), leads to the following current densities in curvilinear compo- nents: −kuv h kuukvv kuvh + ζ = Jv = Ju = 0, (cid:107)z(cid:48) (ξ)(cid:107) (27) Here, the magnitude of the first derivative of the transformation of coordinates (cid:107)z(cid:48) (ξ)(cid:107) is introduced, χ = (kuukvv − k2 uv) /kuv is a real constant that depends on the electric conductiv- ities and ψ = ζχ. It is worth mentioning here that the foregoing curvilinear components of the current density are strikingly identical to the isotropic case whereas the electric potentials and electric fields are different. On the other hand, the electric potential and electric fields take the same expression as the isotropic case when (a) the fiber are aligned with (ρ0 = 0) or orthogonal to (ρ0 = π/2) the field lines or (b) ktt = kmm. In such cases, the coupling term kuv goes to zero and the electric potential takes exactly the same form as in the isotropic case. (cid:18) u − kuu kuv (cid:19) (cid:19) 11 Fig. 4: Electrical resistance of plates made of CTI Vs isotropic materials: (a) effect of the volume fraction of conductive fibers; (b) effect of the conductivity of the matrix phase. Equations (41)a,b were used to estimate the conductivities in the orthonormal basis (t, m, k). The physical properties of the constituents refer to a typical system composed by an epoxy resin reinforced by carbon fibers. The Cartesian components can be found from the curvilinear ones introducing a rotation: Jx + iJy = exp (iα) (Ju + iJv) with α being the angle between the unit vectors i and eu. Recalling that exp (iα) = cos α+i sin α =(cid:112)z(cid:48) (ξ) /¯z(cid:48) (ξ), where the bar indicates the complex conjugate, the Cartesian components can be written as follows: Jx + iJy = iψ z(cid:48) (ξ) (cid:107)z(cid:48) (ξ)(cid:107)2 (28) As expected, this expression differs from the isotropic case even if the magnitude of the current density remains unchanged. This is because the Cartesian components depend on the anisotropic electric conductivities and α, the latter changing continuously in Ω. The case in which ∂Ωn,0 is described by the condition v = v0 can be solved by means of a similar approach. In such a case, the electric potential is calculated as ϕ (u, v) = ζ (v + λu) where the constant λ can be calculated imposing the boundary condition Jv = 0 when v = v0: kvv h ζ + kuv h ζλ = 0 for v = v0 (29) According to the foregoing equation, λ = −kvv/kuv and the electric potential can be written as follows: u (30) (cid:19) (cid:18) v − kvv kuv ϕ (u, v) = ζ 12 Finally, the curvilinear components of the current density read: (cid:18) kuv h Ju = (cid:19) − kuukvv kuvh ζ = − ψ (cid:107)z(cid:48) (ξ)(cid:107), Jv = 0 (31) where ψ was defined before. Then, the Cartesian components of the current density can be found by means of the following expression: Ju + iJv = − ψz(cid:48) (ξ) (cid:107)z(cid:48) (ξ)(cid:107)2 (32) Finally, it is worth noting that in case the fiber paths are aligned with the field lines (ρ0 = 0), the coupling term kuv goes to zero. Accordingly, the boundary conditions reduce to the same as the isotropic case and the governing equation, equation (22), simplifies to: kvv ∂2ϕ ∂v2 + kuu ∂2ϕ ∂u2 = 0 (33) Notwithstanding the presence of the terms kuu and kvv, this equation has the same solution as the isotropic case: for u = u0 describing ∂Ωn,0 and: ϕ (u, v) = ζv ϕ (u, v) = ζu (34) (35) for v = v0 describing ∂Ωn,0. Accordingly, the current density can be calculated using equa- tions (10), (11), (12) and (13) provided that the proper conductivities kvv and kuu are used instead of k in the calculation of ψ. 3.4 Optimal fiber path for conductivity and calculation of the elec- trical resistance Leveraging the foregoing solutions, it is possible to investigate the reduction of electrical resistance enabled by the curvilinear fibers compared to the isotropic case. Towards this goal, let us focus on the case in which ∀u ∈ ∂Ωn,0, u = u0 and consider an infinitesimal flux tube. As was shown in the foregoing sections, the flux tube is defined by curves u = const either when the material is isotropic or when it exhibits curvilinear transverse isotropy since Jv is the only nonzero curvilinear component. Assuming that the same potential difference ∆ϕ0 is applied at the remote extremities of the flux tube, the electric current can be expressed as dI = Jvhdudz where the current density can be calculated by means of the proposed formulation and it is considered uniform through the thickness. Finally, the 13 electrical resistance of the domain can be obtained by integration over all the flux tubes, leading to the following expression: ∆ϕ0 (cid:90) T I (cid:90) uL where R(ISO) = I = v (u) hdudz = ψ(ISO)T (uL − u0) J∞ for the isotropic case and: 0 u0 R(CT I) = I = where ∆ϕ0 (cid:90) T I (cid:90) uL 0 u0 v (u) hdudz = ψ(CT I)T (uL − u0) J∞ (36) (37) for the case of curvilinear transverse isotropy. In the foregoing expressions, T = plate thick- ness and ψ(ISO) and ψ(CT I) are real constants defined in equations (10) and (27) respectively. Thanks to equations (36) and (37), one can easily calculate the resistance. A particularly important conclusion that can be drawn from the foregoing results is that, since the field lines for the isotropic and transversely isotropic case are identical, the resistivity differs only for a multiplying constant dependent on the fiber angle, ρ0, and anisotropic electric conductivities ktt and kmm. This allows a very simple description of the ratio between the electrical resistance of the transversely isotropic body and the isotropic one: Se (ρ0, ktt, kmm) = R(CT I)/R(ISO). In fact, assuming that the same potential differential ∆ϕ0 is applied to the two bodies then one gets ∆ϕ0 = 2ζ (ISO)v0 = 2ζ (CT I)v0kuu/kuv with v0 representing the value of the curvilinear coordinate v where the potential is imposed. Accordingly, ζ (CT I) = ζ (ISO)kuv/kuu and the constant ψ can be written as follows: ψ(CT I) = ζ (ISO) kuukvv − k2 uv kuu = ψ(ISO) kuukvv − k2 uv kuuk(ISO) (38) Leveraging equations (36), (37) and, (38) and using its definition, Se can finally be calculated as follows: Se (ρ0, ktt, kmm) = R(CT I) R(ISO) = kuuk(ISO) kuukvv − k2 (cid:20)cos2 (ρ0) + ktt = k(ISO) (cid:21) uv sin2 (ρ0) kmm (39) where k(ISO) represents the electrical conductivity of the isotropic material. It is worth men- tioning that equation (39) is very useful for design since it provides a very simple expression to estimate the electrical resistance reduction induced by the addition of the conductive 14 Fig. 5: Schematic representation of a patch of 4−node, isoparametric, quadrilateral elements with material orientation assigned to each integration point by the user subroutine ORIENT in ABAQUS/standard [38]. As can be noted, the curves describing the fiber paths are only C 0 (Ω). fibers. Further, it simplifies significantly the computational modeling of materials featuring curvilinear transverse isotropy, in case a numerical simulation is preferred over a closed-form solution (which may be the case when the geometry cannot be described by a relatively simple conformal map). In fact, if the angle ρ0 as well as the conductivities ktt and kmm are known, one can analyze the isotropic medium using standard Finite Element Analysis (FEA) and then use equation (39) to calculate the resistance for the curvilinear, transversely isotropic case. Even the current densities can be calculated from the isotropic case simply leveraging the link between the isotropic constants and ψ provided in equation (38) and by substituting into equation (27). The advantage is that, this way, the fibers do not need to be considered explicitly and the analysis can be conducted in any standard FE software without the need for specialized subroutines. equation (39) enables the identification of the fiber orientation leading to the minimum Se, i.e. the maximum reduction of electrical resistance. Considering that the fibers are conductive and that they act in parallel coupling with the matrix along t and in series coupling along m, it is reasonable to assume that ktt ≥ kmm ≥ k(ISO) > 0. In the particular, unlikely case in which ktt = kmm = k, Se is minimized by the condition that k is maximum, independently on the value taken by ρ0 ∈ [0, π). When ktt (cid:54)= kmm, then the configuration providing the minimum electrical resistance is always the one in which the fibers are aligned with the field lines, i.e. ρ0 = 0. In such a case: Se (0, ktt, kmm) = k(ISO) ktt (40) which depends only on the ratio between the conductivity of the matrix and the conductivity in the fiber direction. The latter conductivity depends on the volume fraction of fibers, Vf , 15 the conductivity of the fibers kf and of the matrix kmat = k(ISO). Considering the coupling between fibers and matrix, a first order approximation of ktt and kmm can be obtained as follows: ktt = Vf kf + (1 − Vf ) kmat, kmm = kf kmat [Vf km + (1 − Vf ) kf ] −1 (41) In case that a more refined estimate is needed, several micromechanical models can be found in the literature [39, 40, 37, 41, 42, e.g.]. A summary of the foregoing conclusions is provided in Figure 4a which shows the effect of the curvilinear fiber angle on the electrical resistance for various volume fractions of fibers assuming a fiber conductivity kf = 4 × 105 S/m and a matrix conductivity k(ISO) = 10−16 S/m. These values correspond to a typical epoxy system reinforced by carbon fibers. As can be noted, the electrical resistance decreases dramatically decreasing the angle ρ0, regardless of the volume fraction. On the other hand, a higher amount of conductive fibers leads to lower electrical resistance for a given angle. The importance of the fiber angle is also confirmed by Figure 4b which reports the electrical resistance against the fiber angle for various conductivities of the matrix, k(ISO). As can be noted, the minimum resistance is always achieved when ρ0 = 0 and the fibers and the matrix are in parallel coupling along the field lines, regardless of the conductivity of the matrix. In contrast, when the conductivity of the matrix is equal to the one of the fibers, the electrical resistance becomes independent of the fiber angle. Finally, it is worth mentioning that the case in which ∀v ∈ ∂Ωn,0, v = v0 provides the same expression for Se as reported in equation (39). 4 Results and discussion In this section, the theoretical framework is applied to several examples of notched plates. In each case, the explicit equations of the the fiber paths minimizing the electrical resis- tance are provided along with the expressions for electric potential and the current density components. The results are validated by comparing the theoretical predictions to a large bulk of Finite Element electrostatic analyses conducted in ABAQUS 6.14 [38]. The electric potential and current density were investigated leveraging a 2D mesh of DC2D8, 8−node isoparametric quadrilateral elements with anisotropic linear electrostatic constitutive laws by means of the implicit solver ABAQUS/standard [38]. The user subroutine ORIENT [38] was used to prescribe a material orientation following the assigned curvilinear paths. It is worth mentioning that ORIENT assigns a uniform fiber orientation to each element inte- gration point. Accordingly, the description of the fiber paths obtainable by this method, 16 Fig. 6: Schematic representation of the conformal map provided by equation (42), trans- forming the right half-plane into a semi-infinite plate featuring a semi-circular notch of radius R γ(F EM ) (s, uc, vc), is only C 0 (Ω) (see Figure 5). For this reason, particular care was devoted in verifying the numerical convergence of all the cases investigated, particularly the ones involving sharp corners leading to singular current density fields. A simulation was assumed to converge when further mesh refinements would lead to changes in the maximum current density values lower than 0.01%. This convergence criterion led to meshes featuring ele- ments typically one order of magnitude smaller compared to the isotropic case due to the poor geometrical description of the fibers. To remedy this issue, a new Finite Element based on Isogeometric Analysis (IGA) [43] enabling a smooth description of the fibers is under development by the authors. Its description is beyond the scope of the present work. 4.1 Semicircular notch To exemplify the application of the foregoing theoretical framework, let us consider the following conformal transformation: (cid:112) (cid:19) (cid:18) ξ + 1 2 2 z (ξ) = R ξ2 + 4 (42) This mapping transforms the right curvilinear half-plane, u ≥ 0, into an semi-infinite plate featuring a semi-circular notch of radius R (Figure 6). The boundary, which is considered isolated, J · n = 0, is described by the condition u = 0 whereas the bisector of the notch is described by the condition v = 0. It is worth mentioning here that, even if the foregoing map considers a semi-infinite medium, it can also be applied to finite bodies provided that the domain is large enough so that the effects of the boundaries are negligible. Further, this transformation can always be used to describe the electric potential and current density sufficiently close to the notch, no matter the size of the domain. In such a case, however, the 17 Fig. 7: Solution of the electrostatic problem for a semi-infinite body featuring curvilinear transverse isotropy weakened by semi-circular notch: (a) fiber paths of minimum electrical resistance; (b) electric potential field; (c) magnitude of the electric current density. The notch radius is R = 1 mm. constant ψ required for the calculation of the current density components (see equation 27) must be estimated by e.g. FEA. Alternatively, in case of finite bodies, one can always find the exact solution of equation (22) by assuming f(cid:48)(cid:48), λ (cid:54)= 0 and leveraging a Laurent series. However, this latter approach would prevent the achievement of a closed-form solution. 4.1.1 Optimal fiber paths In Section 3, it was shown that the fiber paths, γ (s, uc, vc), providing the minimum elec- trical resistance are aligned with the field lines. Leveraging equations (15a,b) and (42) and imposing ρ0 = 0, the paths can be calculated easily as follows: (cid:110)(cid:2)u2 c − (s + vc)2(cid:3)2 (cid:110)(cid:2)u2 c − (s + vc)2(cid:3)2  x (s, uc, vc) = c − (s + vc)2 + 2iuc (s + vc)(cid:3). A schematic representation of the forego- c (s + vc)2(cid:111) 1 + 4u2 c (s + vc)2(cid:111) 1 4 cos (43a) y (s, uc, vc) = Ruc 2 R 2 4 sin η 2 + R 2 η 2 (43b) γ (s, uc, vc) : where η = Arg(cid:2)u2 + 4u2 ing fiber paths for R = 1 mm can be found in Figure 7a. 18 4.1.2 Electric potential The electric potential can be obtained easily by means of equations (34) and (42). In fact, noting that ξ (z) = z/R−R/z, the curvilinear coordinate v can be determined easily, leading to the following result in Cartesian coordinates: ϕ (x, y) = ζ sin [Arg (x + iy)] (44) Alternatively, equation (44) can be written in polar coordinates as follows: (cid:32)(cid:112)x2 + y2 + R (cid:33) R(cid:112)x2 + y2 (cid:19) (cid:18) r + R R r ϕ (r, θ) = ζ sin θ (45) The electric potential for R = 1 and ζ = −1 V/mm is provided in Figure 7b. 4.1.3 Current density Finally, the current density components can be calculated by means of equation (11): (cid:34) (cid:19)2(cid:35) (cid:18) R Jy + iJx = ψ R 1 + z which can be used to calculate the Cartesian components of the current density in polar coordinates: (cid:35) (cid:34) (cid:19)2 (cid:18)R (cid:19)2 (cid:18) R r Jy (r, θ) = ψ R 1 + cos 2θ Jx (r, θ) = − ψ R r sin 2θ (46) (47) (48) Figure 7c shows a contour plot of the magnitude of the current density for R = 1 and ψ = 1 A/mm which corresponds to the case of a unitary remote positive current, i.e. J∞ y = 1 A/mm2. Figure 8a shows a comparison between the current density components along the boundary (u = 0) calculated by means of equations (47) and (48) and FEA. In the numerical analysis, a plate of height H = 100 mm, width D = 100 mm and notch radius R = 1 mm was considered. It can be seen that theoretical and numerical results are in perfect agreement, with differences that are always lower than 0.2%. It should also be noted that, as expected, the notch leads to a concentration of current density. Following equations (47) and (48), the maximum current occurs at the notch tip with the maximum concentration factor K = J max/J∞ y = 2. 19 Fig. 8: Comparison between the theoretical solution and FEA: (a) Cartesian components of the current density along the boundary of a plate featuring a semicircular notch; (b) nor- malized Cartesian components of the current density along the boundary of a plate featuring a hyperbolic notch of opening 2α = 90◦ and tip radius ρ = 95 µm. 4.2 Deep hyperbolic notch in an infinite plate Let us consider the mapping represented in Figure 1b [44, 45]: z (ξ) = ξq (49) where u = u0 represents the condition for the boundary of a deep hyperbolic notch, q is a parameter linked to the notch opening angle, 2α, by the expression q = (2π − 2α) /π with q ∈ [1, 2]. Notwithstanding its simplicity, the conformal map described by equation (49) is rather general since it can describe infinite notches of any opening angle and radius of 0q/(q − 1) curvature at the tip. In fact, the radius can be controlled by the expression ρ = uq which shows that the particular case u0 = 0 describes a sharp notch of opening 2α. The case of a crack, i.e. ρ = 0 and 2α = 0, can be obtained imposing u0 = 0 and q = 2 [44, 45]. When q = 3/2 and u0 = 0.1, the opening angle is 90◦ with a tip radius of 95 µm as shown in Figure 9. 4.2.1 Optimal fiber paths The field lines close to the notch can be calculated easily leveraging equation (15) which, using the composed function z (s) = (z ◦ ξ) (s), leads to the following expression in Cartesian coordinates: 20 Eq. (48)Eq. (47)Eq. (55)Eq. (54) x (s, uc, vc) =(cid:2)u2 y (s, uc, vc) =(cid:2)u2 c + (vc + s)2(cid:3)q/2 c + (vc + s)2(cid:3)q/2 γ (s, uc, vc) : cos{q Arg [uc + i (vc + s)]} sin{q Arg [uc + i (vc + s)]} (50a) (50b) Here s ∈ [0,∞), uc = u (0), vc = v (0) and ρ0 was set to zero. A schematic representation of the field lines, equations (50a) and (50b), is provided in Figure 9a. These curves represent the fiber paths minimizing the electrical resistance of the body. 4.2.2 Electric potential Once the conformal mapping describing the domain is identified, the electric potential can be calculated easily in Cartesian or polar coordinates leveraging equation (9): ϕ (x, y) = ζ(cid:0)x2 + y2(cid:1)1/2q sin (cid:21) Arg (x + iy) (cid:20)1 (cid:18) θ q q (cid:19) (51) (52) (54) (55) and ϕ (x, y) = ζr1/q sin It is worth mentioning that, in this particular case, the electric potential can be calculated but for the multiplying constant ζ since both the notch and the domain are infinite. This constant can be calculated analytically or numerically by analyzing a finite plate with a deep hyperbolic notch and a sufficiently large ligament. Figure 9b shows the distribution of electric potential for ζ = −1 V/mm1/q. 4.2.3 Current density The current density can be calculated from the first derivative of the Neuber's map, equa- tion(49), leading to the following equation: Jy + iJx = z1/q−1 ψ q (53) which can be used to calculate the Cartesian components of the current density in polar coordinates: (cid:20)(cid:18)1 (cid:20)(cid:18)1 q q (cid:19) (cid:19) (cid:21) (cid:21) θ θ − 1 − 1 Jy (r, θ) = r1/q−1 cos ψ q Jx (r, θ) = r1/q−1 sin ψ q 21 Fig. 9: Solution of the electrostatic problem for an infinite body featuring curvilinear trans- verse isotropy weakened by a deep hyperbolic notch: (a) fiber path of minimum electrical resistance; (b) electric potential field; (c) magnitude of the electric current density. The notch is described by z (ξ) = ξq with q = 3/2, u0 = 0.1 corresponding to an opening angle 2α = 90◦ and a notch tip radius ρ = 95 µm. Figure 9c shows the distribution of the magnitude of the current density for ψ/q = 1 A/mm1/q+1 whereas Figure 8b shows a comparison between the normalized current den- sity components along the notch boundary (u = 0.1) calculated by means of equations (54) and (55) and FEA. In the numerical analysis, a plate of height H = 1000 mm, width D = 800 mm, ligament length d = 700 mm, notch opening 2α = 90◦ and tip radius ρ = 95 µm was considered. It can be seen that theoretical and numerical results are in perfect agreement, with differences that are always lower than 0.25%. 4.3 Sharp V-notch of finite depth The link between the electrostatic solution and the conformal map used to describe the domain makes methodologies such as the Schwarz -- Christoffel transformation [34], that allows the geometrical construction of the map for any polygonal domain, particularly convenient. In general terms, the Schwarz -- Christoffel Mapping (SCM) can be written as [34]: z (ξ) = z (ξ0) + C dw (w − aj)1−αj /π (56) (cid:90) ξ n(cid:89) ξ0 j=1 22 Fig. 10: Schematic representation of the conformal map provided by equation (57), trans- forming the half-plane, v ≥ 0, into a semi-infinite plate featuring a finite V-notch of depth a and opening 2α. where C is a constant and z (ξ) maps the real axis, v (x, y) = 0, to the edges of a polygon with n sides and interior angles αj with aj being the vertices of the polygon in the Argand plane. Let us consider the case of a finite, sharp V-notch in an infinite solid subjected to a remote current J∞ or a remote potential differential ∆ϕ0. Thanks to the SCM, the conformal map describing the notch profile and the corresponding first derivative are: z(cid:48) (ξ) = A ξ1−2α/π (ξ2 − 1)1/2−α/π , z (ξ) = B + z(cid:48) (w) dw (57) (cid:90) ξ where A and B are complex constants. Solving the integral in the second of equation (57) and imposing the following conditions:(cid:40) z (ξ = 0) = ia (58a) (58b) (59) (60) one obtains: A = and: y (ξ = 1) = a tan α cos αΓ(cid:0)1 − α √ π a π πξ2(1− α π ) 2 (π − α) π (cid:1) , B = ia (cid:1) Γ(cid:0)1 + α (cid:18)1 − ξ2 (cid:19) α (cid:18)1 H(ζ) π − 1 2 z (ξ) = B + A (cid:19) (cid:82) ∞ 0 xt−1 exp (−x)dx is the gamma function [46] and, 2F1 (a, b, c, z) =(cid:80)∞ In the foregoing equations, a is the depth of the notch, 2α is the opening angle, Γ (t) = k=0 (a)k (b)k / (c)k zk/k! is the Gaussian hypergeometric function [47, 48, e.g.]. A schematic representation of the con- ξ2 − 1 − α π where H(ζ) =2 F1 , 1 − α π , 2 − α π , ξ2 2 23 Fig. 11: Electrostatic solution for an infinite transversely isotropic body featuring a sharp V-notch of finite depth a = 1 and opening angle 2α = 90◦. (a) Electric potential distribution; (b) magnitude of the current density. formal transformation is provided in Figure 10. 4.3.1 Optimal fiber paths In this case, the boundary is described by the condition v = v0 = 0. According to Section 2, the optimal fiber paths can be found by means of equation (16) imposing x (s, uc, vc) = Re [(z ◦ ξ) (s, uc, vc)] and y (s, uc, vc) = Im [(z ◦ ξ) (s, uc, vc)]. The dashed lines in Figure 2a,b show a schematic representation of the fiber paths that minimize the electrical resistance of the body, coinciding with the field lines, both in the physical and transformed domain. The continuous lines represent curves that are rotated of 45◦ with respect to the field lines. 4.3.2 Electric potential The electric potential can be calculated recalling that, in case v = v0 describes the isolated boundary ∂Ωn,0, the potential is ϕ (u, v) = ζu. Since the solution in Cartesian coordinates cannot be found easily in closed-form owed to the complexity of equation (60), only the implicit description of the potential is provided as follows: ϕ (x, y) = ζ Re [ξ (z)] (61) Figure 11a show the contour plot of the electric potential for ζ = 1 V. 24 4.3.3 Current density The current density can be calculated taking advantage of equation (15): (cid:20)dz (ξ) (cid:21)−1 dzξ Jx − iJy = ψ (ξ2 − 1)1/2−α/π ξ1−2α/π = ψ (62) which can be used to calculate the following Cartesian components of the current density in curvilinear coordinates: (u2 − v2 − 1)2 + 4u2v2(cid:105)1/4−α/2π (cid:104) (u2 − v2 − 1)2 + 4u2v2(cid:105)1/4−α/2π (cid:104) (u2 + v2)1/2−α/π cos η (u2 + v2)1/2−α/π sin η Jx (u, v) = ψ Jy (u, v) = −ψ (63) (64) where η = (1/2 − α/π) Arg (u2 − v2 − 1 + 2iuv)−(1 − 2α/π) Arg (u + iv). Figure 11b shows the distribution of the magnitude of the current density for ψ = 1 A/mm2 whereas Figure 12a shows a double-logarithmic plot of the Cartesian component of the current density, Jx, along the bisector of the notch calculated by equation (63) and FEA. For the numerical analysis, a plate of height H = 1000 mm, width D = 500 mm, notch depth a = 10 mm, and opening angles 2α = 45◦, 90◦, and 135◦ was considered. As can be noted, the current density close to the notch tip is singular of order 1/q − 1 with q = (2π − 2α) /π. The theoretical predictions (solid lines) and numerical results (symbols) are in excellent agreement. 4.4 Rectangular notch Let us consider a rectangular cutout in an infinite plate subjected to a remote current density or a potential differential. The conformal mapping to describe the notch profile can be chosen as: z (ξ) = DE2 ξ, /E2c (65) (cid:18) (cid:19) b a (cid:18) b (cid:19) a 0 1 − y2 sin2 θdθ and E2c (y) = E2 (π/2, y) are the incomplete and com- plete elliptic integrals of the second kind respectively [46] and D is a parameter controlling the length of the net section of the plate (Figure 13). The notches are described by the condition u = ±π/2 while the width of the notch, h, can be expressed as: where E2 (x, y) =(cid:82) x (cid:112) h = Im lim v→∞ DE2 (π/2 + iv, b/a) /E2c (b/a) (66) (cid:110) 25 (cid:111) Fig. 12: Comparison between the theoretical solution and FEA: (a) double-logarithmic plot of the current density component, Jx, along the bisector of a finite V-notch for various opening angles 2α = 45◦, 90◦, 135◦. As can be noted, the current density close to the notch tip is singular of order 1/q − 1 with q = (2π − 2α) /π; (b) Cartesian components of the current density as a function of the curvilinear coordinate, u, for v = 3. upon the condition that b ≤ a. Then, the ratio between the width of the notch and the ligament can be calculated explicitly by the following equation [28]: (cid:114) b a = −2 h D where E1c (y) =(cid:82) π/2 0 Re{E2c (1 − a/b) − a/bE1c (1 − a/b)} − Im{E2c (a/b) − (1 − a/b) E1c (a/b)} E2c (b/a) dθ√ 1−y2 sin2 θ is the complete integral of the first kind [46]. (67) 4.4.1 Optimal fiber paths As shown in Figure 13, the boundary of the rectangular cutout is described by the condition u = ±π/2. According to Section 2, the optimal fiber paths can be found by means of equation (15) imposing x (s, uc, vc) = Re [(z ◦ ξ) (s, uc, vc)] and y (s, uc, vc) = Im [(z ◦ ξ) (s, uc, vc)]. The dashed lines in Figure 14a show a schematic representation of the fiber paths that minimize the electrical resistance of the body, coinciding with the field lines. 4.4.2 Electric potential The electric potential can be calculated by means of equation (9) although, in this case, the solution in Cartesian coordinates cannot be found easily in closed-form. The implicit 26 Solid lines: Eq. (62)Symbols: FEASolid lines: Eq. (69)Symbols: FEAFig. 13: Schematic representation of the conformal mapping in equation (65) which trans- forms the infinite strip [−π/2, π/2]×(−∞,∞) to an infinite domain featuring two rectangular notches of width 2h of infinite depth with a finite ligament 2D. description of the potential reads: ϕ (x, y) = ζ Im [ξ (z)] (68) The representation of equation (68) in curvilinear coordinates is, of course, trivial. Figure 14b show the contour plot of the electric potential for ζ = 1 V. 4.4.3 Current density The current density can be calculated from the first derivative of ξ = ξ (z). However, since this function is difficult to calculate explicitly in this case, one can leverage equation (13) and the derivative of equation (65), obtaining the following equation: (cid:20)dz (ξ) (cid:21)−1 dξ (cid:112)1 − b/a sin2 (ξ) E2c (b/a) = ψ D Jy + iJx = ψ (69) which can be used to calculate the Cartesian components of the current density in curvilinear coordinates: a2 27 Jy (u, v) = ψ Jy (u, v) = ψ 4 (cid:114) (cid:114) 4 4b2 sin2(u) cos2(u) sinh2(v) cosh2(v) 1 − b sin2(u) cosh2(v)−cos2(u) sinh2(v) Im(cid:2)E2c Im(cid:2)E2c a2 a (cid:0) b (cid:0) b a (cid:1)(cid:3) sin(cid:0) 1 (cid:1)(cid:3) cos(cid:0) 1 2η(cid:1) + Re(cid:2)E2c (cid:104) 2η(cid:1) − Re(cid:2)E2c (cid:104) + a (cid:0) b (cid:0) b a (cid:1)(cid:3) cos(cid:0) 1 2η(cid:1) (cid:1)(cid:3) sin(cid:0) 1 2η(cid:1) a 4b2 sin2(u) cos2(u) sinh2(v) cosh2(v) 1 − b sin2(u) cosh2(v)−cos2(u) sinh2(v) + a (cid:105)2 (cid:105)2 (70) (71) Fig. 14: Electrostatic solution for an infinite body featuring curvilinear transverse isotropy weakened by a deep rectangular cutout: (a) fiber path of minimum electrical resistance; (b) electric potential field; (c) magnitude of the electric current density. The notch is described by equation (65) leading to a notch width of 2h = 1.45 mm and a ligament length of 2D = 2 mm. where η = Arg(cid:2)1 − b/a sin2(u + iv)(cid:3). Figure 14c shows the distribution of the magnitude of the current density for ψ = 1 A/mm2 whereas Figure 12b shows a comparison between the Cartesian components of the current density along the curve v = 3 calculated by equations (70) and (71) and FEA. For the numerical analysis, a plate of height H = 1000 mm, width W = 500 mm, ligament length 2D = 150 mm was considered. Again, it is worth noting that the theoretical and numerical results are in excellent agreement. 4.5 Electrical resistance of CTI systems vs. traditional composites In the foregoing sections, a theoretical framework for the closed-form solution of linear elec- trostatic problems in media featuring Curvilinear Transverse Isotropy (CTI) was proposed. The analytical formulation enabled the derivation of equation (39) which allows the explicit calculation of the electrical resistance of the CTI system for any fiber orientation. Such an expression can be simplified to equation (40) when the fibers follow the field lines, which was shown to be the case optimizing the electrical conductivity. Leveraging this result, it is interesting to compare the electrical resistance of a plate notched by a finite sharp V-notch of depth a, in case the medium exhibits CTI or the fibers follow a straight longitudinal path. The latter case is representative of the electrical conductivity of a traditional fiber composite plate reinforced by carbon fibers. The comparison is shown in Figure 15 which provides the 28 Fig. 15: Electrical resistance as a function of notch depth, a, for a finite plate featuring a sharp V-notch: straight fibers vs curvilinear fibers. All the material parameters are kept constant. The use of curvilinear fibers leads to lower electrical resistances for a given depth. The advantage of CTI is evident when the notch is deep and the electric field is significantly perturbed. electrical resistance as a function of the notch depth for a constant opening angle 2α = 30◦. The analyzed plate has a height H = 100 mm, a width D = 150 mm whereas a unit thickness was considered. A fiber volume fraction Vf = 40%, fiber electrical conductivity kf = 4× 105 S/m and matrix electrical conductivity k(ISO) = 10−16 S/m were used in equations (41a,b) to calculate the electric properties of the medium in its material coordinate system for both configurations. The electrical resistance was calculated leveraging FEA imposing a unitary potential difference at the left and right boundaries of the plate while the rest of the boundary was considered isolated (J · n = Jn = 0). This plot clearly highlights the superior perfor- mance of CTI systems compared to composites featuring straight fibers in terms of electrical resistance. In fact, for a given notch depth a, R is always lower for CTI than for traditional composites, the difference growing with increasing notch depth and reaching almost 60% for a = 0.8D. This is because, for deeper notches, the perturbation of the electric field from the un-notched solution becomes larger and larger and thus the morphology of the fibers starts to play a key role. 4.6 Comments on the applicability of the theoretical framework The theoretical framework described in the foregoing sections and the related corollaries enabling the straightforward calculation of the electrical resistance of CTI plates depend on the fulfillment of specific hypotheses. The first is that separation of length-scales can be 29 assumed so that the material can be treated as homogeneous and transversely isotropic along the curvilinear paths. It is also assumed that Dirichlet or von Neumann boundary conditions are ascribed to portions of the boundary where one of the curvilinear coordinates describing the domain is constant. This latter requirement seems particularly restricting, yet, it is not. In fact, the literature abounds of handbooks and collections of conformal mappings satisfying such conditions and enabling the description of a plethora of domains [49, 50, 51][e.g]. In case of polygonal domains, the map can be even constructed geometrically [34]. Finally, it should be noted that equation (40) can be used even if the conformal mapping is not known or it is too complicated to allow a closed-form solution. In fact, it is worth recalling that the electric potential is an harmonic function and it is constant where Dirichlet boundary conditions are applied. On the other hand, insulating boundaries are always described by field lines of constant electric field since along ∂Ωn,0 the normal derivative of the electric potential is null: ∇ϕ · n = 0. Accordingly, the electric potential ϕ and the the related orthogonal field lines represent the real and imaginary part of the conformal mapping satisfying exactly the hypotheses. Of course, these quantities are also the unknowns of the problem. Hence, their use does not provide much information if the closed-form solution of the electrostatic potential and the related electric field is of interest. However, the fact that these functions exist, confirms the general validity of equation (40). In other words, this equation can always be used to correlate the electric resistance of CTI plates to the isotropic ones. Finally, it worth noting that the hypotheses on the description of the Dirichlet and von Neumann boundary conditions could be further relaxed. In such a case, however, the solution of the governing equation could be obtained only by means of a Laurent series expansion, not a closed-form solution. 5 Conclusion This work proposed a general framework for the exact, closed-form solution of linear elec- trostatic problems in media featuring Curvilinear Transverse Isotropy (CTI), i.e. materials in which a matrix phase is reinforced by fibers following desired curvilinear paths. The ap- proach enables the exact calculation of the electric potential, the electric field, the electrical resistance as well as the family of fiber paths maximizing the electrical conductivity. Based on the results obtained in this study, the following conclusions can be elaborated: 1. to maximize the electrical conductivity of the system, the fiber paths should follow the field lines of the related isotropic electrostatic problem; 2. in case the curvilinear fibers are rotated of a constant angle ρ0 with respect to the field 30 lines, the current density of the CTI system coincides with the isotropic case but for a multiplying constant that depends on the boundary conditions of the problem and the properties of the constituents. In such a case, the curvilinear components of the current density can be described by a complex analytic function which was proven to be the first derivative of the conformal map required to describe the domain of the problem, Ω ∪ ∂Ω; 3. thanks to the foregoing result, the solution of the electrostatic problem comes down to simply identifying the conformal mapping properly describing the domain; 4. in general, while the electric current densities are formally similar, the electric potential and electric field in CTI and isotropic materials take different expressions. However, when the fibers follow the paths maximizing the electrical conductivity (ρ0 = 0), all the quantities of interest take the same form; 5. the theoretical framework enabled the formulation of a strikingly simple equation which relates the electrical resistance of the isotropic case to the one of the CTI system. This equation depends only on the properties of the constituents, the volume fraction of fibers and fiber angle; 6. the foregoing result is particularly important for the design of CTI components. First, it allows an easy estimate of the reduction of electrical resistance introduced by the addition of conductive fibers. Since the expression is derived in closed-form, it is possible to know, explicitly, the volume fraction of fibers and fiber conductivity required to achieve a desired electrical resistance. Second, the equation simplifies significantly the numerical simulation of CTI structures. In fact, one can simply simulate the isotropic case using standard FE solvers and then use the proposed equation to infer the performance of the related CTI system avoiding the need for advanced user subroutines to describe the fibers and the inaccuracy introduced by the generally poor description of the fiber paths in commercial FE solvers; 7. it was shown that CTI systems guarantee a superior performance compared to e.g. structures featuring straight fibers, for the same properties of the constituents. In the presence of a notch, the advantage of CTIs over traditional systems becomes more and more significant with increasing notch depths or, in other words, increasing perturba- tion of the current density field compared to the un-notched case; 8. all the foregoing results are of utmost importance for the design of material systems leveraging Curvilinear Transverse Isotropy (CTI) to achieve superior electrical or ther- 31 mal conductivity. Possible applications include the use of this technology to enable damage self-sensing based on resistivity measurements in typically insulating materi- als such as e.g. polymers. In such a case, the addition of conductive, curvilinear fibers can lead to sufficient levels of conductivity. Further, the fiber paths can be designed to e.g. maximize the change in electrical resistance in the presence of a crack initi- ating from a notch. Ongoing work by the authors towards this direction is providing outstanding results. Acknowledgement Marco Salviato acknowledges the financial support from the Haythornthwaite Foundation through the ASME Haythornthwaite Young Investigator Award. This work was also partially supported by the Joint Center for Aerospace Technology Innovation (JCATI) of the state of Washington, USA, through the project titled "Manufacturing and Buckling Study of Curved Steered Sandwich Panels using Automated Fiber Placement (AFP) and Out-Of-Autoclave (OOA) for Space Launch Vehicles". References [1] Fleck N, Deshpande V, Ashby M. Micro-architectured materials: past, present and future. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2010;466(2121):2495 -- 2516. [2] Schaedler TA, Jacobsen AJ, Torrents A, Sorensen AE, Lian J, Greer JR, et al. Ultralight metallic microlattices. Science. 2011;334(6058):962 -- 965. [3] Jang D, Meza LR, Greer F, Greer JR. Fabrication and deformation of three-dimensional hollow ceramic nanostructures. Nature materials. 2013;12(10):893. [4] Bauer J, Hengsbach S, Tesari I, Schwaiger R, Kraft O. High-strength cellular ceramic composites with 3D microarchitecture. Proceedings of the National Academy of Sci- ences. 2014;111(7):2453 -- 2458. [5] Bauer J, Schroer A, Schwaiger R, Kraft O. Approaching theoretical strength in glassy carbon nanolattices. Nature materials. 2016;15(4):438. [6] Gao W, Zhang Y, Ramanujan D, Ramani K, Chen Y, Williams CB, et al. The sta- tus, challenges, and future of additive manufacturing in engineering. Computer-Aided Design. 2015;69:65 -- 89. 32 [7] de Obaldia EE, Jeong C, Grunenfelder LK, Kisailus D, Zavattieri P. Analysis of the me- chanical response of biomimetic materials with highly oriented microstructures through 3D printing, mechanical testing and modeling. Journal of the mechanical behavior of biomedical materials. 2015;48:70 -- 85. [8] Martini R, Balit Y, Barthelat F. A comparative study of bio-inspired protective scales using 3D printing and mechanical testing. Acta biomaterialia. 2017;55:360 -- 372. [9] Croft K, Lessard L, Pasini D, Hojjati M, Chen J, Yousefpour A. Experimental study of the effect of automated fiber placement induced defects on performance of composite laminates. Composites Part A: Applied Science and Manufacturing. 2011;42(5):484 -- 491. [10] Dirk HJL, Ward C, Potter KD. The engineering aspects of automated prepreg layup: History, present and future. Composites Part B: Engineering. 2012;43(3):997 -- 1009. [11] Nik MA, Fayazbakhsh K, Pasini D, Lessard L. Optimization of variable stiffness com- posites with embedded defects induced by automated fiber placement. Composite Struc- tures. 2014;107:160 -- 166. [12] Wang F, Guo X, Xu J, Zhang Y, Chen C. Patterning Curved Three-Dimensional Journal of Applied Mechanics. Structures With Programmable Kirigami Designs. 2017;84(6):061007. [13] Nik MA, Lessard L, Pasini D. Size-dependent behavior of laminates with curvilinear fibers made by automated fiber placement. Science and Engineering of Composite Materials. 2015;22(2):157 -- 163. [14] Dodwell TJ, Butler R, Rhead AT. Optimum Fiber Steering of Composite Plates for Buckling and Manufacturability. AIAA Journal. 2016;54(3):1146 -- 1149. [15] Le JL, Du H, Dai Pang S. Use of 2D Graphene Nanoplatelets (GNP) in cement compos- ites for structural health evaluation. Composites Part B: Engineering. 2014;67:555 -- 563. [16] Pierson HO. Handbook of carbon, graphite, diamonds and fullerenes: processing, prop- erties and applications. William Andrew; 2012. [17] Du X, Xu F, Liu HY, Miao Y, Guo WG, Mai YW. Improving the electrical conductivity and interface properties of carbon fiber/epoxy composites by low temperature flame growth of carbon nanotubes. RSC Advances. 2016;6(54):48896 -- 48904. [18] Ohm GS. Die galvanische Kette, mathematisch bearbeitet. TH Riemann; 1827. 33 [19] Fisher SD. Complex Variables. New York, Dover; 1999. [20] Brown JW, Churchill RV. Complex Variables and Applications. New York, McGraw- Hill; 2013. [21] Leathem JG. XIV. Some applications of conformal transformation to problems in hy- drodynamics. Phil Trans R Soc Lond A. 1915;215(523-537):439 -- 487. [22] Shepherd W. The torsion and flexure of shafting with keyways or cracks. Proc R Soc Lond A. 1932;138(836):607 -- 634. [23] Wigglesworth L, Stevenson A. Flexure and torsion of cylinders with cross-sections bounded by orthogonal circular arcs. Proceedings of the Royal Society of London Series A, Mathematical and Physical Sciences. 1939;p. 391 -- 414. [24] Wigglesworth L. The flexure and torsion of an internally cracked shaft. Proc R Soc Lond A. 1939;170(942):365 -- 391. [25] Greenberg MD. Advanced engineering mathematics. vol. 2. Prentice Hall Upper Saddle River, NJ; 1998. [26] Ru C. Eshelby's problem for two -- dimensional piezoelectric inclusions of arbitrary shape. In: Proceedings of the Royal Society of London A: Mathematical, Physical and Engi- neering Sciences. vol. 456. The Royal Society; 2000. p. 1051 -- 1068. [27] Baant MZ. Conformal mapping of some non-harmonic functions in transport theory. In: Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences. vol. 460. The Royal Society; 2004. p. 1433 -- 1452. [28] Salviato M, Zappalorto M. A unified solution approach for a large variety of antiplane shear and torsion notch problems: Theory and examples. International Journal of Solids and Structures. 2016;102:10 -- 20. [29] Salviato M, Zappalorto M, Maragoni L. Exact solution for the mode III stress fields ahead of cracks initiated at sharp notch tips. European Journal of Mechanics-A/Solids. 2018;72:88 -- 96. [30] Anselmo T, Nelson R, Carneiro da Cunha B, Crowdy DG. Accessory parameters in conformal mapping: exploiting the isomonodromic tau function for Painlev´e VI. Pro- ceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2018;474(2216):20180080. 34 [31] Biot MA. General solutions of the equations of elasticity and consolidation for a porous material. J appl Mech. 1956;23(1):91 -- 96. [32] Biot MA. On the instability and folding deformation of a layered viscoelastic medium in compression. J appl Mech. 1959;26:393 -- 400. [33] Zappalorto M, Salviato M, Maragoni L. Analytical study on the mode III stress fields due to blunt notches with cracks. Fatigue & Fracture of Engineering Materials & Struc- tures. 2018;. [34] Driscoll T, Trefethen L. Schwarz -- Christoffel mapping. Vol. 8 of Cambridge Monographs on Applied and Comput. Math. Cambridge University Press, Cambridge, UK; 2002. [35] Sadd MH. Elasticity: theory, applications, and numerics. Academic Press; 2009. [36] Bower AF. Applied mechanics of solids. CRC press; 2009. [37] Sevostianov I, Kachanov M. Connections between elastic and conductive properties of heterogeneous materials. In: Advances in applied mechanics. vol. 42. Elsevier; 2009. p. 69 -- 252. [38] Abaqus. Abaqus user manual (version 6.14). 2014;. [39] Eshelby JD. The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc R Soc Lond A. 1957;241(1226):376 -- 396. [40] Mori T, Tanaka K. Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta metallurgica. 1973;21(5):571 -- 574. [41] Mura T. Micromechanics of defects in solids. Springer Science & Business Media; 2013. [42] Nemat-Nasser S, Hori M. Micromechanics: overall properties of heterogeneous materi- als. vol. 37. Elsevier; 2013. [43] Hughes TJ, Cottrell JA, Bazilevs Y. Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement. Computer methods in applied mechanics and engineering. 2005;194(39-41):4135 -- 4195. [44] Neuber H. Theory of notch stresses: principles for exact stress calculation of strength with reference to structural forms and materials. AEC TR 4547. 1958;. [45] Neuber H. Kerbspannungslehre-Theorie der Spannungskonzentration, Genaue Berech- nung der Festigkeit Springer. Verlag, Berlin. 1985;3. 35 [46] Abramowitz M, Stegun IA, et al. Handbook of mathematical functions: with formulas, graphs, and mathematical tables. vol. 55. Dover publications New York; 1972. [47] Yoshida M. Hypergeometric Functions, My Love: Modular Interpretation of Configura- tion Spaces. Vieweg Verlag, Wiesbaden; 1997. [48] Andrews GE, Askey R, Roy R. Special Functions, volume 71 of Encyclopedia of Math- ematics and its Applications. Cambridge University Press, Cambridge; 1999. [49] Mathews JH. Conformal mapping dictionary;. http://mathfaculty.fullerton.edu/ mathews/. [50] Ivanov VI, Trubetskov MK. Handbook of conformal mapping with computer-aided visualization. CRC press; 1994. [51] Kythe PK. Handbook of Conformal Mappings and Applications. 2019;. 36
1906.05727	1	1906	2019-06-12T09:59:49	High-efficiency thermophotovoltaic system that employs an emitter based on a silicon rod-type photonic crystal	[ "physics.app-ph", "physics.optics" ]	Thermophotovoltaic systems in principle enable utilization of heat that is usually regarded as wasted energy. However, the wavelength selectivity of the thermal emitter required for high efficiencies is rather difficult to control with conventional designs. Here, we design a thermophotovoltaic system comprising silicon rods as thermal emitter with a relatively narrow emission spectrum and a photovoltaic cell with a band gap corresponding to 1.76 $\mu$m, and verify efficient power generation. By accurately measuring the heat flux that enters the emitter, the emitter temperature, and the electrical output power of the photovoltaic cell, we find that the actual system efficiency (ratio of ingoing heat flux to output power) is 11.2% at an emitter temperature of 1338 K, and that the output power density footprint is 0.368 W/cm2. The obtained efficiency is relatively high, i.e., 1.65 times that of the previously reported record value (6.8%). Further efficiency improvements in the future may lead to development of distributed energy supplies using combustion heat.	physics.app-ph	physics	High-efficiency thermophotovoltaic system that employs an emitter based on a silicon rod-type photonic crystal Masahiro Suemitsu1,2, Takashi Asano1, Takuya Inoue1, Susumu Noda1 1Department of Electronic Science and Engineering, Kyoto University, Kyoto 606-8501, Japan 2Energy Technology Laboratories, Osaka Gas Co., Ltd., 6-19-9 Torishima, Konohana-Ku 554-0051, Osaka, Japan Corresponding author: [email protected] Abstract Thermophotovoltaic systems in principle enable utilization of heat that is usually regarded as wasted energy. However, the wavelength selectivity of the thermal emitter required for high efficiencies is rather difficult to control with conventional designs. Here, we design a thermophotovoltaic system comprising silicon rods as thermal emitter with a relatively narrow emission spectrum and a photovoltaic cell with a band gap corresponding to 1.76 m, and verify efficient power generation. By accurately measuring the heat flux that enters the emitter, the emitter temperature, and the electrical output power of the photovoltaic cell, we find that the actual system efficiency (ratio of ingoing heat flux to output power) is 11.2% at an emitter temperature of 1338 K, and that the output power density footprint is 0.368 W/cm2. The obtained efficiency is relatively high, i.e., 1.65 times that of the previously reported record value (6.8%). Further efficiency improvements in the future may lead to development of distributed energy supplies using combustion heat. 1 Introduction Control of the thermal emission spectrum from a hot object has been considered for applications in various fields (1 -- 13). It is particularly important for applications such as thermophotovoltaics (14,15). A thermophotovoltaic system allows us to generate electricity from thermal energy by converting thermal radiation emitted from a hot object (the so-called thermal emitter) into electricity by a photovoltaic (PV) cell. Because such systems have the potential to provide conversion efficiencies and output power densities per volume comparable to that of fuel cells if the thermal emitter is designed correctly, they have been attracting attention in recent years (15 -- 21). The most fundamental parameter for the performance of a thermophotovoltaic system is the relation between the emission spectrum of the thermal emitter and the band-gap energy of the absorber in the PV cell. To illustrate the importance of this relation, first, consider the case where the input heat flux with power density 𝑃in is converted into black-body radiation without losses, and this radiation is converted into electricity by an ideal single-junction PV cell with a band-gap energy defined by the corresponding wavelength 𝜆Eg (Fig. 1A). The PV cell is placed adjacent to the emitter, and they have the same size. In this case, the light with wavelengths longer than 𝜆Eg (the region L) is not absorbed by the PV cell, and only the short-wavelength light (in region S) is absorbed. Hence, the ratio of absorbed energy to the total radiated energy increases as the wavelength 𝜆Eg becomes longer. This ratio is the so-called spectral efficiency denoted by 𝜂spectrum and shown with the red curve in Fig. 1B. For instance, the red curve in Fig. 1B shows that in case of a black-body with a temperature of 1300 K, the spectral efficiency is limited to about 25% for 𝜆Eg near the peak wavelength of the black-body radiation (≈ 2 m), but it increases to about 80% for 𝜆Eg = 5 m. However, the losses through hot-carrier relaxation to the band edge in the PV cell's absorber increase as the wavelength of the absorbed light becomes shorter. Therefore, the internal conversion efficiency 𝜂int (i.e., the fraction of absorbed power that is actually converted into electrical power with power density 𝑃out) decreases for larger 𝜆Eg as shown with the blue curve in Fig. 1B. Because the system efficiency defined by 𝜂system = 𝑃out/𝑃in can be also written as the product of spectral and internal efficiencies, the maximum efficiency of about 30% is obtained at 𝜆Eg ≈ 4 to 6 m (Fig. 1B; black broken curve). Here, a large 𝑃out (green curve) of more than 4 W/cm2 can be obtained. A perfect spectral efficiency would be obtained if the emission spectrum has a shape like that shown with the red curve in Fig. 1C; if a wavelength-selective emitter that perfectly cuts the long-wavelength contribution above 𝜆Eg is available, 𝜂spectrum can reach 100%. In this case, the 𝜂system is solely determined by 𝜂int as shown in Fig. 1D. 2 Furthermore, the Pin required to maintain a certain emitter temperature would be reduced by a factor of S/(S+L). Then we can expect a high system efficiency of more than 57% at 𝜆Eg = 1.8 m and a large 𝑃out on the order of 1 W/cm2. However, while this concept is attractive from the viewpoint of efficiency, the actual control of the thermal emission spectrum is difficult. Furthermore, since there exist losses such as heat conduction from the emitter to other device parts, and also factors like the non-ideal properties of the PV cell, the actual conversion efficiencies of real thermophotovoltaic systems (𝜂system_exp) are still at a low level. Until about 2013, the reported experimental system efficiencies, in which the input power has been directly measured, were around 1% (16 -- 18, 22 -- 24). Then, beginning with 2015, 𝜂system_exp exceeding 5% started to appear in the literature (19 -- 21). For example, Kohiyama et al. have reported 5.1% for 𝜂system_exp by employing a ≈1650 K hot Mo-MIM structure for the emitter and a GaSb PV cell (25) with a 𝜆Eg of 1.85 m (21). Biermann et al. have reported 𝜂system_exp = 6.8% by employing a ≈1300 K hot Si-MIM structure on a tungsten substrate for the emitter (26). and a InGaAsSb PV cell (27) with a 𝜆Eg of 2.2 m (20). The important point is that, both publications have estimated losses such as those accompanying heat conduction, which can be reduced by upscaling, and have presented a system efficiency of about 10% by removing these losses from the denominator in 𝜂system_exp. Novel emitter designs may enable even higher efficiencies. Semiconductor based photonic crystal (PC) structures seem to be advantageous for thermal emitters, because anomalous photonic density of states of a PC and an absorption spectrum of a semiconductor can be utilized to tune the emission spectrum. In this work, we designed a thermophotovoltaic system that employs a Si rod-type PC, whose emission profile can be controlled better than those of the metallic thermal emitters in the previous reports. Owing to the photonic density of states of the rod-type PC and the step-like absorption spectrum of Si (11), we obtain a highly suitable emission spectrum for the employed InGaAs single-junction device. With this we experimentally obtain 𝜂system_exp = 11.2% under an emitter temperature of 1338 K. For the system efficiency excluding the losses that can be reduced via upscaling we obtain 14.5%, and the efficiency that can be expected by further improvements of the contact grid design of the PV cell is 19.4%. With respect to thermophotovoltaic systems in which the input power has been directly measured, our experimentally determined efficiency can be considered as the present world record. Furthermore, the predicted upscaled system efficiency exceeds the previously reported semi-empirical limit. 3 Fig. 1. Differences between a black-body emitter and a selective emitter. (A) Spectral power density of a black-body emitter. The inset shows a schematic for the case of lossless conversion of the input power into black-body radiation, and the conversion of this radiation into electrical power by an ideal single-junction PV cell. (B) The conversion efficiency and output power density as a function of the wavelength equivalent to the band-gap energy of the PV cell in case of the thermophotovoltaic conversion shown in the inset of (A). (C) Spectral power density of an emitter whose spectrum is optimized for conversion. The inset shows a schematic for the case of the conversion of radiation whose wavelength contribution above 𝜆Eg has been cut, and the conversion of this radiation into electrical power by an ideal single-junction PV cell. (D) The conversion efficiency and output power density as a function of the wavelength equivalent to the band-gap energy of the PV cell in case of the thermophotovoltaic conversion shown in the inset of (C). 4 246810020406080100012345Efficiency (%)Power density (W/cm2)24681001020304050Wavelength (m)Radiated power (W/(m2nm))24681001020304050Wavelength (m)Radiated power (W/(m2nm))246810020406080100012345Efficiency (%)Power density (W/cm2)1300 KBlEg1300 KSⅬAInput power (Pin )SElectric power (Pout )Photovoltaic cellBlack-body emitterL lEg1300 KSCBlack-body emitterSelective emitterBandgap wavelength (lEg) (m)Bandgap wavelength (lEg) (m)1300 KhspectrumDInput power (Pin )SElectric power (Pout )Photovoltaic cellSelective emitterResults The thermal emitter used in the present thermophotovoltaic system is a rod-type PC with a rectangular lattice (11, 13) As shown in the illustration in Fig. 2A, this PC was fabricated by processing a polycrystalline Si thin film on a MgO substrate, which is transparent for infrared light. We prepared a 20-nm-thick HfO2 layer on the 500-m-thick MgO substrate and then grew 825 nm of polycrystalline Si using the LP-CVD method. The PC pattern was written by electron-beam lithography, and the Si rod structure was obtained by dry etching. The PC's geometry is defined by d = 360 nm, h = 825 nm, and a = 700 nm as shown in Fig. 2B. The thermal radiation spectrum that is observed along the surface normal direction from the thermal emitter with this structure when it is heated up to 1300 K, is shown with the red solid curve in Fig. 2C. Additionally, the blue curve in Fig. 2C is the theoretically determined thermal radiation spectrum along the surface normal direction for the ideal rod structure in Fig. 2A. While there are some deviations caused by the difference between the ideal structure and the fabricated structure, both spectra agree fairly well. Because the actual radiation that is incident on the PV cell is the hemispherically integrated radiation, we show its theoretical value in Fig. 2D (since the experimental integration over 2 sr is difficult). The peak wavelength is about 1.6 m, and we find that the contribution with wavelengths longer than 1.76 m is strongly suppressed when compared with the black-body radiation. For the PV cell we employ a single-junction device based on InGaAs. The photograph of the fabricated PV cell is shown in Fig. 3A, and the measured external quantum efficiency is shown in Fig. 3B. The light receiving surface of the PV cell is about 36 mm2, the cell's aperture ratio (ratio of active area to total light receiving area) is 77%, the series resistance is 0.01 cm2, the shunt resistance is 1000 cm2, the dark current is 5×10-6 A/cm2, and the diode's ideality factor n is 1.5. As can be seen in Fig. 3B, the 𝜆Eg of the absorber layer is about 1.76 m, and thus the 𝜂spectrum with respect to the spectrum in Fig. 2C is 42.6%. For comparison we note that, since the 𝜂spectrum of this PV cell under direct illumination with black-body radiation at 1300 K is 11.5%, 𝜂spectrum increased by a factor of more than three due to the control of the thermal emission via the Si rod structure. 5 Fig. 2. Properties of the PC thermal emitter. (A) Illustration of the PC thermal emitter. (B) Scanning electron microscope image of the Si rod-type PC before atomic layer deposition. (C) The experimental (red curve) and theoretical (blue curve) thermal radiation spectrum along the surface normal direction of the emitter. (D) The hemispherically integrated spectrum of the theoretical thermal radiation of the thermal emitter at 1300 K (black solid curve), and the hemispherically integrated black-body radiation spectrum for 1300 K (gray broken curve). 6 0.515100123Wavelength (m)Radiated power (W/(m2nm sr))AMgOPolycrystalline Si rodsdhatB360 nm825 nm700 nmMgOHfO2(20 nm)1300 KTheory ExperimentVertical0.5151001020304050Wavelength (m)Radiated power (W/(m2nm))1.76D1300 KIntegratedC1.76Polycrystalline Si rodsFig. 3. PV cell for electricity generation. (A) The visible structure of the fabricated PV cell. The thin lines on the cell is the front surface contact grid, which is connected to the large contact pads via several gold ribbons. (B) External quantum efficiency of the PV cell. By using the above thermal emitter and PV cell, we constructed a thermophotovoltaic system that is optimized for accurate characterization as schematically shown in Fig. 4A. For actual implementation we consider a layout as shown in Fig. 4B; combustion heat or solar heat is transferred to the emitter surface via conduction from the edge, and PV cells are placed on both sides of the emitter. In the present system for characterization, an internal heater (Ohmic resistor controlled via 4 electrical contacts) supplies the emitter with heat, and the thermal radiation from the emitter is converted into electrical power by the PV cells. The present setup also enables direct measurement of the emitter temperature by forming a micro thermocouple on the emitter. As the Si rod-type PC structure is situated on top of a transparent MgO substrate, almost the same thermal radiation spectrum is obtained on both sides of the emitter. Therefore, PV cells are placed on both sides of the emitter. In efficiency tot) and the total power that is not emitted towards the PV cells (𝑃leak measurements of thermophotovoltaic systems, the quantitative evaluation of the total tot ) is input power (𝑃in tot and difficult in general. The present setup enables a quantitative evaluation of 𝑃in tot by suspending the emitter in vacuum with thin Pt wires (for thermal insulation) 𝑃leak and converting the electrical energy transmitted through the thin wires to thermal energy tot, and via the internal heater. The 𝑃in the emitter temperature can be measured at the same time, which is considered essential tot, the total output power from the two PV cells 𝑃out for reliable results. The photograph of the thermal emitter including the heater is shown in Fig. 4C. The Si rod-type PC is formed on a 4.8×4.8-mm large MgO substrate with a 7 0.511.52020406080100Wavelength (m)External quantum efficiency (%)A5.9mm6.4 mmInGaAsphotovoltaic cellAu mountBthickness of 65 m, and the Pt/Ti heater is M-shaped with a total length of 14.3 mm, a width of 10 m, and a thickness of 300 nm. We note that the ratio of the heater's area to the PC area is 0.6%, and thus the directly generated emission of the latter can be neglected when compared to the emission of the PC. Furthermore, for an accurate temperature measurement, a 200-m squared Pt/Ti pad was prepared on the side opposite to the contacts for the heater, and connected with 25-mm long Pt and Pt0.9Rh0.1 wires with 25 m in diameter to directly fabricate an S-type thermocouple on the emitter. In the following, the power generation is tested after placing PV cells on both sides of the emitter and reducing their distance to less than 500 m as shown in Fig. 4D. The view factor between the thermal emitter and the PV cells in this configuration is about 0.97. The PV cells are mounted on copper heat sinks for cooling, and the back-surface temperature is adjusted to about 283 K. During the measurements, the whole system is placed in a vacuum chamber (about 3×10-2 Pa). Fig. 4. Thermophotovoltaic system that employs an emitter based on a Si rod-type PC. (A) Illustration of the designed thermophotovoltaic system. (B) The proposed layout for actual implementation. (C) Photograph of the fabricated thermal emitter. The heating element and the temperature measurement device are also visible. (D) The side view of the fabricated thermophotovoltaic device. 8 APhotovoltaic cell(Cell B: MgOside)Pt wiresPt/ TiheaterPhotovoltaic cell (Cell A: PC side)Thermal emitterThermocoupleD~500 mPhotovoltaic cells(t340 m)Thermal emitterCopper heatsinkCopper heatsinkCell BCell ABConcentrated sunlightMicro-combustor,Photovoltaic cellsThermal emittersConduction heat transfer from the emitter edgeS-type thermocouple4.8 mmThermal emitter(t 65 m)f 25mmPt wiresPt/ Tiheater (w 10 m)CIn the present experiment we applied electrical power to the heater in the emitter and let the emitter temperature increase stepwise while measuring the output characteristics tot applied to the of the PC cells at each temperature. The relation between the 𝑃in emitter and the emitter temperature is plotted with the blue circles in Fig. 5A. The tot = 1.24 W, which is the maximum applied emitter temperature reaches 1338 K for 𝑃in power in this experiment. The inset of Fig. 5A presents the side view of the system at 1338 K. The current -- voltage (I -- V) characteristics of the PV cells for an emitter temperature of 1338 K are presented in Fig. 5B. We find that almost the same I -- V characteristics are obtained for both Cell A (on the PC side) and Cell B (on the MgO tot = 0.139 W. The 𝑃out substrate side). The total output power from both devices at the optimum operating tot as a function of the emitter temperature is shown in point is 𝑃out Fig. 5A with the black triangles. The trend for the system efficiency 𝜂system_exp, which is tot, is plotted with the red circles in Fig 5C. The obtained by dividing 𝑃out increase of the emitter temperature is accompanied by an increase in 𝜂system_exp, and we obtain an efficiency of 11.2% at 1338 K. This experimentally obtained efficiency is 1.65 times larger than the maximum value among the 𝜂system_exp that have been previously reported (6.8%) (15, 20). tot by 𝑃in The 𝑃out of the system, which is defined as 𝑃out tot divided by the total area of the two PV cells, is plotted with black triangles in Fig. 5C. The 𝑃out is about 0.184 W/cm2 at doubles (0.368 W/cm2) 1338 K, but the output power density footprint 𝑃out because the system has two PV cells. Note that the size of each PV cell (𝐴PV cell = 0.378 cm2) is larger than that of the emitter (𝐴emitter = 0.230 cm2) in this prototype system to maintain a large view factor. When we scale up the system, we can make 𝐴PV cell the same as 𝐴emitter without degrading the view factor. 𝑃out case is mentioned in the discussion section. for this footptint footptint 9 Fig. 5. Performance of the present thermophotovoltaic system. (A) The heater's tot blue circles) and the maximum total output power provided by the input power (𝑃in tot , black triangles). The inset shows the side view of the fabricated PV cells (𝑃out thermophotovoltaic device under operation (emitter at 1338 K). (B) The theoretically and experimentally determined I -- V characteristics of Cell A and Cell B for an emitter temperature of 1338 K are shown with the red and blue curves, respectively. (C) The dependences of 𝜂system_exp (red circles) and the output power density (black triangles) on the emitter temperature. 10 80090010001100120013001400051000.10.2Temperature (K)Conversion efficiency (%)Output power density (W/cm2)40060080010001200140000.511.5Temperature (K)Input/ Output power (W) Input power Output power of the photovoltaic cells00.10.20.30.40.500.20.40.6Voltage (V)Current density (A/cm2)1338 K Cell A Cell B TheoryBACDiscussion The energy balance of the present power generation system for an emitter temperature of 1338 K is shown in Fig. 6. Losses such as the radiation from the edges of the thermal emitter, the losses via thermal radiation from the thin Pt wires which are heated up by the emitter, as well as their heat conduction, and also losses caused by the form factor exist. Because these losses can be reduced by increasing the emitter and PV tot ) and discuss the system cell areas, we quantitatively evaluate these energy losses (𝑃leak efficiency that can be achieved after a scale up of the present power generation system tot. Also, when we scale up the system, we can make by excluding 𝑃leak 𝐴PV cell the same as 𝐴emitter without degrading the view factor. tot from 𝑃in Fig. 6. The energy balance of our thermophotovoltaic system for an emitter temperature of 1338 K. Obviously, the thermal radiation that exits the emitter from its edges cannot be used as it does not strike the PV cell. This power can be determined from the total area of the edges and their emissivity as well as the emitter temperature. Firstly, the area of the MgO substrate edges is about 2.6% of that of the substrate surface area. The evaluated total emissivity of the edges is about 60% as calculated from the optical thickness of the MgO substrate when seen from the side and considering the influence of the Si rods that can be seen from the side. As a result, the power that is emitted from the edges at 1338 tot. The calculated loss that is caused by the thermal K becomes about 12.5% of 𝑃in tot. The radiation and heat conduction of the thin Pt wire at 1338 K is about 3.3% of 𝑃in energy loss that originates from the air molecules in the vicinity is only ≈0.04% owing to the measurement under a sufficiently low pressure (3×10-2 Pa), and thus can be omitted in this discussion. Furthermore, we can derive 2.5% for the loss caused by the view factor. When we evaluate the sum of these losses (18.3% of the total input power), tot = 0.183×1.24 W = 0.227 W for an emitter with 𝐴emitter = 0.230 cm2. we find 𝑃leak tot / If we assume that the power density of the above mentioned losses (𝑃leak = 𝑃leak 11 Pin=100 Pout:11.2 The loss caused byemission from the edge: 12.5The loss caused by thermal radiation and heat conduction of the Pt wires:3.3 The loss caused by the view factor: 2.5 S= 34.7L= 47.0Pt wiresPhotovoltaic cellsThermal emitter(1338 K)𝐴emitter) can be reduced by upscaling, the heat flux density required to maintain an emitter temperature of 1338 K becomes 1.01 W/ 0.230 cm2 =4.39 W/cm2. When we would be 0.139 W/ 0.230 cm2 = 0.604 W/cm2. set 𝐴PV cell = 𝐴emitter, the 𝑃out However, we have to consider the effect of the increase in the open-circuit voltage upon enhancement of the short-circuit current density. The 𝑃out when considering this effect would be 0.638 W/cm2. Therefore, the system efficiency can increase to 14.5% by footptint footptint upscaling. Finally, we examine the influence of the aperture ratio, which is determined by the geometry of the contact grid of the PV cell. The present cell's aperture ratio is 77%, but it is possible to reach 100% by employing a back surface contact scheme (28). In case of no shadowing by the electrodes and considering the above mentioned effect, the output power increases by 33.5%. Hence, the 𝑃out that can be provided at 1338 K increases to 0.852 W/cm2. Because the input heat flux density does not differ from the above derived 4.39 W/cm2, the ideal system efficiency for the present PC-emitter design footptint becomes 19.4%. 12 Conclusion In conclusion, we designed an efficient thermophotovoltaic system that employs a Si rod-type PC thermal emitter, which has a relatively narrow emission spectrum. The employed emitter exhibited a high spectral efficiency of 42.6% with respect to the band-gap energy of the absorber layer (𝜆Eg = 1.76 m). In order to quantitatively tot , we evaluate the input heat flux 𝑃in thermally isolated the emitter by suspending it in vacuum with thin Pt wires. We tot and the not usable dissipated power 𝑃leak adopted the method of converting the input electrical power into a heat flux via an internal heater in order to have a well-defined heat source. For an emitter at 1338 K, we measured a high value of 11.2% for the actual system efficiency. This is 1.65 times the highest value among the values reported so far. The 𝑃out is measured to be 0.368 W/cm2. The system efficiency estimated by removing losses that can be reduced by a footptint scale up of the system size, is 14.5%. Furthermore, we estimated that a system efficiency of 19.4% can be reached if the contact grid design of the PV cell is improved. of 0.852 W/cm2. is expected. Our experimental result can In this case, a high 𝑃out be considered as the present world record for the efficiency of a thermophotovoltaic footptint system in which the input power has been directly measured. Furthermore, the predicted upscaled system efficiency exceeds the previously reported semi-empirical limit. We expect that by further improvements in the future, the heat that is usually considered wasted can be utilized more effectively. These improvements may aid the development of distributed energy supplies using combustion heat. 13 References and Notes 1) J. F. Waymouth, Where will the next generation of lamps come from? J. Light Vis. Environ. 13, 51-68 (1989). 2) J. G. Fleming, S. Y. Lin, I. El-Kady, R. Biswas, K. M. Ho, All-metallic three-dimensional photonic crystals with a large infrared bandgap. Nature 417, 52 -- 55 (2002). 3) J. J. Greffet, R. Carminati, K. Joulain, J. P. Mulet, S. Mainguy, Y. Chen, Coherent emission of light by thermal sources. Nature 416, 61 -- 64 (2002). 4) F. Kusunoki, J. Takahara, T. Kobayashi, Qualitative change of resonant peaks in thermal emission from periodic array of microcavities. Electron. Lett. 39, 23-24 (2003). 5) K. Ikeda, H. T. Miyazaki, T. Kasaya, K. Yamamoto, Y. Inoue, K. Fujimura, T. Kanakugi, M. Okada, K. Hatade, S. Kitagawa, Controlled thermal emission of polarized infrared waves from arrayed plasmon nanocavities. Appl. Phys. Lett. 92, 021117 (2008). 6) T. Asano, K. Mochizuki, M. Yamaguchi, M. Chaminda, S. Noda, Spectrally selective thermal radiation based on intersubband transitions and photonic crystals. Opt. Express 17, 19190 -- 19203 (2009). 7) X. Liu, T. Tyler, T. Starr, A. F. Starr, N. M. Jokerst, W. J. Padilla, Taming the blackbody with infrared metamaterials as selective thermal emitters. Phys. Rev. Lett. 107, 045901 (2011). 8) M. De Zoysa, T. Asano, K. Mochizuki, A. Oskooi, T. Inoue, S. Noda, Conversion of broadband to narrowband thermal emission through energy recycling. Nature Photon. 6, 535-539 (2012). 9) T. Inoue, M. De Zoysa, T. Asano, S. Noda, Single-peak narrow-bandwidth mid-infrared thermal emitters based on quantum wells and photonic crystals. Appl. Phys. Lett. 102, 191110 (2013). 10) T. Inoue, M. De Zoysa, T. Asano, S. Noda, Realization of narrowband thermal emission with optical nanostructures. Optica, 2, 27 (2015). 11) T. Asano, M. Suemitsu, K. Hashimoto, M. D. Zoysa, T. Shibahara, T. Tsutsumi, and S. Noda, Near-infrared -- to -- visible highly selective thermal emitters based on an intrinsic semiconductor. Science Adv. 2, e1600499 (2016). 12) K. Du, L. Cai, H. Luo, Y. Lu, J. Tian, Y. Qu and Q. Li, Wavelength-tunable mid-infrared thermal emitters with a non-volatile phase changing material. Nanoscale, 10 (9), 4415-4420 (2018). 13) M. Suemitsu, T. Asano, M. De Zoysa and S. Noda, Wavelength-selective thermal 14 emitters using Si-rods on MgO. Appl. Phys. Lett. 112 011103 (2018). 14) B. Wedlock, Thermo-photo-voltaic energy conversion. Proc. IEEE 51, 694 (1963). 15) D. G. Baranov, Y. Xiao, I. A. Nechepurenko, A. Krasnok, A. Alù and M. A. Kats, Nanophotonic engineering of far-field thermal emitters, Nature Mat., https://doi.org/10.1038/s41563-019-0363-y, (2019). 16) A. Datas, and C. Algora, Development and experimental evaluation of a complete solar thermophotovoltaic system. Prog. Photovolt. Res. Appl. 21, 1025 (2012). 17) M. Shimizu, A. Kohiyama, and H. Yugami, High-efficiency solar-thermophotovoltaic system equipped with a monolithic planar selective absorber/emitter. J. Photon. Energy 5, 053099 (2015). 18) A. Lenert, D. M. Bierman, Y. Nam, W. R. Chan, I. Celanović, M. Soljačić, and E. N. Wang, Nature Nano. 9, 126 (2014). 19) C. Ungaro, S. K. Gray, and M. C. Gupta, Solar thermophotovoltaic system using nanostructures, Opt. Express 23, A1149 (2015). 20) D. M. Bierman, A. Lenert,W. R. Chan, B. Bhatia, I. Celanovic, M. Soljacic, and E. N.Wang, Enhanced photovoltaic energy conversion using thermally based spectral shaping. Nature Energy 1, 16068 (2016). 21) A. Kohiyama, M. Shimizu, and H. Yugami, Unidirectional radiative heat transfer with a spectrally selective planar absorber/emitter for high-efficiency solar thermophotovoltaic systems. Appl. Phys. Express 9, 112302 (2016). 22) F. Demichelis, E. Minetti-Mezzetti, M. Agnello, and E. Tresso, Evaluation of thermophotovoltaic conversion efficiency. J. Appl. Phys. 53, 9098 (1982). 23) J. L. Duomarco, and R. Kaplow, Solar thermophotovoltaics: An assessment Sol. Energy 32, 33 (1984). 24) V. Badescu, Thermodynamic theory of thermophotovoltaic solar energy conversion. J. Appl. Phys. 90, 6476 (2001). 25) L. L. Tang, H. Ye, and J. Xu, A novel zinc diffusion process for the fabrication of high-performance GaSb thermophotovoltaic cells. Sol. Energy Mater. Sol. Cells 122, 94 (2014). 26) W. R. Chan, P. Bermel, R. C. N. Pilawa-Podgurski, C. H. Marton, K. F. Jensen, J. J. Senkevich, J. D. Joannopoulos, M. Soljačić, and I. Celanovic, Toward high-energy-density, high-efficiency, and moderate-temperature chip-scale thermophotovoltaics. Proc. Natl. Acad. Sci. 110, 5309 (2013). 27) M. W. Dashiell, J. F. Beausang, H. Ehsani, G. J. Nichols, D. M. Depoy, L. R. Danielson, P. Talamo, K. D. Rahner, E. J. Brown, S. R. Burger, P. M. Fourspring, W. F. Topper,P. F. Baldasaro, C. A. Wang, R. K. Huan, M. K. Connors, G. W. Turner, Z. 15 A. Shellenbarger, G. Taylor, J. Li, R. Martinelli, D. Donetski, S. Anikeev, G. L. Belenky, and S. Luryi, Quaternary InGaAsSb Thermophotovoltaic Diodes. IEEE Trans. Elect. Dev. 53, 2879 (2006). 28) M. D. Lammert and R. J. Schwartz, The Interdigitated Back Contact Solar CelAl: Silicon Solar Cell for Use in Concentrated Sunlight, IEEE Trans. Elect. Dev. 24, 337, (1977). Acknowledgments: We thank K. Ishizaki for fruitful discussions. Funding: This work was supported in part by Japan Society for the Promotion of Science Grant-in-Aid for Scientific Research (KAKENHI) grant no. 17H06125. Author contributions: M.S. performed the system designs, analyses and experiments. T.A. and S.N. designed the study. T.I. examined the photovoltaic cell. All authors participated in discussing the results. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper s. Additional data related to this paper may be requested from the authors. 16 Supplementary Materials for High-efficiency thermophotovoltaic system that employs an emitter based on a silicon rod-type photonic crystal Masahiro Suemitsu1,2, Takashi Asano1, Takuya Inoue1, Susumu Noda1 Affiliations: 1Department of Electronic Science and Engineering, Kyoto University, Kyoto 606-8501, Japan 2Energy Technology Laboratories, Osaka Gas Co., Ltd., 6-19-9 Torishima, Konohana-Ku, Osaka 554-0051, Japan Correspondence to: [email protected] This PDF file includes: Materials and Methods 17 Materials and Methods Fabrication of the thermal emitter We prepared a 20-nm thick HfO2 layer on a 500-m thick MgO substrate and then grew 825 nm of poly-crystalline Si using the LP-CVD method. The PC pattern was written by electron-beam lithography, and the Si rod structure was obtained by dry etching. The locations of the heater, the thermocouple, and their electrical contact pads were first patterned by photolithography, and the Si rod structure in these locations was removed by dry etching. After the resist mask had been removed, we fabricated a 30-nm thick Al2O3 layer on the whole emitter by atomic layer deposition to improve the heat resistance. For the formation of the heater and thermocouple elements, a film was fabricated by successively sputtering HfO2 and Al2O3 (each time 50 nm). Then the patterns of the electrodes were written by photolithography, and finally the Pt/Ti layers with thicknesses of 300 and 30 nm were made by sputtering. The thermal emitter with the fabricated heater and thermocouple elements was cut into a square with size 4.8×4.8 mm, and thinned to 65 m using a polishing machine. After polishing, Pt and Pt0.9Rh0.1 wires with f 25 m were attached to their corresponding contact pads by using a wedge bonder. We note that at the contact pads for the heater, we attached two Pt wires on each pad. With this four-terminal configuration, we accurately measured the voltage difference between the contact pads of the heater, an evaluated the electrical power consumed by the heater inside the thermal emitter. The wires attached to the contact pads are connected to a universal electrical board via soldering, and we adjusted each wire's length to 25 mm. Fabrication of the photovoltaic cell The PV cell used in this work is based on InGaAs. First, an n-type InP layer with thickness 0.1 m and a doping concentration of n = 3×1018 cm-3 was epitaxially grown on a semi-insulating Fe-InP substrate by MOCVD. This was followed by the growth of n-type In0.53Ga0.47As (n = 3×1018 cm-3, 0.3 m), p-type In0.53Ga0.47As (p = 2×1017 cm-3, 2.0 m), p-type InP (p = 2×1018 cm-3, 0.1 m), and finally a p+-In0.53Ga0.47As layer (p = 5×1018 cm-3, 0.2 m). The area of the electrode for the p-side was defined by photolithography, and a Au/Pt/Ti layer structure was deposited by evaporation. The p+-In0.53Ga0.47As layer in the area outside the region of the electrode for the p-side was removed by etching. The area of the electrode for the n-side was patterned by photolithography, and the p-InP/ p-In0.53Ga0.47As/ n-In0.53Ga0.47As structure was removed by etching. A Au/Pt/Ti layer structure was fabricated on the exposed n-InP area, and then the device was annealed at 673 K for 20 s using a rapid thermal processing 18 (RTA) process. After annealing, the PV cell area without electrodes was coated with a 206-nm-thick SiO2 layer in order to reduce the reflection of light with wavelengths shorter than 1.76 m. After this, we increased the thickness of the electrode part to 7 m by plating, which reduces the series resistance. The backside of the PV cell was prepared by depositing 150 nm Al via evaporation. The fabricated PV cell was attached to a gold-plated copper substrate using a conductive Ag paste (Fujikura Kasei, XA-874), and connected to an external source meter via 10 m thick and 100 m broad gold ribbons. 19
1901.07188	1	1901	2019-01-22T07:10:05	Bistatic scattering forward model validation using GNSS-R observations	[ "physics.app-ph" ]	In this paper we advance a previously developed bistatic scattering forward model to include the circularly polarized incident and scattered waves, which is the case for Global Navigation Satellite System (GNSS) reflectometry. This model development enables retrieval of soil moisture from Signals of Opportunity (SoOp) bistatic observations, e.g., from the Cyclone Global Navigation Satellite System (CYGNSS) observations and GNSS Reflectometer Instrument for Bistatic Synthetic Aperture Radar (GRIBSAR). In order to validate the forward model with measured data, we present a method to construct the Delay Doppler Map (DDM) from simulations of the forward model. The forward model Radar Cross Section (RCS) predictions will be compared with actual measurements. The validated model is intended for use in soil moisture retrievals.	physics.app-ph	physics	BISTATIC SCATTERING FORWARD MODEL VALIDATION USING GNSS-R OBSERVATIONS Amir Azemati(1), Mahta Moghaddam(1), Arvind Bhat(2) (1) Ming Hsieh Department of Electrical Engineering University of Southern California, Los Angeles, CA, 90089 (2) Intelligent Automation INC. (IAI), Rockville, MD, 20855 ABSTRACT In this paper we advance a previously developed bistatic scattering forward model to include the circularly polarized incident and scattered waves, which is the case for Global Navigation Satellite System (GNSS) reflectometry. This model development enables retrieval of soil moisture from Signals of Opportunity (SoOp) bistatic observations, e.g., from the Cyclone Global Navigation Satellite System (CYGNSS) observations and GNSS Reflectometer Instrument for Bistatic Synthetic Aperture Radar (GRIBSAR). In order to validate the forward model with measured data, we present a method to construct the Delay Doppler Map (DDM) from simulations of the forward model. The forward model Radar Cross Section (RCS) predictions will be compared with actual measurements. The validated model is intended for use in soil moisture retrievals. Index Terms -- SoOp, CYGNSS, GRIBSAR, DDM, RCS 1. INTRODUCTION Global land surface observations of soil moisture are essential for better understanding of how water, carbon, and energy cycles are linked. Soil moisture mapping has crucial impacts on many areas of human interest including weather and climate forecasting, wildfire and flood prediction, agriculture and drought analysis, and human health [1-5]. Soil moisture can be retrieved using active or passive microwave observations. Active microwave systems, or radars, can provide higher resolution estimates of soil moisture than their passive counterparts. Conventional monostatic radars, however, are expensive, and it is a total loss if their transmitter fails, as in the case of the Soil Moisture Active Passive (SMAP) mission. In contrast, bistatic radars are much more resilient in that their transmitters are pervasively present, such as in the case of GPS/GNSS transmitters. Therefore, it is important to advance novel soil moisture Fig. 1. An example of overland (Herbaceous Amazon forest) Delay Doppler Map (DDM), which is derived from CYGNSS raw IF data. The raw IF processed data are measured in "counts" unit, which is proportional to the total received signal power. The center bin of this DDM represents scattering in specular direction, which has the highest Signal to Noise Ratio (SNR). retrieval techniques by using the available SoOp from these sources [6-12]. In our previous works [3-5], a fully bistatic scattering forward model for vegetated terrain has been developed, which provides the possibility to exploit SoOp for soil moisture retrieval over various land covers including forests. The proposed bistatic scattering model has three main scattering mechanisms: direct ground scattering, vegetation volume scattering, and double bounce scattering from ground and vegetation layer. It also includes the Right-Hand Circularly Polarized (RHCP) incident and linearly polarized scattered wave cases [3- 5]. The RHCP GNSS incident waves produce both RHCP (𝝈𝑹𝑹) and Left-Hand Circularly Polarized (LHCP) scattered waves (𝝈𝑳𝑹) upon reflection from the ground surface. Thus, in section 2 of this paper we extend our proposed bistatic scattering model to include the RHCP incident and circularly polarized (RHCP and LHCP) scattered waves scenarios. In section 3 of this paper we present a method for constructing DDMs by using the forward bistatic scattering model. A validation technique for the forward model is presented in section 4, in which the developed bistatic scattering model will be used to generate the DDMs (e.g. Fig. 1) for Tonzi ranch area in California. The accuracy of the bistatic model will be assessed with data available from bistatic platforms such as from CYGNSS and GRIBSAR [2-4]. then equations (5) and (6) will be utilized to derive the cross section values related to the circularly polarized incident and scattered waves (𝝈𝑹𝑹,𝝈𝑳𝑹). For the case reflection coefficients, namely (S𝑯𝑯, S𝑽𝑽) [4-5]. of scattering in the non-specular direction we will follow the same approach described above, however, we need to use the absolute value of co-polarization scattering elements of scattering matrix in lieu of 3. CONSTRUCTING DDM BY USING THE ADVANCED BISTATIC MODEL the GNSS signals as the sources of opportunity, the forward bistatic scattering model needs to handle circularly polarized incident and scattered wave 2. CIRCULAR/CIRCULAR BISTATIC MODEL The current version of bistatic scattering model includes the linearly polarized incident and scattered wave cases received signal scattered in specular direction can be derived from their corresponding reflection coefficients relationship between circularly The measurements taken from SoOp reflections from the Earth surface are represented as Delay Doppler Maps (DDMs) (Fig. 1) [2-3]. It is shown in section 2 that it is possible to use a polarimetric bistatic scattering model to compute the relationship between circularly and linearly polarized radar cross sections (𝝈𝑹𝑹, 𝝈𝑳𝑹) and (𝝈𝑯𝑯,𝝈𝑽𝑽 ,𝝈𝑯𝑽 ,𝝈𝑽𝑯). The next step is to validate Moreover, as shown in the compact polarimetry study of reflection coefficients can be written in terms of the linearly these RCS values predicted by the forward bistatic scattering model with the actual measurements of CYGNSS and GRIBSAR receivers. Since the CYGNSS and GRIBSAR data are presented as DDMs, the proposed bistatic scattering model needs to be utilized to generate these DDMs for validation purposes. In addition, the DDMs' unit will be converted from Watts to normalized circularly polarized RCS in order to be consistent with the forward model. Therefore, the conversion method, which is described in [3], will be applied to convert the scattered signal power DDM in Watts (Fig. 2) to normalized bistatic scattering RCS. (𝝈𝑯𝑯,𝝈𝑽𝑽 ,𝝈𝑯𝑽 ,𝝈𝑽𝑯) [4]. However, in order to utilize scenarios (𝝈𝑹𝑹, 𝝈𝑳𝑹). The circularly polarized (Γ𝑹𝑹, Γ𝑳𝑹). The (Γ𝑹𝑹, Γ𝑳𝑹) and circularly polarized specular radar cross sections (𝝈𝑹𝑹, 𝝈𝑳𝑹) can be expressed as [13]: 𝝈𝑹𝑹 = 𝟒𝝅Γ𝑹𝑹, (1) 𝝈𝑳𝑹 = 𝟒𝝅Γ𝑳𝑹, (2) polarized reflection coefficients (Γ𝑯𝑯, Γ𝑽𝑽) as follows: Γ𝑹𝑹 = - (Γ𝒉𝒉.Γ𝒗𝒗) 𝟐 Γ𝑳𝑹 = - (Γ𝒗𝒗1Γ𝒉𝒉) 𝟐 circularly polarized RCS (𝝈𝑹𝑹,𝝈𝑳𝑹) in terms of the linearly polarized RCS (𝝈𝑯𝑯,𝝈𝑽𝑽): 𝝈𝑹𝑹 = 𝝈𝑯𝑯.𝝈𝑽𝑽 + 3𝝈𝑯𝑯𝝈𝑽𝑽 (5) 𝟒 - 3𝝈𝑯𝑯𝝈𝑽𝑽 (6) 𝝈𝑳𝑹 = 𝝈𝑯𝑯.𝝈𝑽𝑽 𝟒 Therefore, in order to compute the 𝝈𝑹𝑹 and 𝝈𝑳𝑹 , first the linearly polarized received signal (𝝈𝑯𝑯,𝝈𝑽𝑽), and Fig. 2. DDM of the Tonzi ranch area in California, which demonstrates the GPS scattered signal power measured by the GRIBSAR. According to Fig. 2, the bright yellow bin at zero Doppler, which has the highest SNR, corresponds to the specular bin and the area around this bin is called the glistening zone. Moreover, the size of the glistening zone is determined by the root mean square of surface (3) our forward bistatic scattering model will be used to predict the radar cross section values corresponding to Equations (1)-(4) can be utilized to express the (4) [13-14], the circularly polarized polarized reflection coefficients roughness distribution [2]. In order to construct the over-land DDM with the proposed bistatic scattering model, we first compute the specular bin RCS [3] and set the specular bin as the reference point. Each bin around the specular direction has a specific set of scattering (𝜃9) and azimuth angles (𝜑9). Depending on set separate ranges for scattering (𝜃9) and azimuth (𝜑9) the size of the glistening zone, which is proportional to the random distribution of surface roughness [2], and the receiver distance from that region, it is possible to angles. The specular direction is located at the center of these ranges and the size of the glistening zone (area of interest) [2] specifies the lower and upper bounds of azimuth and scattering angle ranges. After setting the ranges for scattering and azimuth angles, we construct the over-land DDM by running the forward bistatic scattering model for each of the DDM bins, including scattering contributions from their corresponding scattering and azimuth angles. For terrains with small surface roughness a combination of Small Perturbation Method (SPM) and Kirchoff approximation (KA) is used to compute the bistatic scattering from glistening zone and specular direction (bin), respectively [15-18]. However, for terrains with high surface roughness, SPM + KA is no longer acceptable and the Stabilized Extended Boundary Condition Method (SEBCM) [19] will be applied to compute the ground bistatic scattering contribution in the future. Unlike the analytical methods (SPM+KA), the SEBCM calculates the coherent (specular direction) and the non-coherent (glistening- zone) scattering RCS all in one shot at the cost of higher computational time. 4. VALIDATION OF BISTATIC MODEL In the previous section we discussed the method used to construct the over-land DDMs with the developed forward scattering model. The next step after constructing the DDMs is the forward model validation with real measured data available from sources such as CYGNSS and GRIBSAR, using in-situ data from the Soil moisture Sensing Controller and oPtimal Estimator (SoilSCAPE) network [20] to parameterize the forward model. In order to validate the forward model, the measured DDMs will be compared with the DDMs simulated by utilizing our forward bistatic scattering model and the method described in Section 3. In August 2018, Intelligent Automation Inc. (IAI) flew the GRIBSAR instrument onboard an airplane in order to measure bistatic GPS signals scattered from Tonzi- ranch in California (Fig. 3). This site is home to one of Fig. 3. Estimated path delay of reflected GPS signal and the amplitude ratio of reflected and direct GPS signal, which are measured by the GRIBSAR for the Tonzi ranch region in California. the SoilSCAPE in-situ sensor networks, providing several nodes for soil moisture profile observations at sub-hourly temporal resolution. Thus, it provides an excellent opportunity to validate the bistatic scattering model predictions with actual GNSS-R data from GRIBSAR. 5. RESULTS AND SUMMARY To validate the forward model and the associated DDMs from Tonzi ranch, we first simulate the DDM. To do so, we use the field-measured values of soil moisture profiles from the SoilSCAPE network and our existing database of surface roughness and vegetation parameters that we have previously measured in the field, to parametrize the radar scattering model. We then compute the circularly polarized bistatic scattering cross sections corresponding to each DDM bin according to their scattering and azimuth angles, followed by filling in each DDM bin with the points on the ground that contribute to that bin. We compare the DDM thus simulated with the observed DDM, after making proper adjustments to measurement units and applying the required calibration parameters. Results of this analysis and comparison will be shown at the presentation. In the future, we will use the calibrated data from GRIBSAR, proposed validated forward model, and well-tested inverse scattering algorithms [21] to retrieve soil moisture at Tonzi ranch and other locations where data might be available. Once demonstrated, the method will be used more broadly for soil moisture retrievals from CYGNSS data. in IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 1, pp. 71-84, Jan. 2011. [13] Ulaby, F.T., Long, D.G., Blackwell, W.J., Elachi, C., Fung, A.K., Ruf, C., Sarabandi, K., Zebker, H.A. and Van Zyl, J., 2014. Microwave radar and radiometric remote sensing (Vol. 4, No. 5, p. 6). Ann Arbor: University of Michigan Press. [14] J. D. Ouellette et al., "A Simulation Study of Compact Polarimetry for Radar Retrieval of Soil Moisture," in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 9, pp. 5966-5973, Sept. 2014. [15] Ishimaru, Akira. Wave propagation and scattering in random media. Vol. 2. New York: Academic press, 1978. [16] L. Tsang et al., "Active and Passive Vegetated Surface Models With Rough Surface Boundary Conditions From NMM3D," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 6, no. 3, pp. 1698-1709, June 2013. [17] A. Tabatabaeenejad and M. Moghaddam, "Bistatic scattering from three-dimensional layered rough surfaces," in IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no. 8, pp. 2102-2114, Aug. 2006. [18] Tsang, Leung, et al. Scattering of electromagnetic waves, numerical simulations. Vol. 25. John Wiley & Sons, 2004. [19] X. Duan and M. Moghaddam, "Stabilized extended boundary condition method for 3D electromagnetic scattering from arbitrary random rough surfaces," 2010 IEEE Antennas and Propagation Society International Symposium, Toronto, ON, 2010, pp. 1-4. [20] M. Moghaddam et al., "A Wireless Soil Moisture Smart Sensor Web Using Physics-Based Optimal Control: Concept and Initial Demonstrations," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 3, no. 4, pp.522-535, Dec.2010. doi: 10.1109/JSTARS.2010.2052918 [21] Tabatabaeenejad, A., and M. Moghaddam, "Inversion of subsurface properties of layered dielectric structures with random rough interfaces using the method of simulated annealing," IEEE Trans. Geosci. Remote Sensing, 47(7), 2035- 2046, 2009 6. REFERENCES [1] D. Entekhabi et al., "The Soil Moisture Active Passive (SMAP) Mission," in Proceedings of the IEEE, vol. 98, no. 5, pp. 704- 716, May 2010. [2] Cyclone Global Navigation Satellite System (CYGNSS) Handbook, University of Michigan, Ann Arbor, MI, 2016. pp. Conference 7480-7482. [3] A. Azemati, M. Moghaddam and A. Bhat, "Relationship Between Bistatic Radar Scattering Cross Sections and GPS Reflectometry Delay-Doppler Maps Over Vegetated Land in Support of Soil Moisture Retrieval," IGARSS 2018 - 2018 International Geoscience and Remote Sensing IEEE Symposium, Valencia, 2018, doi: 10.1109/IGARSS.2018.8517345 [4] A. Azemati and M. Moghaddam, "Circular-linear polarization signatures in bistatic scattering models applied to signals of opportunity," 2017 on Electromagnetics in Advanced Applications (ICEAA), Verona, 2017, pp. 1825-1827. International [5] A. Azemati and M. Moghaddam, "Modeling and analysis of bistatic scattering from forests in support of soil moisture retrieval," 2017 IEEE International Symposium on Antennas and Propagation & USNC/URSI National Radio Science Meeting, San Diego, CA, 2017, pp. 1833-1834. [6] N. Pierdicca, L. Pulvirenti, F. Ticconi and M. Brogioni, "Radar Bistatic Configurations for Soil Moisture Retrieval: A Simulation Study," in IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 10, pp. 3252-3264, Oct. 2008. [7] N. Rodriguez-Alvarez et al., "Soil Moisture Retrieval Using GNSS-R Techniques: Experimental Results Over a Bare Soil Field," in IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 11, pp. 3616-3624, Nov. 2009. [8] A. Camps et al., "Sensitivity of GNSS-R Spaceborne Observations to Soil Moisture and Vegetation," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 10, pp. 4730-4742, Oct. 2016. [9] R. Shah, C. Zuffada, C. Chew, M. Lavalle, X. Xu and A. Azemati, "Modeling bistatic scattering signatures from sources of opportunity in P-Ka bands," 2017 International Conference on Electromagnetics in Advanced Applications (ICEAA), Verona, 2017, pp. 1684-1687. [10] C. C. Chew, E. E. Small, K. M. Larson and V. U. Zavorotny, "Effects of Near-Surface Soil Moisture on GPS SNR Data: Development of a Retrieval Algorithm for Soil Moisture," in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 1, pp. 537-543, Jan. 2014. [11] K. M. Larson, J. J. Braun, E. E. Small, V. U. Zavorotny, E. D. Gutmann and A. L. Bilich, "GPS Multipath and Its Relation to Near-Surface Soil Moisture Content," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 3, no. 1, pp. 91-99, March 2010. [12] N. Rodriguez-Alvarez et al., "Land Geophysical Parameters Retrieval Using the Interference Pattern GNSS-R Technique,"
1709.03229	3	1709	2019-06-21T20:54:34	On the wave propagation analysis and supratransmission prediction of a metastable modular metastructure for non-reciprocal energy transmission	[ "physics.app-ph" ]	In this research, we investigate in-depth the nonlinear energy transmission phenomenon in a metastable modular metastructure and develop efficient tools for the design of such systems. Previous studies on a one-dimensional (1D) reconfigurable metastable modular chain uncover that when the driving frequency is within the stopband of the periodic structure, there exists a threshold input amplitude, beyond which sudden increase in the energy transmission can be observed. This onset of transmission is caused by nonlinear instability and is known as supratransmission. Due to spatial asymmetry of strategically configured constituents, such transmission thresholds could shift considerably when the structure is excited from different ends and therefore enabling the non-reciprocal energy transmission. This one-way propagation characteristic can be adaptable via reconfiguring the metastable modular system. In this new study, we build upon these findings and advance the state of the art by (a) exploring the different mechanisms that are able to activate the onset of supratransmission and their implications on wave energy transmission potential, and (b) developing an effective design tool - a localized nonlinear-linear model combined with harmonic balancing and transfer matrix analyses to analytically and efficiently predict the critical threshold amplitude of the metastable modular chain. These investigations provide important new understandings of the rich and intricate dynamics achievable by nonlinearity, asymmetry, and metastability, and create opportunities to accomplish adaptable non-reciprocal wave energy transmission	physics.app-ph	physics	On the wave propagation analysis and supratransmission prediction of a metastable modular metastructure for non-reciprocal energy transmission Z. Wu, and K.W. Wang Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109-2125 Corresponding author, email: [email protected] Abstra In this research, we investigate in-depth the nonlinear energy transmission phenomenon in a metastable modular metastructure and develop efficient tools for the design of such systems. Previous studies on a one-dimensional (1D) reconfigurable metastable modular chain uncover that when the driving frequency is within the stopband of the periodic structure, there exists a threshold input amplitude, beyond which sudden increase in the energy transmission can be observed. This onset of transmission is caused by nonlinear instability and is known as supratransmission. Due to spatial asymmetry of strategically configured constituents, such transmission thresholds could shift considerably when the structure is excited from different ends and therefore enabling the non-reciprocal energy transmission. This one-way propagation characteristic can be adaptable via reconfiguring the metastable modular system. In this new study, we build upon these findings and advance the state of the art by (a) exploring the different mechanisms that are able to activate the onset of supratransmission and their implications on wave energy transmission potential, and (b) developing an effective design tool - a localized nonlinear-linear model combined with harmonic balancing and transfer matrix analyses to analytically and efficiently predict the critical threshold amplitude of the metastable modular chain. These investigations provide important new understandings of the rich and intricate dynamics achievable by nonlinearity, asymmetry, and metastability, and create opportunities to accomplish adaptable non-reciprocal wave energy transmission. 1 Introduction Non-reciprocity of wave propagation refers to the unidirectional wave transmission between two points in space and it necessitates breaking the time-reversal symmetry of the system [1] [2]. Motivated by one- way flow of electrical energy using diodes, in recent years, significant research attention have been devoted to realize non-reciprocal wave propagation in other energy forms such as acoustic [3] [4] [5] [6] [7] [8] [9] [10] [11], elastic [12] [13], thermal [14] [15] and optical energy [16] [17]. These contributions can be loosely categorized into two domains: using linear systems with symmetry breaking mechanism [5] [10] [12] [17] or utilizing nonlinear systems [3] [4] [6] [7] [8] [9] [11] [13] [14] [15] [16]. While 1 interesting, many of these studies only focused on designing systems that exhibit unidirectional energy transmission and limited attention has been paid to systems that are capable of attaining on-demand tuning of such non-reciprocal wave propagations, which are beneficial in many engineering applications [6] [18]. Boechler et al demonstrated tunable rectification factor by adjusting the static load of a defected granular chain [3]. Chen and Wu presented a tunable topological insulator using a 2D phononic crystal through both analysis and numerical experiments [19]. Recently, Raney et al also realized a tunable soft mechanical diode capable of one-way solitary wave propagation using elastomeric bistable lattices [20]. Amongst these studies [3] [19] [20], the tunable non-reciprocal wave propagation was achieved by adjusting global topology or external constraint of the proposed periodic structure. On the other hand, individual reconfiguration of the numerous repeating internal modules/components for a fixed global confinement, which can significantly expand the adaptation space and greatly benefit adaptive wave control, are not allowed with these approaches. Recently, we proposed a novel approach to accomplish adaptable non-reciprocal wave propagation for a broad frequency range with large adaptation space [21]. This was achieved by exploiting the nonlinear supratransmission property of a spatially asymmetric reconfigurable metastable modular metastructure, systems that exhibit coexisting metastable states for the same topology [21]. Supratransmission refers to a wave propagation phenomenon that is unique to nonlinear periodic structures. For spatial periodic linear systems, there exhibits band-pass characteristics of energy transmission owning to the existence of bandgaps in the frequency spectrum. When excitation frequency is outside the bandgaps, vibrational energy is able to transmit through the structure with propagation speed determined by the input frequency. Whereas when frequency is inside the bandgaps, structure will scatter or absorb the energy and prohibit the transmission in all or specific directions [22] [23]. Such characteristic has been widely used in the development of mechanical filters [24] [25]. Nonlinear supratransmission, on the other hand, describes a sudden transmission of wave energy in periodic nonlinear structure subjected to continuous boundary driving with input frequency inside the linear bandgaps [26]. For small excitation amplitude, similar to a linear system, injected energy is spatially attenuated away from the input and does not propagate through the chain. When input exceeds a threshold amplitude, even though the excitation frequency is still inside the linear bandgaps, energy transmission becomes possible due to nonlinear instability [27]. This intriguing phenomenon has been observed and extensively investigated in many discrete nonlinear systems, such as sine-Gordon and Klein-Gordon chains of coupled oscillators [26] [28] [29], Fermi-Pasta- Ulam (FPU) nonlinear chains [30] , and periodic bistable chains [31] [32]. Due to the symmetric nature of the systems in these studies, non-reciprocal wave energy transmission is not possible. To facilitate unidirectional energy transmission, we proposed a metastable module as the building block [21]. The unit 2 module which consists of a bistable spring connected in series with a linear spring, hence the spatial symmetry is broken on both the module level and system level [33] [34]. We provided experimental and numerical evidences on non-reciprocal wave propagation using a metastable chain realized by connecting the modules in series and we discovered that due to spatial asymmetry, this non-reciprocal energy propagation is facilitated by triggering the onset of nonlinear supratransmission at different input amplitude levels when metastructure is excited in opposite directions [21]. In addition to non-reciprocal wave propagation, endowed with metastability, we discovered that the proposed metastructure can exhibit on-demand tuning and adaptation of non-reciprocal wave propagation characteristics by switching amongst metastable states for a large parameter space. Our prior investigation [21] presented a novel approach to achieve adaptable non-reciprocal wave propagation and showed great potential, however deep understanding of the proposed system is still lacking and an efficient tool is needed to design complex systems consists of the proposed modules for non-reciprocal wave energy propagation and adaptation. To advance the state of the art, building upon the authors' foundational idea, the objectives of this research are two-folds. First, through in-depth numerical studies, we seek to explore the different mechanisms that are able to activate the onset of supratransmission and their implications on wave energy transmission potential. Second, we aim to develop effective analytical tools to analyze and synthesize systems capable of attaining non-reciprocal wave propagation for desired operation ranges. In doing so, this investigation presents an opportunity to thoroughly explore the rich dynamics of the proposed system and provides design guidelines for creating real-world engineering systems with desired non-reciprocal characteristics. To present the approach and outcomes, this paper is organized as follows. In Sec II, the equations describing the nonlinear dynamics of the metastructure are presented, followed by dispersion analysis of the linearized system. In Sec. III, we present in detail different route to supratransmission the proposed metastructure imparts. In Sec IV, we discuss a localized nonlinear-linear model that can analytically predict the critical threshold amplitude for supratransmission and exemplify non-reciprocal characteristics of the proposed system. Lastly in Sec. V, influence of some key parameters on the threshold amplitude to trigger supratransmission is explored. 2 Mathematical model and governing equations Figure 1 depicts a schematic of discrete lattice representation of 𝑁 identical metastable modules connected in series. Each metastable module, highlighted with red dashed box, consists of two masses 𝑚1 3 and 𝑚2 (represented with orange and yellow dots) coupled via a linear constituents represented with coil springs, highlighted with blue dashed box; the modules are inter-connected by bistable springs, exemplified with buckled beams, green dashed box. Free length configuration of the structure 𝐿𝑓𝑟𝑒𝑒 is defined to be the zero force position when all the bistable elements are buckled to the left and the global displacement 𝑧 is defined as the additional deformation of the structure starting from the free length 𝐿𝑓𝑟𝑒𝑒, Figure 1(a). Without loss of generality, the bistable and linear restoring forces are assumed of the form, 𝐹𝑁𝐿 = 𝑘1𝑥 + 𝑘2𝑥2 + 𝑘3𝑥3 and 𝐹𝐿 = 𝑘𝐿𝑦, where 𝑥 and 𝑦 are the deformations of bistable and linear springs respectively. The total potential energy of the metastable chain for a fixed global displacement z measured from its free length 𝐿𝑓𝑟𝑒𝑒 can be expressed as: + 𝑘2 3 2 (𝑥[𝑖+1]1 − 𝑥[𝑖]2) 2 (𝑥[𝑖]2 − 𝑥[𝑖]1) 𝑈 = ∑ 𝑘𝐿 𝑁−1 [ 𝑖=1 2 3 (𝑥[𝑖+1]1 − 𝑥[𝑖]2) + 𝑘3 4 4 (𝑥[𝑖+1]1 − 𝑥[𝑖]2) ] + + 𝑘1 2 3 + 𝑘3 𝑥[1]1 4 𝑘𝐿 2 2 (𝑧 − 𝑥[𝑁]1) + 𝑘1 2 2 + 𝑘2 𝑥[1]1 3 4 𝑥[1]1 (1) which is a function of internal mass displacements 𝑥[𝑖]1, 𝑥[𝑖]2, where subscript 𝑖 in square bracket refers to the 𝑖𝑡ℎ metastable module. All internal displacements 𝑥[𝑖]1 and 𝑥[𝑖]2 are measured from the individual positions of the free length configuration. For a fixed global displacement z, the equilibrium positions of metastructure satisfies 𝜕𝑈 𝜕𝑥[𝑖]1 ⁄ = 0 and 𝜕𝑈 𝜕𝑥[𝑖]2 ⁄ = 0 under the constraint that ∑ 𝑁−1 𝑖=1 (𝑥[𝑖]1 + 𝑥[𝑖]2) + 𝑥[N]1 = 𝑧. Based on the minimum potential energy principle [35] [36], metastable states of the chain satisfy 𝜕2𝑈 𝜕𝑥[𝑖]𝑘𝜕𝑥[𝑗]𝑙 > 0, i.e. the Hessian matrix of the potential is positive definite. ⁄ 4 Figure 1. (a) Schematic and discrete mass spring representation of an 𝑁 metastable module assembled in series. Metastable module, bistable and linear constituents are highlighted with dashed boxes. (c) Force displacement profile of a 10 module metastructure assembly. (b) and (d) Linearized periodic structure for dispersion analysis and corresponding band structure of the metastructure with free length 𝑧 = 0 for configuration A in (c). In general, for a fixed global displacement 𝑧 , a chain of 𝑁 metastable modules can have up to 2𝑁 metastable states (internal configurations), Figure 1(c) [33] [34]. Starting from one of the metastable states, equations of motion for the 𝑖𝑡ℎ metastable module can be expressed as: 𝑚1𝑥[𝑖]1 + 𝐹𝑁𝐿(𝑥[𝑖]1 − 𝑥[𝑖−1]2) + 𝑐(𝑥[𝑖]1 − 𝑥[𝑖−1]2) + 𝑘𝐿(𝑥[𝑖]1 − 𝑥[𝑖]2) + ζ(𝑥[𝑖]1 − 𝑥[𝑖]2) = 0 𝑚2𝑥[𝑖]2 + 𝐹𝑁𝐿(𝑥[𝑖]2 − 𝑥[𝑖+1]1) + 𝑐(𝑥[𝑖]2 − 𝑥[𝑖+1]1) + 𝑘𝐿(𝑥[𝑖]2 − 𝑥[𝑖]1) + ζ(𝑥[𝑖]2 − 𝑥[𝑖]1) = 0 (2a) (2b) where c represents the damping coefficients. Eq. (2a) is applicable to ∀𝑖 = 2 to 𝑁 − 1 and Eq. (2b) is applicable to ∀𝑖 = 1 to 𝑁 − 1. To investigate the non-reciprocal effect, two actuation scenarios are considered: one is forward actuation with input 𝑥𝑖𝑛 on the left side of the lattice chain and the other is backward actuation with input 𝑥𝑖𝑛 on the right hand side of the chain. Discrete representations of the excitation scenarios are depicted in Figure 2. For both scenarios, displacement input 𝑥𝑖𝑛 is directly applied to the mass next to the boundary of the chain and output signal 𝑥𝑜𝑢𝑡 is measured one unit (two degrees of freedom) away from the other side of 5 the boundary. Owning to asymmetry of the metastable module, periodic repeating unit for the two excitation scenarios are inherently different, highlighted with red dashed boxes respectively, Figure 2. With fixed boundary conditions and displacement input 𝑥𝑖𝑛, for forward actuation, equations of motion of the last mass in the chain can be modified as 𝑚1𝑥[𝑁]1 + 𝐹𝑁𝐿(𝑥[𝑁]1 − 𝑥[𝑁−1]2) + c(𝑥[𝑁]1 − 𝑥[𝑁−1]2) + 𝑘𝐿(𝑥[𝑁]1 − 𝑧) + ζ(𝑥[𝑁]1 − 𝑧) = 0 (3a) and displacement of the first mass is prescribed by 𝑥𝑖𝑛. Similarly, for backward actuation, equations of motion for first mass can be expressed as: 𝑚1𝑥[1]1 + 𝐹𝑁𝐿(𝑥[1]1) + c(𝑥[1]1) + 𝑘𝐿(𝑥[1]1 − 𝑥[1]2) + ζ(𝑥[1]1 − 𝑥[1]2) = 0 (3b) To determine the dispersion relation, we first linearize the equations of motion about its metastable state with damping coefficient 𝑐 = 0 and the band structure is determined by modeling a repeating periodic unit cell of an unforced, infinite chain, Figure 1(b). For the diatomic chain depicted in Figure 1(a), the periodic unit cell is the same as a linearized metastable module, Figure 1(b), and the linearized equation can be written as: 𝑚1𝜁𝑖 + 𝑘𝑁𝐿(𝜁𝑖 − 𝜂𝑖−1) + 𝑘𝐿(𝜁𝑖 − 𝜂𝑖) = 0 𝑚2𝜂𝑖 + 𝑘𝑁𝐿(𝜂𝑖 − 𝜁𝑖+1) + 𝑘𝐿(𝜂𝑖 − 𝜁𝑖) = 0 where 𝜁𝑖 and 𝜂𝑖 are small perturbation of mass 𝑚1 and 𝑚2 around its initial positions and 𝑘𝑁𝐿 is the corresponding linearized stiffness of the bistable spring depending on the starting metastable states. Assuming solutions in the form of a traveling wave, i.e, 𝜁𝑖 = 𝐴𝑒𝑥𝑝[𝑗(𝜔𝑡 − 𝑘𝑖𝐿)] and 𝜂𝑖 = (4b) (4a) 𝐵𝑒𝑥𝑝[𝑗(𝜔𝑡 − 𝑘(𝑖 + 1)𝐿)], where k is the wave number and L is unit length, the model is reduced to a standard eigenvalue problem: (𝑘𝐿 + 𝑘𝑁𝐿)/𝑚1 −(𝑘𝑁𝐿 + 𝑘𝐿𝑒−𝑗𝑘𝐿)/𝑚1 [ −(𝑘𝑁𝐿 + 𝑘𝐿𝑒𝑗𝑘𝐿)/𝑚2 The band structure can then be determined by sweeping the wave number 𝑘 from 0/𝐿 to 𝜋/𝐿. It can be (𝑘𝐿 + 𝑘𝑁𝐿)/𝑚2 (5) ] [ ] = 𝜔2 [ 𝐴 𝐵 𝐴 𝐵 ] determined that the bandgaps are within [√𝜔1, √𝜔2] and [√𝜔3, ∞) , where 𝜔1 = (𝑘𝐿+𝑘 𝑁𝐿)(𝑚1+𝑚2) 2𝑚1𝑚2 − 2 √(𝑘𝐿+𝑘 𝑁𝐿) (𝑚1+m2 )2−16𝑘𝐿𝑘 𝑁𝐿𝑚1𝑚2 2𝑚1𝑚2 , 𝜔2 = (𝑘𝐿+𝑘 𝑁𝐿)(𝑚1+𝑚2) 2𝑚1𝑚2 + 2 √(𝑘𝐿+𝑘 𝑁𝐿) (𝑚1+m2 )2−16𝑘𝐿𝑘 𝑁𝐿𝑚1𝑚2 2𝑚1𝑚2 , and 𝜔3 = (𝑘𝐿+𝑘 𝑁𝐿)(𝑚1+𝑚2) 𝑚1𝑚2 . For demonstration purposes and without loss of generality, parameters used in the analysis are chosen to be of arbitrary unit. With 𝑘1 = 2, 𝑘2 = −3, 𝑘3 = 1, 𝑘𝐿 = 0.2, 𝑚1 = 1and 𝑚2 = 1, band structures of the diatomic metastable chain in free length and zero external force configuration depicted by point A is shown in Figure 1(c). Two bandgaps of the diatomic chain are [0.633, 2] and [2.098, ∞). 6 Figure 2. Conceptual diagrams of metastable module assembled in series under forward (excitation from left) and backward (excitation from right) actuations. For both scenarios, displacement input 𝑥𝑖𝑛 is directly applied to the mass next to the boundary of the chain, and output signal 𝑥𝑜𝑢𝑡 is measured one module away from the boundary. Repeating unit for both excitation scenarios are highlighted with red dashed box. The first and last units are highlighted with green and blue dashed boxes, respectively. 3 In-depth understanding of route to supratransmission of the metastructure -- A numerical study With the same system parameters used to determine the bandgaps in Sec. 2.1, 𝑘1 = 2, 𝑘2 = −3, 𝑘3 = 1, 𝑘𝐿 = 0.2 and masses 𝑚1 = 𝑚2 = 1, wave propagation characteristics are explored using a 10 module metastructure over a wide spectrum of input parameters starting from free length configuration, i.e. 𝑧 = 0. Given the system parameters, distance between the two equilibrium positions for the bistable constituents is 2. Small damping coefficients 𝑐 = 0.001 is applied between lattices and input 𝑥𝑖𝑛 = 𝑝𝑠𝑖𝑛(𝜔𝑡) is prescribed to one end of the structure depending on excitation scenario with numerical simulations running for sufficiently long time (30000 periods where one period is defined as 2𝜋/√𝑘1 𝑚1⁄ ) to reach steady state. Small damping coefficients are chosen to limit the influence of energy dissipation on wave propagation phenomenon. Figure 3 depicts the contour plot of the transmittance ratio (TR) of the metastructure as input frequency and amplitude varies for forward actuation scenario, Figure 2. The transmittance ratio (TR) is defined as the ratio of output RMS displacement over input RMS displacement 𝑇𝑅 = 𝑅𝑀𝑆(𝑥𝑜𝑢𝑡)/𝑅𝑀𝑆(𝑥𝑖𝑛). Hence, a higher transmittance ratio indicates more energy is transmitted through the metastructure. Contour plot is in log scale with brighter region corresponds to larger TR values. As shown in Figure 3, for input frequency inside the passband (PB) of the structure, since wave energy is able to propagate through the chain [23], transmittance ratio (TR) is always large. When excitation frequency is inside the stopband of the structure, an amplitude dependent wave transmission characteristic can be observed due to supratransmission [21] [26] [30] [37]. Upon close inspection, three distinct regions, labeled with I, II and III in Figure 3, with different input amplitude dependency can be identified. In this section, we will explore in detail different mechanisms that trigger the supratransmission phenomenon, which enable the energy transmission inside the bandgap. 7 Figure 3. Contour plot on transmittance ratio (vs. input frequency and amplitude) for forward actuation from free length configuration. Passbands (PB) from dispersion analysis are within (0, 0.632) and (2, 2.098). Passbands are bounded by the red dash-dotted lines. For frequency inside the bandgaps, three regions, each with unique energy transmission characteristics depending on input frequency and amplitude can be identified, labeled with I, II and III. Representative frequency for each region is labeled with (a), (b) and (c) with yellow dashed lines, respectively. 3.1 Region I To explore the characteristic dynamics in region I, input frequency 𝜔 = 1 labeled with (a) in Figure 3 is chosen while excitation amplitude varies from 0.1 to 3 for forward actuation, Figure 2. From previous study [21], we find that energy transmission is closely related to the dynamics of the first unit near the excitation; therefore, Figure 4 depicts both the displacement amplitudes 𝑌 of the last 100 excitation periods of the first bistable constituent and corresponding transmittance ratio as input amplitude increases. The displacement amplitude is defined to be the peak value of relative displacement between 𝑚1 and 𝑚2 at steady state and is chosen to represent the dynamics of the first metastable unit. If the steady state response of the bistable constituent is periodic oscillation, displacement amplitude is represented with a single point in Figure 4 for a fixed excitation level, whereas for aperiodic or chaotic oscillation, amplitude at steady state covers a wide range of magnitude over the last 100 excitation periods and therefore is represented by a line. As input amplitude varies, 6 input levels 𝑝 = [0.15, 0.45, 0.8, 1.5, 1.77, 1.87], labeled A through F, are carefully inspected. Figure 5 depicts the phase diagram of the bistable 8 displacement 𝑦 and corresponding bistable velocity 𝑣 of the first unit as well as the FFT of bistable velocity 𝑣. Bistable displacement 𝑦 is defined to be the relative displacement between two masses at steady state and likewise bistable velocity 𝑣 is the relative velocity. Black dots represent the equilibrium positions of the bistable constituents and red dots correspond to the stroboscopic map of the first bistable constituent. With small excitation level 𝑝 = 0.15 , point A in Figure 4, the first bistable constituent undergoes intrawell oscillation (small amplitude oscillation around one of the equilibrium positions) with dominant frequency the same as the input frequency, Figure 5(A). Since input frequency is inside the stopband, transmittance ratio is small as expected, Figure 4(b), indicating that energy does not propagate through the chain. As excitation increases to 𝑝 = 0.45, point B in Figure 4, displacement amplitude bifurcates into two magnitudes. FFT of bistable velocity in Figure 5(B) reveals that the vibration exhibits characteristics of superharmonic oscillation with dominant frequency twice as much as the driving frequency. Since all frequency spectrums with dominant magnitude are in the stopband, the TR remains to be low. Further increasing excitation level, steady state magnitude of the first bistable element becomes aperiodic. For instance, at 𝑝 = 0.8, intrawell aperiodic oscillation occurs with frequency contents covering a broader spectrum and some of the frequency content now are inside the passband, Figure 5(C) and hence transmittance ratio starts to increase to 0.04, indicating that vibration energy starts to propagate through the chain. As input level increases to approximately 1.3, displacement amplitude suddenly jumps and the vibration becomes chaotic interwell oscillation (large amplitude oscillation encompassing both equilibrium positions), point D and F in Figure 4. FFT of corresponding bistable velocity in Figure 5(D) and (F) also illustrates that vibration energy of the constituent is diffused to a broader frequency spectrum with more frequency content inside the passband. Hence, such interwell chaotic oscillation leads to a noticeable increase in the transmittance ratio to 0.617 and 0.845 respectively for D and F, Figure 4(b). Lastly, at excitation 𝑝 = 1.77 , displacement amplitude encompass a shorter range near one of its equilibrium positions at y = 2 (the other equilibrium position is at y = 0), point E in Figure 4(a), indicating it undergoes intrawell oscillation. Phase diagram in Figure 5(E) also corroborates this observation. From the FFT of bistable velocity, Figure 5(E), it is determined that bistable constituent of the first unit undergoes a quasi-periodic intrawell oscillation, with more energy localized on frequency contents in the stopband compare to chaotic interwell oscillations, point D and F in Figure 4. Therefore, transmittance ratio at this excitation level drops correspondingly, point E in Figure 4(b). These observations illustrate that the transmittance ratio is closely related to the dynamics of first unit. It is generally high when the first bistable constituent undergoes interwell or intrawell aperiodic/chaotic oscillation. Furthermore, for frequency range inside region I, supratransmission is shown to be facilitated through the nonlinear 9 instability by transitioning from superharmonic vibration to aperiodic intrawell oscillation of the bistable constituent and transmittance ratio grows progressively as input amplitude increases. Figure 4. Displacement amplitude of the first bistable constituent and corresponding transmittance ratio as input amplitude increases for forward excitation scenarios with input frequency 𝜔 = 1 . Displacement amplitude is determined to be the peak value of relative displacement between the two oscillators at steady state. A through F corresponds to input amplitude 𝑝 = [0.15, 0.45, 0.8, 1.5, 1.77, 1.87]. 10 Figure 5. Phase diagram of the bistable constituent of the first unit under forward excitation, Figure 2, and corresponding FFT of the bistable velocity for different input excitation levels. (A) through (F) matches the same labels in Figure 4. Black dots represent the equilibrium positions of the bistable constituents and red dots correspond to the stroboscopic map of the first bistable constituent. 𝜔 = 1 denotes the input frequency. 11 Figure 5. (Cont'd) Phase diagram of the bistable constituents of the first unit under forward excitation and corresponding FFT of the bistable velocity for different input excitation levels. 3.2 Region II Similarly, to explore the characteristic dynamics in region II, input frequency 𝜔 = 1.6 labeled with (b) in Figure 3 is chosen while excitation amplitude varies from 0.1 to 3 for forward actuation, Figure 2. Figure 6 depicts the displacement amplitude of the first bistable unit and corresponding transmittance ratio as input amplitude increases. As input amplitude varies, representative input levels 𝑝 = [0.5,1.27,1.46,1.85], are selected and labeled A through D. Figure 7 depicts the phase diagram of the bistable constituent for the first unit and FFT of corresponding bistable velocity. Red dots correspond to the stroboscopic map of the first bistable constituent. Comparable to cases in region A, when excitation level is small, transmittance ratio is low, for instance 𝑝 = 0.5 point A in Figure 6. As shown in Figure 6(A), the first bistable element undergoes a small amplitude periodic intrawell oscillation with dominant frequency the same as the input frequency and corresponding transmittance ratio is only 0.01. As input amplitude increases to 𝑝 = 1.27 , vibration amplitude of the first bistable constituent increases significantly and the resultant vibration is characterized as chaotic interwell oscillation, Figure 7(B) and (D). Correspondingly, more than two orders 12 of magnitude increases in transmittance ratio is observed as dynamical characteristics of the first bistable constituent suddenly changes. Transitioning from intrawell to interwell oscillation for a bistable element as input amplitude increases is known to be caused by a saddle node bifurcation [38] [39] [40], therefore, for frequency range inside region II, supratransmission is triggered through the nonlinear instability of a saddle node bifurcation. Note that point C in Figure 6 corresponds to a near periodic large amplitude interwell oscillation yet the transmittance ratio is lower compared to the chaotic interwell oscillation case point B or D. This is because when first bistable element undergoes chaotic oscillation, it converts the frequency spectrum from the stopband to the passband more effectively, leading to more energy propagation through the chain and hence resulting in a higher TR [21] [32]. Figure 6. Displacement amplitude of the first bistable constituent and corresponding transmittance ratio as input amplitude increases for forward excitation scenarios with input frequency 𝜔 = 1.6. Displacement amplitude is determined to be the relative displacement amplitude between the two moving masses at steady state. A through D corresponds to input amplitude 𝑝 = [0.5, 1.27, 1.46, 1.85]. 13 Figure 7. Phase diagram of the bistable constituents of the first unit under forward excitation Figure 2, and corresponding FFT of the bistable velocity for different input excitation levels. (A) through (D) matches the same labels in Figure 6. Black dots represent the equilibrium positions of the bistable constituents and red dots correspond to the stroboscopic map of the first bistable constituent. 𝜔 = 1.6 denotes the input frequency. 14 3.3 Region III Lastly, when excitation frequency is inside region III, we explored the characteristic dynamics with input frequency 𝜔 = 2.2, labeled (c) in Figure 3. As indicated by the complicated TR pattern in Figure 8(b), for parameters in region III, we expect the dynamics of the first bistable element to be more intricate. Depicted in Figure 8(a), as input amplitude increases, dynamics of the first bistable element constantly transitions amongst periodic intrawell (point A and F), quasi-periodic intrawell (point B), aperiodic intrawell (point C), chaotic interwell (point D and G) and even subharmonic intrawell oscillation (point E). As a result, transmittance ratio frequently fluctuates up and down and is generally low for periodic intrawell and high for chaotic/aperiodic interwell/intrawell oscillations. In contrast to the previous two cases, for input frequencies in this region, there is no clear threshold on input amplitude beyond which vibration energy always propagates through the chain. However, when energy does propagate through the chain, supratransmission is still enabled via nonlinear instability, exemplified by the sudden change of vibration amplitude of the bistable element. 15 Figure 8. Displacement amplitude of the first bistable constituent and corresponding transmittance ratio as input amplitude increases for forward excitation scenarios with input frequency ω=2.2. Dispalcement amplitude is determined to be the relative displacement amplitude between the two moving masses at steady state. A through G corresponds to input amplitude 𝑝 = [0.1, 0.2, 0.3, 0.75, 1.38, 1.7 1.85]. 16 Figure 9. Phase diagram of the bistable constituents of the first unit under forward excitation, Figure 2, and corresponding FFT of the bistable velocity for different input excitation levels. (A) through (G) matches the same labels in Figure 8. Black dots represent the equilibrium positions of the bistable constituents and red dots correspond to the stroboscopic map of the first bistable constituent. 𝜔 = 2.2 denotes the input frequency. Figure 9. (Cont'd) Phase diagram of the bistable constituent of the first unit under forward excitation Figure 2, and corresponding FFT of the bistable velocity for different input excitation levels. 4 Predicting the onset of supratransmission and region of non-reciprocal transmission -- An analytical approach To facilitate the design and synthesis of metastructures with desired non-reciprocal wave energy propagation characteristics, in this study, we seek to develop analytical tools that is able to effectively and efficiently predict the region of non-reciprocal transmission of the proposed metastructure. 17 The proposed metastructure is a multi-degree of freedom highly nonlinear systems that are known to be challenging and often times impossible to directly conduct analytical study on. On the other hand, from investigations discussed in Sec. 3, we noticeably observed the correlation between dynamics of the first bistable element with respect to the transmission ratio (TR) under forward actuation. This is mainly for two reasons: first is the weak coupling between the individual units (to maintain adaptable and metastable characteristics, strong coupling is not desired [33] [34]) and second is because the driving frequency is inside the stopband, vibration is prone to be localized closer to the excitation initially. Therefore, we propose to predict the onset of supratransmission of the fully nonlinear metsatructure by analyzing an equivalent but simplified mathematical model that only retains the nonlinear element of the first unit while linearizing the rest of the chain. With such a localized nonlinear-linear model, we are endowed with the opportunity to derive analytical solutions that approximate the supratransmission threshold of the fully nonlinear system, by using the transfer matrix method for the linear part [41] and interface with the harmonic balance method of the first nonlinear unit [40]. Similar concepts of treating the nonlinearity locally has been demonstrated to be effective in analyzing vibration characteristics for other nonlinear system [32] [42] [43]. Additionally, we noticed that when the first bistable element transitions from intrawell to interwell oscillation via a saddle node bifurcation (region II), a clear threshold on input amplitude can be identified beyond which drastic increase in transmittance ratio is observed. Such a rapid growth in transmittance ratio could potentially be used as a practical means for identifying regions of non- reciprocal response [21]. Therefore, in the following investigation discussed in this paper, we focus on analytically predicting the onset of supratransmission induced by the saddle-node bifurcation. 4.1 Forward actuation Figure 10(a) depicts the schematic of the localized nonlinear-linear model for forward actuation in which nonlinearity is only preserved in the first unit. Governing equations of the 𝑗𝑡ℎ unit, Figure 10(b) can be expressed as: 𝑚1𝑤 [𝑗+1]1 + 𝑘1(𝑤[𝑗+1]1 − 𝑤[𝑗]2) + 𝑘𝐿(𝑤[𝑗+1]1 − 𝑤[𝑗+1]2) = 0 𝑚2𝑤 [𝑗]2 + 𝑘1(𝑤[𝑗]2 − 𝑤[𝑗+1]1) + 𝑘𝐿(𝑤[𝑗]2 − 𝑤[𝑗]1) = 0 for 1 ≤ 𝑗 ≤ 𝑁 − 1, where 𝑘1 is the linearized stiffness of the bistable constituents and 𝑤[𝑗]1 and 𝑤[𝑗]2 are (6) (7) displacements from the new equilibrium positions for a fixed z and configuration. After nondimensionalization, the governing equations become ′′ 𝜇𝑤[𝑗+1]1 + 𝛽1(𝑤[𝑗+1]1 − 𝑤[𝑗]2) + (𝑤[𝑗+1]1 − 𝑤[𝑗+1]2) = 0 𝑤[𝑗]2 ′′ + 𝛽1(𝑤[𝑗]2 − 𝑤[𝑗+1]1) + (𝑤[𝑗]2 − 𝑤[𝑗]1) = 0 where the parameters are defined as 18 (8) (9) and the operator ( )′ represents a derivative with respect to nondimensional time 𝑡. ω2 = 𝑘L 𝑚2⁄ ; 𝜇 = 𝑚1 𝑚2⁄ ; 𝛽1 = 𝑘1 𝑘L⁄ Figure 10(b) shows a schematic representation of two adjacent units in the assembled periodic structure. On both side of each unit, there is a pair of force and displacement. It should be noted that in this case 𝑗 = 𝑤[𝑗+1]1 and 𝑢𝑅 𝑗+1 = 𝑤[𝑗+2]1. Balancing force at the left and right ends of the (𝑗 + 1)𝑡ℎ unit, we can 𝑢𝑅 derive the following relations: (𝑗+1) = 𝑢𝑅 −𝜔2+𝛽 1 𝛽 1 (𝑗+1) + 𝑢𝐿 𝜔2−1−𝛽 1 𝛽 1 (𝑗+1) = 𝑇11 𝑢𝐿 𝐹 𝑢𝐿 (𝑗+1) + 𝑇12 𝐹 𝐹𝐿 (𝑗+1) (𝑗+1) = 𝐹𝑅 𝜇𝜔4−ω2𝛽 1(1+𝜇) 𝛽 1 (𝑗+1) + 𝑢𝐿 1+𝜇+𝛽 −𝜇𝜔4+ω2(𝛽 𝛽 1 1𝜇)−𝛽 1 (𝑗+1) = 𝑇21 𝐹𝐿 𝐹 𝑢𝐿 (𝑗+1) + 𝑇22 𝐹 𝐹𝐿 (𝑗+1) From compatibility and force equilibrium between adjacent units we have 𝑢𝐿 (𝑗+1) = 𝑢𝑅 (𝑗) and 𝐹𝐿 (𝑖+1) = (𝑗) −𝐹𝑅 . The transfer matrix from 𝑗th unit to (𝑗 + 1)𝑡ℎ unit can then be derived as: [ (𝑗+1) 𝑢𝑅 (𝑗+1) ] = [ 𝐹𝑅 Moving along the finite chain from the second to last unit to the first unit, we can write 𝐹 𝐹 −𝑇12 𝑇11 𝐹 ] [ 𝐹 −𝑇22 𝑇21 (𝑗) 𝑢𝑅 (𝑗)] 𝐹𝑅 [ (𝑁−2) 𝑢𝑅 (𝑁−2)] = [ 𝐹𝑅 𝐹 −𝑇12 𝐹 𝑇11 𝐹 ] 𝐹 −𝑇22 𝑇21 𝑁−2 [ (1) 𝑢𝑅 (1)] = [ 𝐹𝑅 𝑀11 𝑀21 𝐹 𝐹 𝑀12 𝐹 ] [ 𝐹 𝑀22 (1) 𝑢𝑅 (1)] 𝐹𝑅 For last unit, Figure 10(c), with fixed boundary conditions, governing equations can be expressed as: 𝜇𝑤[𝑁]1 ′′ + 𝛽1(𝑤[𝑁]1 − 𝑤[𝑁−1]2) + (𝑤[𝑁]1) = 0 ′′ 𝑤[𝑁−1]2 + (𝑤[𝑁−1]2 − 𝑤[𝑁−1]1) + 𝛽1(𝑤[𝑁−1]2 − 𝑤[𝑁]1) = 0 (10) (11) (12) It is then easy to get that 𝐹𝐿 (𝑁−1) = 𝐶1𝑢𝐿 (𝑁−1) where 𝐶1 = (ω4 𝜇−𝜔2 (1+𝛽 ω4𝜇−𝜔2 (1+𝛽 1+𝛽 1𝜇)+𝛽 1) 1) (1+𝜇)+(2+𝛽 1) . Combining with Eq. (10), we can derive that (1) = 𝐶𝑢𝑅 𝐹𝑅 (1) = −𝐶1𝑀11 (𝐶1𝑀12 F −𝑀21 𝐹 F +𝑀22 F ) (1) 𝑢𝑅 (13) Lastly, governing equations for the first nonlinear unit, Figure 10(d) can be derived as: 1 𝜇(𝑥 + 𝑦 + 𝑥𝑖𝑛)′′ + 𝛽1𝑦+𝛽2𝑦2 + 𝛽3𝑦3 = 𝐹𝑅 (𝑥 + 𝑥𝑖𝑛)′′ + (𝑥) − (𝛽1𝑦+𝛽2𝑦2 + 𝛽3𝑦3) = 0 where 𝑥 and y are deformations of linear and nonlinear springs and 𝛽1 = 𝑘1 𝑘L⁄ 𝛽3 = 𝑘3 𝑘L⁄ . Without loss of generality, we assume 𝛽1 = 2𝛽3 and 𝛽2 = −3𝛽3 to realize a symmetric , 𝛽2 = 𝑘2 𝑘L⁄ , and (15) (14) bistable potential with distance between the two equilibria to be 2, similar to the potential profile used in previous numerical analysis. Assuming 1-term expansion with slowly varying coefficients 𝑥(𝑡) = 𝑎(𝑡) sin(𝜔𝑡) + 𝑏(𝑡) cos(𝜔𝑡) and 1 = 𝑥𝑖𝑛 + 𝑥(𝑡) + 𝑦(𝑡) and 𝑦(𝑡) = 𝑐(𝑡) sin(𝜔𝑡) + 𝑑(𝑡) cos(𝜔𝑡) + 𝑒(𝑡) , plugging equation (13), 𝑢𝑅 19 𝑥𝑖𝑛 = 𝑝𝑠𝑖𝑛(𝜔𝑡) into the equations of motion, eliminating higher order terms, and grouping the constant, sin(𝜔𝑡) and cos(𝜔𝑡) terms yields five equations for the coefficients, 𝑎(𝑡), 𝑏(𝑡), 𝑐(𝑡), d(𝑡) and e(𝑡) 2𝜔𝑏′ = −𝑝𝜔2 + (1 − 𝜔2)𝑎(𝑡) + (−2 − 3 4 𝑅𝑦 2 + 6𝑒(𝑡) − 3𝑒(𝑡)2) 𝛽3𝑐(𝑡) −2𝜔𝑎′ = (1 − 𝜔2)𝑏(𝑡) + (−2 − 3 4 𝑅𝑦 2 + 6𝑒(𝑡) − 3𝑒(𝑡)2) 𝛽3𝑑(𝑡) 2𝜇𝜔𝑏′ + 2𝜇𝜔𝑒′ = (−𝐶 − 𝜇𝜔2)𝑝 + (−𝐶 − 𝜇𝜔2)𝑎(𝑡) + (2𝛽3 + 3 4 𝑅𝑦 2 − 𝐶 − 𝜇𝜔2 − 6𝛽3𝑒(𝑡) + 3𝛽3𝑒(𝑡)2) 𝑐(𝑡) −2𝜇𝜔𝑐′ − 2𝜇𝜔𝑎′ = (−𝐶 − 𝜇𝜔2)𝑏(𝑡) + (2𝛽3 + 3 4 𝑅𝑦 2 − 𝐶 − 𝜇𝜔2 − 6𝛽3𝑒(𝑡) + 3𝛽3𝑒(𝑡)2) 𝑑(𝑡) 3 0 = −2𝛽3𝑒(𝑡) + 3𝛽3𝑒(𝑡)2 − 𝛽3𝑒(𝑡)3 + ( 2 − 3 2 2 𝑒(𝑡)) 𝛽3𝑅𝑦 where 𝑅𝑦 is the bistable displacement amplitude with 𝑅𝑦 2 = 𝑐(𝑡)2 + 𝑑(𝑡)2 Combining the five equations, a third order polynomial in terms of 𝑅 = 𝑅𝑦 𝜌3𝑅3 + 𝜌2𝑅2 + 𝜌𝑅 + 𝜌0 = 0 2 can be derived [40]: where 𝜌3 = 𝜌2 = 𝜌1 = 1 16 (−1+𝜔2)2 (225 𝐶2 𝛽3 2 + 450 𝐶 𝛽3 2 (−1 + (1 + 𝜇)𝜔2) + 225 𝛽3 2(−1 + (1 + 𝜇)𝜔2)2) 1 16 (−1 + 𝜔2)2 (−120 𝐶2𝛽3 − 240 𝐶2𝛽3 2 + 120 𝐶2𝛽3𝜔2 − 120 𝛽3 (−1 + (1 + 𝜇)𝜔2) (−𝜇 𝜔2 (−1 + 𝜔2) + 2 𝛽3 (−1 + (1 + 𝜇)𝜔2)) − 120 𝐶 𝛽3 (4 𝛽3 (−1 + (1 + 𝜇)𝜔2) − (−1 + 𝜔2) (−1 + (1 + 2 𝜇)𝜔2))) 1 16 (−1 + 𝜔2)2 (64 𝐶2 − 128 𝐶2𝜔2 + 64 𝐶2𝜔4 − 128 𝐶2𝛽3 + 128 𝐶2𝜔2𝛽3 + 64𝐶2𝛽3 2 2 + 16 (ω2 (−1 + 𝜔2)𝜇 − 2 𝛽3 (−1 + 𝜔2(1 + 𝜇))) + 64 𝐶 (−1 + 𝜔2 + 𝛽3) (−𝜔2 (−1 + 𝜔2) 𝜇 + 2 𝛽3 (−1 + 𝜔2 (1 + 𝜇)))) 𝜌0 = 1 16(−1+𝜔2)2 (−16C2𝑝2 − 32C𝑝2𝜇𝜔2 − 16𝑝2𝜇2𝜔4) This polynomial can be further simplified as 𝑅3 + 𝑎2𝑅2 + 𝑎1𝑅 + 𝑎0 = 0 with leading coefficients to be 1. Transmission threshold corresponding to a saddle-node bifurcation can then be determined by solving the equation 𝑎0 = −2 × (𝑃 + √𝑄3) (16) where 𝑃 = (𝑎2 3)/27 − (𝑎1𝑎2)/6 and Q = (𝑎2 2)/9 − 𝑎1/3 20 Figure 10.(a) Discrete chain of localized nonlinear-linear model for forward actuation; Only the first unit adjacent to input is nonlinear (b) Schematic of 𝑗𝑡ℎ and (𝑗 + 1)𝑡ℎ unit for transfer matrix analysis. Subscript in square bracket corresponds to the index of metastable module shown in Figure 2. (c) Schematic of the last unit for transfer matrix analysis. (d) Schematic of first unit for harmonic balancing analysis. 4.2 Backward actuation Following the same procedure, transfer matrix from the second to last unit to the first unit for backward actuation, Figure 11, can be derived as [ (𝑁−2) 𝑢𝑅 (𝑁−2)] = [ 𝐹𝑅 𝑁−2 𝐵 −𝑇12 𝐵 𝑇11 𝐵 ] 𝐵 −𝑇22 𝑇21 [ (1) 𝑢𝑅 (1)] = [ 𝐹𝑅 𝑀11 𝑀21 𝐵 𝐵 𝑀12 𝐵 ] [ 𝐵 𝑀22 (1) 𝑢𝑅 (1)] 𝐹𝑅 (17) 𝐵 = −𝜔2 + 1, 𝑇12 where 𝑇11 𝜔2(1 + 𝜇 + 𝛽1𝜇) + 𝛽1)/𝛽1. With fixed boundary conditions, we can derive that 𝐵 = (𝜔2 − 1 − 𝛽1)/𝛽1 , 𝑇21 𝐵 = 𝜔4𝜇 − 𝜔2(1 + 𝜇) , and 𝑇22 𝐵 = (−𝜔4𝜇 + (1) = 𝐶𝑢𝑅 𝐹𝑅 (1) = −𝐶1𝑀11 (𝐶1𝑀12 F −𝑀21 𝐹 F +𝑀22 F ) (1) 𝑢𝑅 (18) where 𝐶1 = 𝛽 1 (𝛽 1 (2+𝛽 1+𝜔4𝜇−𝜔2(1+𝛽 1)+𝜔4𝜇−𝜔2(1+𝛽 1+𝜇)) 1) (1+𝜇)) (𝛽 . For the first nonlinear unit, Figure 11(d), assuming 1-term expansion with slowly varying coefficients (17), 𝑢𝑅 𝑥(𝑡) = 𝑎(𝑡) sin(𝜔𝑡) + 𝑏(𝑡) cos(𝜔𝑡) and 𝑦(𝑡) = 𝑐(𝑡) sin(𝜔𝑡) + 𝑑(𝑡) cos(𝜔𝑡) + 𝑒(𝑡), plugging equation 1 = 𝑥𝑖𝑛 + 𝑥(𝑡) + 𝑦(𝑡) and 𝑥𝑖𝑛 = 𝑝𝑠𝑖𝑛(𝜔𝑡) into equations of motion, eliminating higher order terms, grouping the constant, sin(𝜔𝑡) and cos(𝜔𝑡) terms and combining the equations, a third order 2 where 𝑅𝑦 is the bistable displacement amplitude with polynomial can be similarly derived for 𝑅 = 𝑅𝑦 2 = 𝑐(𝑡)2 + 𝑑(𝑡)2: 𝑅𝑦 𝜌3𝑅3 + 𝜌2𝑅2 + 𝜌𝑅 + 𝜌0 = 0 where 𝜌3 = 225 16 2 𝛽3 1 16 (−1 + 𝐶 + 𝜔2𝜇)2 ( 120 𝛽 (−1 + 𝐶 + 𝜔2𝜇) (𝐶(−1 + 𝜔2 − 2 𝛽) + 2 𝛽 + 𝜔4𝜇 − 𝜔2(1 + 𝜇 + 2 𝛽 𝜇))) 21 𝜌2 = 1 1 2 ) 𝜌1 = 𝜌0 = 16 (−1 + 𝐶 + 𝜔2𝜇)2 ( 16 (𝐶(−1 + 𝜔2 − 2 𝛽) + 2 𝛽 + 𝜔4 𝜇 − 𝜔2 (1 + 𝜇 + 2 𝛽 𝜇)) 16 (−1+𝐶+𝜔2𝜇)2 ( −16 𝑝2 (𝐶(−1 + 𝜔2) + 𝜔2 (−1 + (−1 + 𝜔2)𝜇)) ) 2 This polynomial can be further simplified as 𝑅3 + 𝑎2𝑅2 + 𝑎1𝑅 + 𝑎0 = 0 with leading coefficients to be 1. Transmission threshold of the bifurcation can then be determined by solving the same Eq.(16) with newly derived coefficients 𝑎1, 𝑎2 and 𝑎3 for backward actuation. Figure 11.(a) Discrete chain of localized nonlinear-linear model for backward actuation; Only the first unit adjacent to input is nonlinear (b) Schematic of 𝑗𝑡ℎ and (𝑗 + 1)𝑡ℎ unit for transfer matrix analysis. Subscript in square bracket corresponds to the index of metastable module shown in Figure 2. (c) Schematic of the last unit for transfer matrix analysis. (d) Schematic of first unit for harmonic balancing analysis. 4.3 Analytical results With the localized nonlinear-linear method described in Sec 4.1 and 4.2 for forward and backward actuations, we compare the analytical prediction of the simplified model with numerical investigations of the fully nonlinear metastructure to evaluate the effectiveness of the analytical method. Parameters used for simulations are the same as in Sec. 3. Figure 12 depicts the contour plot on transmittance ratio as input frequency and amplitude varies for both forward and backward actuations, with dark area corresponds to non-propagation zone. Therefore, the boundaries of supratransmission in the parameter space correspond to the transitions of color from dark to bright. Solid red and black lines are analytical predicted onset of supratransmission for forward and backward actuations, respectively. As shown in Figure 11(a), the proposed method is able to predict accurately the onset of the supratransmission induced by the saddle- node bifurcation in region II discussed in Sec. 3 for forward actuation over a range of the parameters investigated. For backward actuation, analytical predicted boundary (black lines) also closely resembles the trend of the numerical simulation: input amplitude level required to facilitate the onset of supratransmission increases as input frequency decreases. Combining the two predicted boundaries, Figure 12(b), since it predicts that wave energy can only propagate through the chain with input parameters above the red (black) line for forward (backward) actuation scenario, there exists a region in parameter space that are bounded by the two predicted transition boundaries such that wave energy can 22 only transmit in one direction. Hence, the proposed metastable structure is capable of attaining supratransmission at different excitation level depending on the excitation directions. Therefore, by combining the supratransmission property of a nonlinear periodic chain with spatial asymmetry, we are able to enable non-reciprocal wave propagation and predict the non-reciprocal region using the proposed analytical approach. Overall, the localized nonlinear-linear model and prediction method proposed in the study are beneficial for identifying the unidirectional wave propagation region in the input parameter space, which can greatly reduce the computational time and complexity. It should be noted that in this study, as a proof of concept, we have focused on analytically predicting the onset of supratransmission induced mainly by the saddle-node bifurcation for one of the metastable states. On the other hand, as discussed in Sec. 3, onset of supratransmission can be facilitated through other intricate nonlinear instabilities as well, which could be predicted using similar approach. That is, utilizing the same principle of localized nonlinear-linear model, we could adopt different forms of harmonic expansion for the first nonlinear unit to derive other supratransmission boundaries of interest [44] [45] [46] [47]. Additionally, the proposed analytical method can be readily extended to other metastable states by linearizing the system around different starting positions. Figure 12. (a) Contour plot on transmittance ratio (vs. input frequency and amplitude) of forward and backward actuations for free length configuration with analytically predicted supratransmission boundaries. (b) Non-reciprocal region is bounded in between the two analytically predicted boundaries. 5 Parametric studies on supratransmission threshold From both numerical and analytical investigations in Sec. 3 and 4, we find that when the excitation frequency is within the stopband of the metastructure, there exhibits a threshold input amplitude beyond which wave energy is able to propagate through the chain due to supratransmission. Additionally, when this onset of supratransmission is induced by saddle-node bifurcation, a drastic increase in transmittance ratio can be observed as input amplitude exceeds the threshold value and this sudden change could potentially be used as a practical means to design non-reciprocal regions. Therefore, in the following 23 sections, we seek to explore influences of important parameters on the threshold amplitude to trigger supratransmission and to provide more insight for synthesizing systems with desired non-reciprocal characteristics. 5.1 Influence of damping coefficient In this section, the influence of damping coefficients 𝑐 on the threshold amplitude is studied. System parameters used are the same as in Sec 3 and 4 and excitation frequency is chosen to be 𝜔 = 1.6. For exploration purposes, the metastructure is actuated under forward excitation scenario with damping coefficients 𝑐 varying from 0.001, 0.005, 0.01, 0.05 to 0.1. Figure 13 depicts the transmittance ratio as input amplitude increases for different damping coefficients. With small damping coefficients, 𝑐 = 0.001, 𝑐 = 0.005, and 𝑐 = 0.01, threshold amplitudes to trigger the onset of supratransmission gradually increase from 𝑝 = 1.27 to 𝑝 = 1.28 and 𝑝 = 1.31. Additionally, due to the increased damping coefficient, energy transmitted through the structure in general decreases hence the transmittance ratio also decreases. As the damping ratio further increases to 𝑐 = 0.05 and 𝑐 = 0.1 , the threshold amplitude required to enable supratransmission increases to 1.54 and 1.75 respectively and the transmitted energy drops drastically with transmittance ratio (TR) decreases from more than 1 to approximately 0.26. Therefore, increasing damping coefficients will increase the threshold amplitude required to trigger the onset of supratransmission and decrease the amount of energy propagated through the structure. 24 Figure 13. Transmittance ratio of metastructure under forward excitation as input amplitude increases for different damping coefficients 𝑐 = [0.001, 0.005, 0.01, 0.05, 0.1]. 5.2 Influence of the number of modules Results presented in the previous sections are all based on a 10 module assembly, in this section, the influence of module number 𝑁 on the threshold amplitude is studied. System parameters used are the same as those used in Sec 3 and 4 with damping coefficients 𝑐 = 0.001. Again, for exploration purposes, the metastructue is actuated only under the forward excitation scenario. Figure 14 depicts the threshold amplitude that enables supratransmission as module number increases from 5, 10, 20, 50 to 100 for a range of frequencies inside Region II, Figure 3. For a fixed excitation frequency, threshold amplitude initially may fluctuates with small module number and as the number of modules increases, this value becomes constant and is invariant of module number. Such an invariant characteristic is similarly observed in other nonlinear metamaterials [48]. Additionally, from the study of the 10 module assembly, we noticed that as the input frequency increases, the threshold amplitude to trigger supratransmission deceases, Figure 12. This trend is also preserved as module number changes from 5 to 100: the threshold amplitude requires to facilitate supratransmission always decreases as frequency increases. Figure 14. Threshold amplitude of metastructure under forward excitation for module number 𝑁 = [5, 10, 20, 50,100] and excitation frequency 𝜔 = [1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9]. 25 6 Conclusion In this paper, we numerically and analytically explore the supratransmission phenomenon in a metastable modular metastructure. From numerical investigations on a 1D metastable modular chain, we explore intricate dynamics affordable by the metastructure and elucidate different kinds of nonlinear instabilities that facilitate the onset of supratransmission. In the case of saddle-node bifurcation instability, the transmission ratio could suddenly increase significantly as the input amplitude increases and exceeds certain threshold. We find that when the excitation frequency is inside the stopband, dynamics of the first nonlinear constituent close to the input can considerably influence the wave energy transmission characteristics of the overall system. Therefore, we propose a localized nonlinear-linear model to analytically estimate the input threshold amplitude required to trigger supratransmission of the nonlinear metastable modular chain. Through intelligently integrating the supratransmission property in a nonlinear periodic chain with spatial asymmetry, the metastable structure is capable of realizing non-reciprocal wave propagation by attaining sudden propagation of wave energy at different excitation levels depending on the input directions. Utilizing the proposed analytical tool, we are able to accurately and efficiently predict the non-reciprocal region depending on input frequency and amplitude. Additionally, we explore the influence of system damping and module number on the threshold amplitude to trigger the onset of supratransmission. The outcomes of this study provide effective methodology and great potential to design systems with desired adaptable non-reciprocal wave energy propagation characteristics utilizing the proposed reconfigurable metastable modular metastructure concept. Acknowledgment The authors gratefully acknowledge the support of the U.S. Army Research Office under grant number W911NF-15-1-0114. Reference [1] L. Rayleigh, "Some general theorems relating to vibrations," Proceedings of the London Mathematical Society, vol. 4, pp. 357-368, 1873. [2] H. Lamb, "On reciprocal theorems in dynamics," Proceedings of the London Mathematical Society, vol. 19, pp. 144-151, 1888. [3] N. Boechler, G. Theocharis and C. Daraio, "Bifurcation-based acoustic switching and rectification," Nature Materials, vol. 10, pp. 665-668, 2011. 26 [4] B. Liang, X. S. Guo, J. Tu, D. Zhang and J. C. Cheng, "An acoustic rectifier," Nature Materials, vol. 9, pp. 989-992, 2010. [5] R. Fleury, D. L. Sounas, C. F. Sieck, M. R. Haberman and A. Alù, "Sound isolation and giant linear nonreciprocity in a compact acoustic circulator," Science, vol. 343, pp. 516-519, 2014. [6] R. Fleury, D. Sounas, M. R. Haberman and A. Alù, "Nonreciprocal acoustics," Acoustics Today, vol. 11, pp. 14-21, 2015. [7] T. Devaux, V. Tournat, O. Richoux and V. Pagneux, "Asymmetric acoustic propagation of wave packets via the self-demodulation effect," Physical Review Letters, vol. 115, p. 234301, 2015. [8] B. Popa and S. Cummer, "Non-reciprocal and highly nonlinear active acoustic metamaterials," Nature Communications, vol. 5, p. 3398, 2014. [9] C. Liu, Z. Du, Z. Sun, H. Gao and X. Guo, "Frequency-preserved acoustic diode model with high forward-power-transmission rate," Physical Review Applied, vol. 3, p. 064014, 2015. [10] Q. Wang, Y. Yang, X. Ni, Y. Xu, X. Sun, Z. Chen, L. Feng, X. Liu, M. Lu and Y. Chen, "Acoustic asymmetric transmission based on time-dependent dynamical scattering," Scientific Reports, vol. 5, p. 10880, 2015. [11] A. Maznev, A. Every and O. Wright, "Reciprocity in reflection and transmission: What is a 'phonon diode'?," Wave Motion, vol. 50, no. 4, pp. 776-784, 2013. [12] N. Swinteck, S. Matsuo, K. Runge, J. O. Vasseur, P. Lucas and P. A. Deymier, "Bulk elastic waves with unidirectional backscattering-immune topological states in a time-dependent superlattice," Journal of Applied Physics, vol. 118, p. 063103, 2015. [13] B. Deng, P. Wang, Q. He, V. Tournat and K. Bertoldi, "Metamaterials with amplitude gaps for elastic solitons," Nature Communications, vol. 9, p. 3410, 2018. [14] B. Li, L. Wang and G. Casati, "Thermal diode: Rectification of heat flux," Physical Review Letters, vol. 93, p. 184301, 2004. [15] B. Hu, L. Yang and Y. Zhang, "Asymmetric heat conduction in nonlinear lattices," Physical Review Letters, vol. 97, p. 124302, 2006. [16] M. Tocci, M. Bloemer, M. Scalora, J. P. Dowling and C. M. Bowden, "Thin‐film nonlinear optical 27 diode," Applied Physics Letters, vol. 66, p. 2324, 1995. [17] D.-W. Wang, H.-T. Zhou, M.-J. Guo, J.-X. Zhang, J. Evers and S.-Y. Zhu, "Optical diode made from a moving photonic crystal," Physical Review Letters, vol. 110, p. 093901, 2013. [18] S. A. Cummer, J. Christensen and A. Alù, "Controlling sound with acoustic metamaterials," Nature Review Materials, vol. 1, p. 16001, 2016. [19] Z.-G. Chen and Y. Wu, "Tunable topological phononic crystals," Physical Review Applied, vol. 5, p. 054021, 2016. [20] J. R. Raney, N. Nadkarni, C. Daraio, D. M. Kochmann, J. A. Lewis and K. Bertoldi, "Stable propagation of mechanical signals in soft media using stored elastic energy," Proceedings of the National Academy of Sciences, vol. 113, pp. 9722-9727, 2016. [21] Z. Wu, Y. Zheng and K. W. Wang, "Metastable modular metastructures for on-demand reconfiguration of band structures and nonreciprocal wave propagation," Physical Review E, vol. 97, p. 022209, 2018. [22] M. Hussein, M. Leamy and M. Ruzzene, "Dynamics of phononic materials and structures: Historical origins, recent progress, and future outlook," Applied Mechanics Reviews , vol. 66 , no. 4, p. 040802, 2014. [23] L. Brillouin, Wave propagation in periodic structures: Electric filters and crystal lattices, Dover Publications, Inc., 1953. [24] I. El-Kady, R. Olsson and J. Fleming, "Phononic band-gap crystals for radio frequency communications," Applied Physics Letters, vol. 92, p. 233504, 2008. [25] R. Johnson, Mechanical filters in electronics, John Wiley & Sons Inc, 1983. [26] F. Geniet and J. Leon, "Energy transmission in the forbidden band gap of a nonlinear chain," Physical Review Letters, vol. 89, p. 134102, 2002. [27] J. Leon, "Nonlinear supratransission as a fundamental instability," Physics Letters A, vol. 319, pp. 130-136, 2003. [28] J. E. Macías-Díaz, "Numerical study of the transmission of energy in discrete arrays of sine-Gordon equations in two space dimensions," Physical Review E, vol. 77, p. 016602, 2008. 28 [29] P. Anghel-Vasilescu, J. Dorignac, F. Geniet, J. Leon and M. Taki, "Nonlinear supratransmission in multicomponent systems," Physical Review Letters, vol. 105, p. 074101, 2010. [30] R. Khomeriki, S. Lepri and S. Ruffo, "Nonlinear supratransmission and bistability in the Fermi- Pasta-Ulam model," Physical Review E, vol. 70, p. 066626, 2004. [31] M. J. Frazier and D. M. Kochmann, "Band gap transmission in periodic bistable mechanical systems," Journal of Sound and Vibration, vol. 388, pp. 315-326, 2017. [32] B. Yousefzadeh and A. S. Phani, "Supratransmission in a disordered nonlinear periodic structure," Journal of Sound and Vibration, vol. 380, pp. 242-266, 2016. [33] Z. Wu, R. L. Harne and K. W. Wang, "Exploring a modular adaptive metastructure concept inspired by muscle's cross-bridge," Journal of Intelligent Material Systems and Structures, vol. 27, pp. 1189- 1202, 2016. [34] R. L. Harne, Z. Wu and K. W. Wang, "Designing and harnessing the metastable states of a modular metastructure for programmable mechanical properties adaptation," Journal of Mechanical Design, vol. 138, p. 021402, 2015. [35] M. Caruel, J. M. Allain and L. Truskinovsky, "Mechanics of collective unfolding," Journal of the Mechanics and Physics of Solids, vol. 76, pp. 237-259, 2015. [36] M. Caruel, J. M. Allain and L. Truskinovsky, "Muscle as a metamaterial operating near a critical point," Physical Review Letters, vol. 110, p. 248103, 2013. [37] J. Lydon, G. Theocharis and C. Daraio, "Nonlinear resonances and energy transfer in finite granular chains," Physical Review E, vol. 91, p. 023208, 2015. [38] R. L. Harne, M. Thota and K. Wang, "Concise and high-fidelity predictive criteria for maximizing performance and robustness of bistable energy harvesters," Applied Physics Letters, vol. 102, no. 5, p. 053903, 2013. [39] R. L. Harne and K. Wang, Harnessing bistable structural dynamics: for vibration control, energy harvesting and sensing, John Wiley & Sons, ISBN: 978-1-119-12804-5, 408 pages, 2017. [40] Z. Wu, L. R. Harne and K. Wang, "Energy harvester synthesis via coupled linear-bistable system with multistable dynamics," Journal of Applied Mechanics, vol. 81, no. 6, p. 061005, 2014. 29 [41] D. Mead, "Wave propagation in continuous periodic structures: research contributions from Southampton, 1964-1995," Journal of Sound and Vibration, vol. 190, no. 3, pp. 495-524, 1996. [42] T. Butlin and R. Langley, "An efficient model of drillstring dynamics," Journal of Sound and Vibration, vol. 356, pp. 100-123, 2015. [43] R. Khajehtourian and M. Hussein, "Dispersion characteristics of a nonlinear elastic metamaterial," AIP Advances, vol. 4, p. 124308, 2014. [44] J. Summers, "Variable-coefficient harmonic balance for periodically forced nonlinear oscillators," Nonlinear Dynamics, vol. 7, pp. 11-35, 1995. [45] G. R. Itovich and J. L. Moiola, "On period doubling bifurcations of cycles and the harmonic balance method," Chaos, Solitons and Fractals, vol. 27, p. 647 -- 665, 2006. [46] M. Guskov and F. Thouverez, "Harmonic balance-based approach for quasi-periodic motions and stability analysis," Journal of Vibration and Acoustics, vol. 134, p. 031003, 2012. [47] T. Detroux, L. Renson, L. Masset and G. Kerschen, "The harmonic balance method for bifurcation analysis of large-scale nonlinear mechanical systems," Computer Methods in Applied Mechanics and Engineering, vol. 296, pp. 18-38, 2015. [48] X. Fang, J. Wen, B. Bonello, J. Yin and D. Yu, "Wave propagation in one-dimensional nonlinear acoustic metamaterials," New Journal of Physics, vol. 19, p. 053007, 2017. 30
1907.08853	1	1907	2019-07-20T18:43:37	Anomalous energy transport in laminates with exceptional points	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Recent interest in metamaterials has led to a renewed study of wave mechanics in different branches of physics. Elastodynamics involves a special intricacy, owing to a coupling between the volumetric and shear parts of the elastic waves. Through a study of in-plane waves traversing periodic laminates, we here show that this coupling can result with unusual energy transport. We find that the corresponding frequency spectrum contains modes which simultaneously attenuate and propagate, and demonstrate that these modes coalesce to purely propagating modes at exceptional points, a property that was recently reported in parity-time symmetric systems. We show that the laminate exhibits metamaterial features near these points, such as negative refraction, and beam steering and splitting. While negative refraction in laminates has been demonstrated before by considering pure shear waves impinging an interface with multiple layers, here we realize it for coupled waves impinging a simple single-layer interface. This feature, together with the appearance of exceptional points, are absent from the model problem of anti-plane shear waves which have no volumetric part, and hence from the mathematically identical electromagnetic waves. Our work further paves the way for applications such as asymmetric mode switches, by encircling exceptional points in a tangible, purely elastic apparatus.	physics.app-ph	physics	Anomalous energy transport in laminates with exceptional points Ben Lustig, Guy Elbaz, Alan Muhafra and Gal Shmuel∗ Faculty of Mechanical Engineering, Technion -- Israel Institute of Technology, Haifa 32000, Israel Abstract Recent interest in metamaterials has led to a renewed study of wave mechanics in dif- ferent branches of physics. Elastodynamics involves a special intricacy, owing to a coupling between the volumetric and shear parts of the elastic waves. Through a study of in-plane waves traversing periodic laminates, we here show that this coupling can result with unusual energy transport. We find that the corresponding frequency spectrum contains modes which simultaneously attenuate and propagate, and demonstrate that these modes coalesce to purely propagating modes at exceptional points -- a property that was recently reported in parity-time symmetric systems. We show that the laminate exhibits metamaterial features near these points, such as negative refraction, and beam steering and splitting. While negative refraction in lam- inates has been demonstrated before by considering pure shear waves impinging an interface with multiple layers, here we realize it for coupled waves impinging a simple single-layer in- terface. This feature, together with the appearance of exceptional points, are absent from the model problem of anti-plane shear waves which have no volumetric part, and hence from the mathematically identical electromagnetic waves. Our work further paves the way for applica- tions such as asymmetric mode switches, by encircling exceptional points in a tangible, purely elastic apparatus. Keywords: metamaterial, negative refraction, wave propagation, Bloch-Floquet waves, composite, laminate, phononic crystal, exceptional points 9 1 0 2 l u J 0 2 ] h p - p p a . s c i s y h p [ 1 v 3 5 8 8 0 . 7 0 9 1 : v i X r a 1 Introduction Metamaterials possess properties not found in nature, stemming from their architectured microstruc- ture (Wegener, 2013, Kadic et al., 2019). Perhaps the most prominent thrust in metamaterials ∗Corresponding author. Tel.: +1 972 778871613. E-mail address: [email protected] (G. Shmuel). 1 research aims at controlling waves for potential applications such as lensing, cloaking and noise reduction (Milton et al., 2006, Chen et al., 2010, Parnell and Shearer, 2013, Bigoni et al., 2013, Colquitt et al., 2014, Cummer et al., 2016). This interest led to a renewed study of wave mechanics in optics (Markos and Soukoulis, 2008, Banerjee, 2011), acoustics (Craster and Guenneau, 2012, Deymier, 2013), mechanical lattices (Phani, 2011, Raney et al., 2016, Phani and Hussein, 2017, Zelhofer and Kochmann, 2017, Ma et al., 2018) and elastodynamics (Brun et al., 2010, Shmuel and Band, 2016, Chen and Elbanna, 2017, Aghighi et al., 2019, Li and Reina, 2019). While electromagnetic-, sound- and anti-plane shear waves are mathematically identical (Adams et al., 2008, Torrent and Sánchez-Dehesa, 2011), in-plane elastic waves are physically richer (Sigalas and Economou, 1992), since they comprise both volumetric and distortional parts, coupled through interfaces in the transmission medium. In this work, we show that in the simplest elastic compos- ite -- a laminate -- this coupling gives rise to anomalous energy transport, which in other systems is achieved by significantly more complicated means. The anomalies reported here and their connec- tion with recent studies in the field are summarized next. Firstly, we show that the spectrum of in-plane waves in laminates exhibits exceptional points, accessible in a purely elastic setting. Exceptional points are states of a system at which two (or more) of its normal modes coalesce, together with their natural frequencies (Moiseyev and Fried- land, 1980, Ding et al., 2015), and are the source of counterintuitive phenomena such as enhanced sensitivity, wave stopping and asymmetric transmission (Hodaei et al., 2017, Achilleos et al., 2017, Goldzak et al., 2018, Merkel et al., 2018). Exceptional points occur only in non-Hermitian systems (Moiseyev, 2011), a property which usually describes systems that interact with the environment. The current paradigm1 to access exceptional points is by balancing external gain and loss through the system, to create parity-time (PT ) symmetry (Rüter et al., 2010, Shi et al., 2016, El-Ganainy et al., 2018). This procedure brings with additional complexity to the elastic medium, as such re- alizations require incorporating optomechanical, acoustoelectric or piezoelectric elements into the system (Xu et al., 2015, Christensen et al., 2016, Hou and Assouar, 2018). Our findings thus suggest a simpler, purely elastic setting to access exceptional points, thereby evading these complexities. Recently, it was demonstrated that encircling these points leads to fascinating asymmetric mode switching of microwaves in a metallic waveguide (Doppler et al., 2016). We argue that our setting constitutes a tangible platform to realize analogous encirclement for elastic wave switching, having the wavenumber, which is related to the excitation angle, as the trajectory parameter. As we show in the sequel, this is made possible owing to the intrinsic Riemann surface structure of our spectrum near these points in the complex wave vector space. In contrast with cases where this structure is artificially obtained by an analytical continuation of an arbitrary parameter (e.g., the frequency 1To break reciprocity, we note that another emerging paradigm is to employ spatiotemporal composites (Trainiti and Ruzzene, 2016, Nassar et al., 2017, Milton and Mattei, 2017) 2 in Shanin et al., 2018), here, complex wavenumbers are an intrinsic and accessible part of the spectrum. Secondly, we demonstrate that the exceptional points foreshadow anomalous energy transport, including negative refraction. In this regard, perhaps our most striking result is the excitation of negative refraction in laminates by an incoming wave from a homogeneous medium whose interface with the laminate is parallel to the layers (Fig. 1). To put this result into context, we recall that the first report of negative refraction was for electromagnetic waves, and required a two-dimensional composite made of complicated split ring resonators (Smith et al., 2000, Shelby et al., 2001, Pendry, 2004). Interest in this phenomenon has been disseminated to elastodynamics, accompanied with ongoing studies (Craster and Guenneau, 2012, Chen et al., 2017, Bordiga et al., 2019, Nemat-Nasser, 2019). Notably, Willis (2013) was the first to show that a simple laminate is capable of negatively refracting anti-plane shear waves. While in the conventional arrangement (interface parallel to the layers), refraction is always positive, Willis realized that negative refraction is possible when the interface is normal to the layers (Fig. 1c). Based on this interface configuration, further studies of such waves in laminated media were carried out (Willis, 2015, Nemat-Nasser, 2015, Srivastava, 2016, Morini et al., 2019). Hence, we demonstrate that in-plane waves may refract negatively in the conventional arrangement, without the need for this complex apparatus, nor for gain and loss (cf. Hou et al., 2018). For completeness, we also analyze the transmission problem through an interface normal to the layers. As highlighted by Srivastava and Willis (2017) and the references therein, a wave incident to such interface induces an infinite number of transmitted waves, which are required to satisfy corresponding continuity conditions across the interface. We suggest a method to calculate the resultant normal mode decomposition owing to incident in-plane waves, based on suitable orthog- onality conditions (Mokhtari et al., 2019). Subsequently, we demonstrate analogous phenomena to those reported by Srivastava (2016), who studied the anti-plane problem. These include beam steer- ing -- small changes in the incident angle leading to large changes in the transmission angle; beam splitting -- an incident wave transmitted as simultaneous negative and positive refracted beams; and pure negative refraction. Our study is detailed in the forthcomings Secs. as follows. Sec. 2 firstly revisits the equations governing in-plane waves in infinite periodic laminates, and subsequently provides two formula- tions -- the extended plane wave expansion method and the hybrid matrix method -- for their so- lution (Laude et al., 2009, Tan, 2010). By analyzing the pertinent equations, we explain why in contrast with the anti-plane problem, we here obtain complex wavenumbers, exceptional points, and negative refraction in the simple configuration. We numerically solve the eigenvalue problems for an exemplary infinite laminate, and present its spectrum in Sec. 3. As predicted, the complex spectrum contains exceptional points with hallmarks of negative refraction. Sec. 4 links the studied 3 Fig. 1. (a) An infinite laminate comprising alternating a and b phases. The medium admits Bloch modes with vertical wavenumber k2, and macroscopic horizontal wavenumber kB. (b) A homogeneous half-space bonded to a semi-infinite laminate at x1 = 0, and at (c) x2 = 0. We schematically denote the incident, re- flected and transmitted waves by I, R and T , respectively. The incident and transmission angles are denoted by θi and θ, respectively. spectrum and eigenmodes to their excitation by incoming waves from a homogeneous half-space that shares an interface with the exemplary laminate. Finally, our main results and conclusions are summarized in Sec. 5. 2 Equations of in-plane waves in laminated media The equations governing in-plane waves in elastic solids can be found in Graff (1975), and their extension to laminated media appears, e.g., in Lowe (1995) and the references therein (see also Adams et al., 2009 for in-plane Bloch waves in infinitely periodic strips). In this Sec., we firstly revisit the equations for an infinite laminate comprising two alternating layers, and provide two formulations to determine the resultant normal modes via different eigenvalue problems. Based on these formulations, we explain next why in contrast with anti-plane waves, in-plane waves in laminates admit complex eigenvalues and exceptional points, and connect these features of the spectrum to the transport of energy. 2.1 Governing Equations We study an infinite elastic laminate made of a periodic repetition of phases a and b in the x1 direc- tion (Fig. 1a). The thickness, mass density and Lamé coefficients of each phase are h(p), ρ(p), λ(p) and µ(p), respectively, where p = a or b, denoting the corresponding phase. The objective is to determine the propagation of free time-harmonic in-plane waves in the lam- (n) in each nth layer using inate. To this end, we seek solutions for the displacements u1 the Naiver-Lamé equations (n) and u2 4 x2x1kBk2x2x1IRT✓i✓x2x1IRT✓i(a)(b)(c)(cid:0)λ(p) + 2µ(p)(cid:1) u1,11 (cid:0)λ(p) + 2µ(p)(cid:1) u2,22 (n) +(cid:0)λ(p) + µ(p)(cid:1) u2,21 (n) +(cid:0)λ(p) + µ(p)(cid:1) u1,12 (n) + µ(p)u1,22 (n) + µ(p)u2,11 (n) = ρ(p) u1 (n) = ρ(p) u2 (n), (n), (1) subjected to boundary conditions that will be specified later. We employ the Helmholtz decompo- sition to write the in-plane components of u(n) in terms of scalar potential φ(n) and ψ(n), namely, u1 (n) = φ,1 (n) + ψ,2 (n), u2 (n) = φ,2 (n) − ψ,1 (n) . (2) This decomposition simplifies Eq. (1) to the form ∇2φ(n) = φ(n), ∇2ψ(n) = L =(cid:112)(λ(p) + 2µ(p)) /ρ(p) and c(p) S =(cid:112)µ(p)/ρ(p) are the pressure and shear wave veloci- where c(p) ties of phase p, respectively. At a fixed frequency ω, Eq. (3) is solved by ψ(n), (3) L 1 c2(p) S 1 c2(p) (cid:16) (cid:16) L+ei S+ei φ(n) = A(n) ψ(n) = A(n) ωt+k(p) L x1−k2 (p)x2 ωt+k(p) S x1−k2 (p)x2 (cid:17) (cid:17) (cid:16) (cid:16) L−ei S−ei + A(n) + A(n) ωt−k(p) L x1−k2 (p)x2 ωt−k(p) S x1−k2 (p)x2 (cid:17) (cid:17) , , (4) where and(cid:8)AL/S,±(n)(cid:9) are integration constants to be determined from the conditions on the boundaries (p), k2(p) k2(p) L = S = (p), (5) L S 2 − k2 2 ω2 c2(p) − k2 ω2 c2(p) of layer n. These correspond to the continuity of the traction and displacements at interfaces be- tween adjacent layers, which immediately requires k(a) 2 = k(b) 2 =: k2. The corresponding equations are compactly written in terms of the state vector s(n) (x1, x2, t) (n)(cid:1)T s(n) =(cid:0)u1 (n)(cid:1) = s(n+1)(cid:0)x1 = x0 simply s(n)(cid:0)x1 = x0 say the boundary between layers n and n + 1 is at x1 = x(n) (n), σ11 (n), u2 (n), σ21 ; (6) (n)(cid:1). The coupling between shear and pressure modes 0 , then the continuity conditions are enters through the latter condition, since the components of s(n) depend both on φ(n) and ψ(n). The remaining equations stem from the Bloch-Floquet theorem, which states that over the course of one period the governing fields are related via the Bloch wavenumber kB, namely, s(n)(cid:0)x1 = x0 (n) − h(n)(cid:1) = eikBhs(n+1)(cid:0)x1 = x0 (n) + h(n+1)(cid:1) . (7) 5 2.2 The hybrid matrix method to determine kB (ω) Eqs. (6) and (7) can be combined in different ways to deliver an eigenproblem for kB and s(n) as functions of real ω at prescribed k2, given the laminate composition. The transfer matrix formu- lation is the most intuitive and common approach, however it suffers from numerical instabilities (Pérez-Álvarez et al., 2015). Here, we employ the stable hybrid matrix method (Tan, 2010), which reads · sm (x1, x2) = eikBh · sm (x1, x2) , (8) (cid:32) −I2 H11 (cid:33) 02 H21 (cid:33) (cid:32) −H12 02 −H22 I2 where 02 (resp. I2) is the 2 × 2 zeros (resp. unit) matrix, and sm and Hij are given in Appendix A, together with a detailed derivation of Eq. (8). The resultant quartic equation for kB is solved by −a2 ±(cid:112)a2 2 − 4a1 (a3 − 2a1) 4a1 cos kBh = , (9) where ai are given in Appendix A. Eq. (9) provides the dispersion relation which relates the mi- crostructure and mechanical properties of the laminate to the waveform at each frequency. The structure of Eq. (9) implies that if kB is a solution, then so are ±kB + 2πm for any m ∈ Z. Hence, the real part of all dispersion curves is representable over the irreducible 1st Brillouin zone2 0 < kB < π/h. The wavenumber kB can be real, pure imaginary (henceforth referred as imag- inary) or complex, as demonstrated in the sequel, and contrary to the case of anti-plane shear (Willis, 2015, Srivastava, 2016). This property is essential for the spectrum to exhibit a Riemann structure, and, in turn, exceptional points. Real kB corresponds to a propagating Bloch mode along x1, where imaginary kB corresponds to attenuating modes, with the exponential decay e−kBx1; since einπ = ±1 for integer n, we will also refer to kB with Re kBh = nπ as imaginary, and in- terpret corresponding modes as non-propagating. We note that modes with kBh = nπ are at the boundary of the Brillouin zones, and represent standing waves. Complex kB describes a progressive mode that exponentially decay according to Im kB. We emphasize that wave attenuation is not an indication of energy loss, as our system is non-dissipative; it is the result of a gradual scattering of energy to incoherent waves with zero mean. Bands of frequencies without real kB roots are termed directional band gaps (or simply gaps) since there is a gap in the spectrum in the x1 direction. When there are no real k2 solutions across these bands, they are termed complete gaps, since all directions of propagation are prohibited. 2See, e.g., Zhang (2019) for a discussion on the reciprocal space symmetries and degeneracies in the general case. 6 2.3 An eigenvalue problem for k2 using a plane wave approach The standard plane wave expansion method to obtain the dispersion relation in elastodynamics dates back to Sigalas and Economou (1992) and Kushwaha et al. (1993). In photonics, the method has been extended by Hsue et al. (2005) to a formulation in which the Bloch wavenumber is the eigenvalue, and later on by Laude et al. (2009) for elastodynamics. A general analysis of the method, associated linear operators, and the properties of the eigenvalues was carried out recently by Mokhtari et al. (2019). Building upon the approach of Laude et al. (2009), we formulate for our settings an eigenvalue problem in which k2 is the eigenvalue3. We begin by deriving the standard plane wave method by substituting into the Cauchy equations of motion {(ui,j + uj,i) µ (x1) + λ (x1) δijuk,k},j = ρ (x1) ui, (10) the Bloch form ui = ui (x1) ei(ωt−kBx1−k2x2), (11) where here and henceforth, the periodic part of (◦) is denoted by (◦). Since ρ (x1), µ (x1), λ (x1) and ui (x1) are periodic in x1 with a period h, they can be written as ui (x1) = ui (x1 + h) , (cid:88) G ζ (x1) = ζ (G) e−iGx1, ζ = ρ, µ, λ, u1, u2, (12) (13) (14) where G = 2πm h , m ∈ Z, with the Fourier coefficients {ζ (G)} (cid:90) h 0 ζ (G) = 1 h ζ (x1) eiGx1dx1. We have that ζ (G) =  h(a) −1(cid:0)ζ (a) − ζ (b)(cid:1)(cid:16) h ζ (a) + h(b) (πm) h ζ (b), 1 − eiGh(a)(cid:17) G = 0, , G (cid:54)= 0. 3We recall that k2 is not a Bloch wavenumber and thus is not identified with any irreducible Brillouin zone (or, conversely, identified with an infinite one), as the laminate is homogeneous in the x2 direction. 7 Substituting Eqs. (11) and (12) into Eq. (10) and factoring out ei(ωt−kBx1−k2x2) yield G,G(cid:48) − (G + G(cid:48) + kB) [k2λ (G) u2 (G(cid:48)) + (G(cid:48) + kB) (λ (G) + 2µ (G)) u1 (G(cid:48))]} e−i(G+G(cid:48))x1 = {−k2 [k2 u1 (G(cid:48)) + (G(cid:48) + kB) u2 (G(cid:48))] µ (G) − (cid:88) = −(cid:88) (cid:88) (cid:8)−k2 [(G(cid:48) + kB) λ (G) + (G + G(cid:48) + kB) µ (G)] u1 (G(cid:48)) − k2 − µ (G)(cid:2)2k2 2 + (G(cid:48) + kB) (G + G(cid:48) + kB)(cid:3) u2 (G(cid:48))(cid:9) e−i(G+G(cid:48))x1 = = −(cid:88) ω2ρ (G) u1 (G(cid:48)) e−i(G+G(cid:48))x1, ω2ρ (G) u2 (G(cid:48)) e−i(G+G(cid:48))x1. G,G(cid:48) G,G(cid:48) G,G(cid:48) 2λ (G) u2 (G(cid:48)) − (15) We multiply Eqs. (15) by eiG(cid:48)(cid:48)x1 and integrate over one period. Due to Fourier orthogonality, only terms satisfying G(cid:48)(cid:48) = G + G(cid:48) remain, and we end up with an infinite set of equations that can be cast in matrix form as (cid:2)A0 + k2A1 + k2 (cid:3) u (G(cid:48)) = ω2Bu (G(cid:48)) , 2A2 (16) where u (G(cid:48)) is a column vector comprising the Fourier coefficients of u1 and u2, and the matrices A0, A1, A2 and B are given in Appendix B. Eq. (16) constitutes a generalized eigenproblem for ω2 and u (G(cid:48)) at prescribed k2 and kB. A mixed generalized eigenproblem for k2 at prescribed ω and kB follows from Eq. (16) using a state space-like formulation (Chapt. 10 in Deymier, 2013, see also Hussein et al., 2014), namely, (cid:35)(cid:32) (cid:33) (cid:34) (cid:35)(cid:32) (cid:33) 0 ω2B − A0 −A1 I u (G(cid:48)) k2u (G(cid:48)) = k2 I 0 0 A2 u (G(cid:48)) k2u (G(cid:48)) (17) (cid:34) For completeness, note that an equivalent formulation can be obtained using the displacement to- gether with the stress and the eigenvector (Mokhtari et al., 2019). For computational purposes, the number of terms in the Fourier series is truncated, say by −N ≤ m ≤ N. The matrices Ai, B, 0 and I are accordingly of dimension (2N + 1) × (2N + 1). 2.4 Some properties of kB and k2 and their effect on energy transport To highlight how the coupling between shear and pressure waves affect kB and k2, we record next their properties in the case of anti-plane shear in laminates, where such coupling is absent, and subsequently point out the differences. These differences have significant implications on frequency spectrum, and in turn the energy flow, since the slopes of propagating branches are a 8 measure of its mean. More formally, the average energy flow is (cid:104)P(cid:105) = −1 2 (cid:104)Re [σ u∗](cid:105) , and satisfies (Willis, 2015) (cid:104)P(cid:105) = ∂ω ∂k E, E = (cid:10)σiju∗ i,j (cid:11) + 1 4 1 4 ω2 (cid:104)ρu∗ kuk(cid:105) ≡ 1 2 ω2 (cid:104)ρu∗ kuk(cid:105) , (18) (19) where (cid:104)◦(cid:105) denotes averaging over one spatial and temporal period, and E is the total mean energy density; P is called the acoustic Poynting vector, and the k-gradient of ω is identified as the group velocity. Exceptional points. -- In anti-plane shear, the size of the transfer matrix is 2 × 2, whose two roots are λ1,2 = eikB1,B2h. When the transfer matrix is real, one can show that if λ is a solution, then so are λ∗ and λ−1 . Since there are only two roots, either λ = λ∗ and then kB is imaginary, or λ∗ = λ−1 and then kB is real. Accordingly, eigenvalue degeneracies occur only at the edge of the irreducible Brillouin zones, where the modes are standing and not propagating. In this usual degeneracy, the eigenmodes remain linearly independent, and the splitting of the eigenvalues from the so-called diabolic points scales linearly. The situation is significantly different when there are two coupled displacements as considered here. The size of the corresponding transfer matrix is 4 × 4, thus has four roots. Hence, the fact that λ∗ and λ−1 are also solutions does not enforce that λ = λ∗ or λ∗ = λ−1, and, in turn the exclusion of complex kB roots. As we show in Sec. 3, not only such complex roots exist, they coalesce together with their eigenmodes at exceptional points inside the irreducible 1st Brillouin zone, according to a square-root scaling, constituting a Riemann surface structure. Negative refraction. -- We quantify the flow direction of each mode that is propagating in the laminate plane using the angle (Fig. 1) ∂ω/∂kB ∂ω/∂k2 . (cid:104)P1(cid:105) (cid:104)P2(cid:105) = arctan θ = arctan (20) Srivastava (2016) showed analytically for anti-plane shear waves that k2 and P2 share the same sign, and hence so does (cid:104)P2(cid:105). Therefore, the macroscopic transport of energy of anti-plane shear waves in the x2 direction is aligned with the local transport. This implies that in the canonical configuration of excitation (Joseph and Craster, 2015), i.e., when the laminate is impinged by a wave at a boundary normal to the lamination direction (Fig. 1b), anti-plane shear will always refract positively. The reason is that continuity requires the excited Bloch waves to share the same vertical wavenumber as the incident wave, and hence the incident angle and the angle of the Bloch 9 waves have the same sign. In the present problem, where P2 is 1 2 P2 = ωRe{µ (k2 u1 + kB u2 + iu(cid:48) 2) u∗ 1 + (kBλu1 + k2 (λ + 2µ) u2 + iλu(cid:48) , (21) this is no longer the case; as we will demonstrate in the sequel, there are positive k2 for which simultaneously (cid:104)P2(cid:105) < 0 and (cid:104)P1(cid:105) > 0, where ∂x1 1) u∗ 2} , (·) (cid:48) ≡ ∂ (·) P1 = 1 2 ωRe{µ (k2 u1 + kB u2 + iu(cid:48) 2) u∗ 2 + (λ + 2µ) (kB u1 + iu(cid:48) 1) u∗ 1 + λk2 u2 u∗ 1} . (22) This implies that negative refraction is realizable in the simple interface configuration. Willis (2015) devised a complex configuration to achieve negative refraction of anti-plane shear waves, by considering waves impinging at an interface parallel to the lamination direction (Fig. 1c). We will additionally show that in-plane waves can refract negatively in this excitation setup as well. 3 Mode spectrum of an exemplary laminate In this Sec., we study the frequency spectrum of in-plane waves propagating through an infinite laminate comprising steel and acrylic layers, whose properties are µ(1) = 78.85 GPa, λ(1) = 118.27 GPa, ρ(1) = 7800 kg m−3, h(1) = 1.3 mm, ρ(2) = 1200 kg m−3, h(2) = 3 mm. µ(2) = 1.21 GPa, λ(2) = 4.86 GPa, (23) The same laminate was considered by Nemat-Nasser (2015), Willis (2015) and Srivastava (2016) to study anti-plane shear waves. The forthcoming study of the spectrum is the basis for the trans- mission analysis in Sec. 4. To simplify the presentation of the spectrum, we fix one of the three parameters {ω, kB, k2}, and evaluate the relation between the remaining two. Our analysis begins by plotting in Fig. 2 the ordinary frequency f = ω/2π versus 0 < kBh < π at prescribed k2 values. Panels 2(a)-(c) correspond to k2h = 0.5, 0.5i, and 0.5 + 0.5i, respectively. Re kBh and Im kBh (Im kBh in panel c) are shown in solid and dashed curves, respectively, where the real and imaginary parts of a certain branch are plotted using the same color. Gaps are indicated by the red shading. Panel 2(a) exhibits the following notable features. Firstly, there are no branches with real kB from f = 0 to 14.1 kHz. In a study of the dependency of this gap on Re k2h (not shown here), we found that it widens as Re k2h increases. Across 136.2 < f < 181.2 kHz, the branches are complex conjugates of each other; from f = 159.3 kHz the imaginary parts decrease until they vanish at f ≈ 181.2 kHz. At this exceptional point the eigenvalues coalesce. Beyond this point there is a special narrow range where both branches are real, while their slopes have an opposite sign, i.e., 10 (a) (b) (c) Fig. 2. The ordinary frequency f (in kHz) versus kBh restricted to 0 < kBh < π for laminate (23) at (a) k2h = 0.5, (b) 0.5i, and (c) 0.5 + 0.5i. Re kBh and Im kBh (Im kB in panel c) are shown in solid and dashed curves, respectively. The real and imaginary parts of a certain branch are plotted using the same color. Gaps are indicated by the red shading. the modes are propagating in opposite directions; this is the fingerprint of exceptional points. Panel 2(b) exhibits different characteristics. Firstly, across the studied frequencies, one of the branches always has an imaginary part, hence the maximal number of modes that propagate in x1 at any frequency is one. Contrary to the case in panel 2(a), here the first gap emerges above f = 0, namely, at f = 180.6 kHz. In a study of the dependency of this gap on Im k2h (not shown here), we found that gaps starting at f = 0 emerge at higher values of Im k2h. Thus, the diagram starts with a propagating band, where the horizontal group velocity changes sign at f = 148 kHz. This flip of sign inside the irreducible first Brillouin zone is unique to in-plane waves. Panel 2(c) shows an anomalous scenario without passbands, except at a discrete frequency (f ≈ 54.2 kHz) for which one branch is propagating. To facilitate the visual identification of this frequency, here we plot the signed imaginary part instead of its absolute value. Accordingly, this frequency is spotted by the zero crossing of Im kBh. There is no counterpart to this phenomenon in anti-plane waves, where complex k2 are not accessible. Fig. 3 shows the (a) real and (b) imaginary parts of kBh as functions of f at k2h = 0.5, when we carry out an analytic continuation to the dispersion relation, such that the domain of the spectrum is formally extended to complex-valued frequencies. We emphasize that here, this is an artificial continuation (see, e.g., Shanin et al., 2018), as complex frequencies are not accessible (we are concerned with time-harmonic waves in a system with no dissipation). Lu and Srivastava (2018), for example, considered an analytic continuation of the shear modulus, which has the physical interpretation of a system with gain or loss. The spectrum exhibits a structure of a Riemann surface in the vicinity of the exceptional point: Fig. 2(a) is thus a section of that surface, at the plane Im f = 0. In view of the duality of between the frequency and the wavenumbers (Torrent et al., 2018) in the different forms of the eigenvalue 11 00100181.230040050000100200300400500-0.2 00100200300400500(a) (b) Fig. 3. The (a) real and (b) imaginary part of kBh versus the real and imaginary parts of f, for k2h = 0.5. (a) (b) Fig. 4. Pure real kBh branches as function of the real and imaginary parts of k2h for laminate (23) at (a) f = 100 kHz, and (b) 160 kHz. Segments of real, imaginary and complex k2 are denoted by blue, red, and grey, respectively. problem (Mokhtari et al., 2019), a Riemann surface is expected when f is fixed and the roots of kB are evaluated against k2, as we will demonstrate in the sequel. Importantly, the states of the system in the complex wave vector space are accessible, and are not merely a formal extension. Towards this end, we fix f and examine the spectrum in the complex kB − k2 space. Figs. 4(a) and 4(b) show branches of purely real kB versus Re k2h and Im k2h, at f = 100 kHz and 160 kHz, respectively4. Segments of real, imaginary and complex k2 are denoted by blue, red, and grey, respectively. Notably, the number of modes with real or imaginary k2 is finite. In our study (not shown here) the number of modes with real or imaginary k2 increases with the frequency. We further note that in our study on branches with complex k2, we found that their number is infinite, although it cannot be observed from the truncated diagram we show. By contrast, Srivastava (2016) showed that in the anti-plane motion there are no complex k2, and the number of branches with 4The diagrams were evaluated using the extended plane wave method with 51 plane waves in the expansion. A comparison with the exact hybrid matrix method can be found in Appendix C. 12 2.22.6131810181.2-1181.4-0.5010.51810181.2-1181.40410250004102500(a) (d) (b) (e) (c) (f) Fig. 5. The real (solid curves) and imaginary (dashed curves) parts of kBh as functions of Re k2h for laminate (23), at Im k2h = 0 and (a) f = 60, (b) 60.1, (c) 100, (d) 160, (e) 181.3, and (f) 340 kHz. Parts of the same branch are plotted using the same color. imaginary k2 is infinite. In both cases (anti-plane and in-plane waves), the existence of an infinite number of decaying modes conforms with the need of such a set in satisfying the continuity of the displacement and traction across certain interfaces (Srivastava and Willis, 2017). Modes that propagate along both x1 and x2 are studied in Fig. 5, by restricting attention to the plane Im k2h = 0, and evaluating kBh versus Re k2h at prescribed frequencies. The real and imag- inary parts of kB are shown in solid and dashed curves, respectively, where each color corresponds to a different branch. At f = 60 kHz (panel 5a), two modes of different branches are propagating when 0 < kBh < 0.919; only one mode propagates when 0.919 < kBh < 3.101, and at larger Bloch wavenumbers there are no propagating modes. Here and at higher frequencies, we find that beyond the illustrated k2-interval there are no propagating branches -- solutions at higher k2 values are always given by kB with an imaginary part, such that kB is either imaginary or complex. At a slightly higher frequency (f = 60.1 kHz, panel 5b), the whole 1st Brillouin zone admits real solutions, contrary to the case at f = 60 kHz. This change in the spectrum between the frequencies has a significant implication on beam steering, as we will show in Sec. 4.2. In a narrow range near k2h = 0, the blue mode has a non-zero imaginary part and its real part equals π. Thus, for the frequencies in panels 5(a)-(b) and displayed range of k2h values, the Bloch wavenumber of the modes is either real, or has an imaginary part with Re kBh = 0 mod π. Panel 5(c), which evaluates the modes at f = 100 kHz, also shares the latter characteristic. Here, however, the real part of the blue is a non-monotonic function of k2h, as it changes trend twice. This is the fingerprint of an exceptional point, and an indicator of negative refraction as 13 00.511.5203.10100.511.520012300123001234002460will be demonstrated in Sec. 4.2. Similarly to the observation made regarding the modes near the exceptional point in Fig. 2, the slopes of the two modes have different sign. There are accordingly three propagating modes across the range 1.87 (cid:62) kBh (cid:46) 2.03 associated with the same branch, which exhibit different vertical wavelengths. These features are unique to the in-plane motion, as in the anti-plane motion the real part of kBh is either monotonically increasing to π or monotonically decreasing to 0 (Srivastava, 2016). At f = 160 kHz (panel 5d) across the interval 0 < Re k2h (cid:46) 0.79, there are two complex conjugate pairs of kB solutions with and identical real part, up to k2h ≈ 0.79. At this exceptional point, the imaginary part of all the branches vanishes, and the modes coalesce. Beyond this point, the real part of the branches diverges in a different directions, similarly to the feature observed in Fig. 2(a). At k2h ≈ 0.8, the blue branch becomes attenuating, and at k2h ≈ 1 the green branch also becomes imaginary. The blue mode is propagating again for 3.08 < k2h < 3.22; beyond that range there no more propagating modes. When f = 181.3 kHz (panel 5e), there are two exceptional points at k2h = 0.503 and 0.93. Again, in the vicinity of these points there are modes with slopes of different sign. As in panel 2(d), there is a range of Bloch wavenumbers with three propagating modes when kBh > 1.26, where for lower kBh there are two propagating modes. We recall that in Fig. 2(a), we exhibited an exceptional point for f = 181.2 kHz and k2h = 0.5, i.e., when perturbing one of the parameters of the system about this point (e.g., k2h is perturbed from 0.5 to 0.503), another exceptional point emerges at a variation of another parameter of the system (e.g., is perturbed f from 181.2 to 181.3 kHz). This is not accidental; in fact, the excep- tional points we demonstrate for each set of parameters comprise exceptional curves in a higher dimensional space. At f = 340 kHz (panel 5f), the propagating green mode is a non-monotonic function of k2, where initially it has a positive slope up to a maximal point, beyond which the slope changes sign, until the mode becomes imaginary at k2h = 1.72. The blue mode is either real or imaginary with Re kBh mod π = 0. The number of propagating modes for prescribed kB is higher than at lower frequencies, namely, a minimal number of three and a maximal number of five across 1.624 < kBh < 1.88. As mentioned earlier, we expect the spectrum to exhibit the structure of a Riemann surface near exceptional points in the complex kB − k2 space. This is shown in Fig. 6, where the (a) real and (b) imaginary parts of kBh are plotted against Re k2h and Im k2h, at f = 160 kHz: Fig. 5(d) is thus the section Im k2 = 0 of this surface. We emphasize again the the states of the system over this manifold are accessible since they correspond to realizable wavenumbers, and are not merely a formal extension. Finally, in Fig. 7 we fix kBh and evaluate the spectrum in the f − k2 space. Specifically, panels 14 (a) (b) Fig. 6. The (a) real and (b) imaginary parts of kBh as functions of Re k2h and Im k2h, at f = 160 kHz. (a) (b) (c) Fig. 7. The ordinary frequency f (in kHz) versus real and imaginary parts of k2h solutions for laminate (23) at (a) kBh = 0.5, (b) 0.5i, and (c) 0.5 + 0.5i. Segments of real, imaginary and complex k2 are denoted by blue, red, and grey, respectively. 7(a)-(c) correspond to kBh = 0.5, 0.5i and 0.5 + 0.5i, respectively; segments of real, imaginary and complex k2 are denoted by blue, red, and grey, respectively. We observe that the number of modes in general -- and propagating in particular -- increases with frequency. In panel 7(a) we observe an exceptional point when k2h = 0.9i at 13 kHz, where a single complex mode branches to two imaginary modes; as frequency increases to 33 kHz, one of these branches becomes a propagating mode5. A similar exceptional point is identified in panel 7(b), when k2h = 2.3i at 72 kHz. There are no propagating modes in panel 7(c). 4 Transmission across an interface of a semi-infinite laminate As mentioned, modes with either prescribed kB or prescribed k2 are excitable through two different configurations. We analyze the transmission and excited modes in these configurations, linking them to the exceptional points, negative refraction, and beam steering found in the spectrum, as reported in Sec. 3. 5Laude et al. (2009) discussed how the shifting of modes between the complex and real domains serves as a mech- anism to preserve the total number of modes at a prescribed frequency. 15 0.781.82.60.10.793.100.8-0.10.78-100.10.79100.8-0.10410062420004100624200041006242004.1 Metamaterial refraction via an interface parallel to the layers To excite modes with designated vertical lengths, we connect the laminate at x1 = 0 to a homo- geneous half-space whose properties are denoted by the script 0, from which an incident wave propagates in an angle θi towards the interface (Fig. 1b). The wave is described by the potential V V = Iei(ωt−k sin θix1−k cos θix2), (24) S for a shear wave, or ω/c(0) where the wavenumber k = ω/c(0) is either ω/c(0) L for a pressure wave. The incident displacements are derived according to Eq. (2), by setting ψ(0) = V for incident shear, or φ(0) = V for incident pressure wave. The incident wave is partially reflected back by the interface, and partially transmitted to the laminate as Bloch modes. The continuity conditions at x1 = 0 enforces all these waves to have an x2-dependency in the form e−ik cos θix2. Corresponding Bloch modes are extracted by the procedure in Sec. 2.1 with k2 = k cos θi, in the form 2(cid:88) m=1 T = Tm u(m) (x1) ei(ωt−kBx1−k cos θix2), (25) (26) (cid:17) , where the reflected waves are derived from (cid:16) R = Rφei φ(0) ωt+k(0) L x1−k cos θix2 (cid:17) , ψ(0) R = Rψei (cid:16) ωt+k(0) S x1−k cos θix2 L and k(0) S are related to k2 = k cos θi via Eq. (5). Finally, the reflection coefficients l and transmission coefficients Tm are determined from the continuity of the state vector at such that k(0) Rφ l , Rψ x1 = 0. We numerically demonstrate how negative refraction is obtained, using a half-space with the properties µ(0) = 0.178 GPa, λ(0) = 0.714 GPa, ρ(0) = 3000 kg/m3. (27) We set an incident pressure wave at θi = 77.22◦ and f = 100 kHz, for which k2h = 1. The resultant Bloch wavenumbers are extractable from Fig. 5, where at the pertinent k2 there is one attenuating mode and one propagating mode. We recall that the slope in Fig. 5 of the propagating mode at this k2 is positive, while at higher k2 the slope is negative, which is an indicator of negative refraction. This is verified by evaluating Eqs. (21)-(22), to find that the propagating angle is θ = 107.03◦. As a consistency check, we examine the balance of energy at x1 = 0, namely, −(cid:12)(cid:12)Rφ(cid:12)(cid:12)2 (cid:104)P1(cid:105)φ −(cid:12)(cid:12)Rψ(cid:12)(cid:12)2 (cid:104)P1(cid:105)ψ (cid:35) = 1, (28) (cid:34) 2(cid:88) 1 I2 (cid:104)P1(cid:105)I Tm2 (cid:104)P1(cid:105)(1) Tm m=1 16 Fig. 8. Frequency domain finite element simulations using COMSOL Multiphysics(cid:114) of the energy flux along x2 (P2) resulting from an (a) incident pressure wave at θi = 77.22◦ and f = 100 kHz, and (b) an incident shear incident wave at θi = 89.62◦ and f = 340 kHz. The obtained negative angles of refraction agree with the analytical prediction. where the scripts Tm, φ, ψ and I denote quantities associated with the corresponding fields. Indeed, we find that the terms in left hand side are 0.617, 0.316 and 0.067, respectively, which sum to 1. Note that only the propagating mode contributes to the first term, since the energy flux of the mode with the imaginary kB vanishes. As a second example, we set an incident shear incident wave at θi = 89.62◦ and f = 340 kHz, for which k2h = 0.25. The resultant Bloch wavenumbers are extractable from Fig. 5(f). Here again, at the prescribed k2 there is one attenuating mode, and one propagating mode with a positive slope, that will change sign at greater k2. Calculation of Eqs. (21)-(22) confirms a negative refraction at θ = 106.2◦. Energy balance is also verified, where the calculation of the terms in the left hand side of Eq. (28) provides 0.148, 0.007 and 0.845, respectively, which again sum to 1. Fig. 8 shows frequency domain finite element simulations using COMSOL Multiphysics(cid:114) the energy flux along x2 (P2) resulting from an (a) incident pressure wave at θi = 77.22◦ and f = 100 kHz, and (b) an incident shear incident wave at θi = 89.62◦ and f = 340 kHz. We used low reflecting boundary conditions to avoid reflections due to the finite truncation of the spatial domain. The incident wave was excited using load control over a defined line. We used triangular elements, whose maximal size was third of the unit-cell thickness. Indeed, the resultant negative angles of refraction agree with the analytical prediction. 4.2 Negative refraction via an interface normal to the layers It is possible to excite designated Bloch wavenumbers using an incident wave from a homoge- neous half-space that is bonded to the laminate by an interface along the lamination direction, say at x2 = 0 (Fig. 1c). This configuration was proposed by Willis (2013) to achieve negative refrac- 17 tion of anti-plane shear waves, and was studied in greater detail by Willis (2015), Nemat-Nasser (2015), Srivastava (2016), and Morini et al. (2019). We recall that in this setup, an incident wave will excite an infinite number of decaying waves in the laminate, required to enforce continuity conditions (Srivastava and Willis, 2017). Here, we firstly develop a method to resolve the normal mode decomposition and energy partition of the excited in-plane waves in this configuration. Consider again an incident wave according to the potential (24), now impinging on the horizon- tal interface x2 = 0. In this setting, the continuity conditions at the interface enforce the reflected and transmitted waves to share the same horizontal length with the incident wave. Accordingly, the transmitted waves have the form T = Tm u(m) (x1) ei(ωt−kBx1−kmx2), (29) ∞(cid:88) m=1 where kB = k sin θi, and {km} are extractable from Eq. (17). In view of the foregoing observations, the reflected waves is constructible from potentials that must have the form φR = l Ul (x1) ei(ωt−kBx1+κφ Rφ l x2), ψR = l Ul (x1) ei(ωt−kBx1+κψ Rψ l x2), (30) ∞(cid:88) l=−∞ where κφ l = (cid:114)(cid:16) ω/c(0) L and Ul (x1) = e−2lπix1/h. (cid:17)2 − (k sin θ + 2πl/h)2, κψ (cid:110) conditions at x2 = 0, which are compactly written as l = ω/c(0) S (31) (cid:17)2 − (k sin θ + 2πl/h)2, (cid:111) The transmission and reflection coefficients Tm, Rφ l , Rψ l are determined by the continuity l s⊥ Rφ φ (l) − l s⊥ Rψ ψ (l) = Is⊥ I , (32) ∞(cid:88) Tms⊥ T (m) − ∞(cid:88) s⊥(n) =(cid:0)u1 l=−∞ m=1 using ∞(cid:88) (cid:114)(cid:16) l=−∞ ∞(cid:88) l=−∞ (n)(cid:1)T (33) For computational purposes, we truncate the sums over l and m such that −NR ≤ l ≤ NR and 1 ≤ m ≤ 2 (2NR + 1), and apply the elegant orthogonality relation6 (Mokhtari et al., 2019) (n), σ21 (n), σ22 (n), u2 . (cid:2)s⊥ (cid:90) h 6Alternatively, we can use the Fourier orthogonality(cid:82) h ∗(α) −(cid:0)s⊥ ∗(α) + s⊥ (β)s⊥ (β)s⊥ (β)s⊥ 3 2 4 0 1 3 ∗(α)(cid:1)(cid:3) dx1 = ∗(α) + s⊥ 4 (β)s⊥ 2 1 0 Uα (x1) U∗ β (x1) dx1 = hδαβ. (cid:104)P2(cid:105) (α)δαβ, 4ih ω (34) 18 (a) (b) Fig. 9. The transmitted angle θ of Bloch modes in laminate (23), induced by a pressure wave impinging at x2 = 0 from the half-space (37), as function of incoming angle θi (upper axis) and its corresponding kBh (lower axis). The frequency is 60 kHz in panel (a), and 60.1 kHz in panel (b). The colors identify corresponding branches of the same color in Fig. 5. to obtain an algebraic system of equations for Tm, Rφ l , Rψ l in the form (cid:110)   Tm Rφ l Rψ l (cid:111)  = I  ,  Is T Is φ Is ψ  Ms Ms Ms T T Ms φT Ms 0 ψT T φ Ms φφ 0 T ψ Ms ψψ (35) T T is with diagonal submatrices whose elements are given in Appendix D; thus, the size of Ms 2 (2NR + 1) × 2 (2NR + 1), of Ms ψT is φT and Ms (2NR + 1) × 2 (2NR + 1), and the size of the remaining submatrices is (2NR + 1) × (2NR + 1). In summary, Eq. (35) constitutes 4 (2NR + 1) algebraic equations for 4 (2NR + 1) scattering coef- ficients. The resultant coefficients should satisfy the following energy balance T ψ is 2 (2NR + 1) × (2NR + 1), of Ms T φ and Ms (cid:34)(cid:88) T −(cid:88) (cid:12)(cid:12)(cid:12)Rφ l (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(l) φ −(cid:88) (cid:12)(cid:12)(cid:12)Rψ l (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(l) ψ (cid:35) 1 I2 (cid:104)P2(cid:105)I Tm2 (cid:104)P2(cid:105)(m) m l l = 1. (36) An example of the fields at the interface as obtained from the method is provided in Appendix E. We proceed to analyze the transmission of Bloch modes in laminate (23) that are induced by impinging pressure wave from the half-space µ(0) = 0.033 GPa, λ(0) = 0.132 GPa, ρ(0) = 3000 kg/m3. (37) To this end, we plot in Fig. 9 the propagation angle θ of the modes analyzed in Fig. 5, as functions of the incoming angle θi and its corresponding kBh, indicated by the upper and lower axes, respec- tively. The calculation of θ was carried out through direct evaluation of Eqs. (21)-(22) for calculate arctan (cid:104)P1(cid:105) (cid:104)P2(cid:105). Panel 9(a) is for f = 60 kHz, which corresponds to Fig. 5(a), and uses the same color legend. 19 0° 29.546° 30.388°87.414°03.101 3.1822-90°-60°-30°0°30°60°90°0° 29.911°85.804°02-90°-60°-30°0°30°60°90°Fig. 10. The transmitted angle θ of Bloch modes in laminate (23), induced by a pressure wave impinging at x2 = 0 from the half-space (37), as function of incoming angle θi (upper axis) and its corresponding kBh (lower axis), f = 181.3 kHz. Purple denotes the solutions with k2h < 0.5, where grey denotes solutions with 0.93 < k2h < 1.21. Segments above and below the exceptional points Fig. 5(e) are marked using circle marks and diamond marks, respectively. Our first observation concerns the change of sign in θ with respect to kBh = π, owing to the 2π- periodicity of the spectrum in Re kBh, and its and reflection symmetry within that Brillouin zone period. Accordingly, ∂ω/∂kB changes sign, and hence so does (cid:104)P1(cid:105), as noted by Willis (2015) in his analysis of anti-plane shear. The diagram exhibits 2-fold rotational symmetry about (π, 0), which corresponds to θi ≈ 30◦. Notably, the transmission angle of the blue mode increases very fast near kBh = 3.101, and then discontinuously flips from 90◦ to −90◦ at kBh = 3.182. This discontinuity occurs since the blue branch does not exist (a gap) when 3.101 < kBh < π, as highlighted in Fig. 5(a). Therefore, in the vicinity θi ≈ 30◦, it is sufficient to slightly change the incident angle in order to significantly steer the transmitted mechanical beam. Wide steering by small changes of the incident angle, albeit less extreme, occurs also about θi ≈ 10◦ and 55◦. We further note that since there is a range of kBh -- and hence of incident angles -- in which there is only one propagating mode (see Sec. 3), it is possible to achieve purely negative transmission. Analogous observations were made by Srivastava (2016) for anti-plane shear waves. Panel 9(b) is for f = 60.1 kHz, which corresponds to Fig. 5(b). We observe that the discon- tinuous sign flip of θ and significant beam steering are lost, as the θ (kBh)-curve smoothly passes through the point (0, π). This result agrees with the extension of the blue branch in Fig. 5(b) to π near k2h = 0, where it has an imaginary part, hence becomes attenuating. Thus, in the vicinity of f = 60 kHz, energy transport is also very sensitive to the excitation frequency. Fig. 10 is for f = 181.3 kHz, corresponding to Fig. 5(e), which we recall exhibits two ex- ceptional points; a zoom in about these points is shown in the right side of the figure. The color legend is different here: purple denotes solutions with k2h < 0.5, i.e., left to the first excep- tional point in Fig. 5(e), where grey denotes solutions right to the second exceptional point, with 20 6.768°9.514°12.283°2.244.04-5°0°5°0°9.514°19.305°02-90°-60°-30°0°30°60°90°Zoom in0.93 < k2h < 1.21. (Note that we do not include the third propagating branch with 3.4 < k2h.) We mark the segments above and below the exceptional points Fig. 5(e) using circle marks and diamond marks, respectively. Up to the first exceptional point (kBh ≈ 2.564, θi ≈ 7.753◦), the green branch of Fig. 5(e) ad- mits two propagation angles which -- remarkably -- one of them is negative inside the first Brillouin zone. Beyond the exceptional point, mode switching occurs between the green branch (purple diamonds) and the blue branch (purple circles) of Fig. 5(e), as the former becomes attenuating and the latter becomes propagating with positive refraction. Similar switching occurs between the green and blue branches beyond the second exceptional point (kBh ≈ 3.021, θi ≈ 9.146◦). These phenomena are unique to the in-plane motion, owing to exceptional points, and are not accessible in the anti-plane setting. Finally, we note that here again, the diagram exhibits 2-fold rotational symmetry about (π, 0), which corresponds to θi ≈ 9.514◦, followed by the inversion of the afore- mentioned phenomena with respect to the incident angle.s 5 Conclusions We have revisited the problem of in-plane waves propagation in periodic laminates, were the mo- tivation was twofold. Firstly, this study is a necessary complement to the reports of Willis (2015), Nemat-Nasser (2015) and Srivastava (2016) on metamaterial phenomena in the model problem of anti-plane waves traversing elastic laminates. Secondly, this problem allows us to show that the coupling between shear and pressure parts can be harnessed for anomalous energy transport, within a relatively simple analytical study. We have shown that the corresponding spectrum contains exceptional points at which two Bloch modes coalesce, and further showed that these points are the source of anomalous energy transport. To this end, we have determined how energy is scattered when an incident in-plane wave impinges the laminate in two model interface problems. We found metamaterial transmission through the laminate, such as pure negative refraction, and beam splitting and steering, at states of the system near the exceptional points. Notably, we achieved negative refraction in the canonical transmission problem, where the laminate is impinged by an incoming wave from a homogeneous medium whose interface with the laminate is parallel to the layers. We emphasize again that these phenomena emerge from the unique coupling in elastodynamics between the volumetric and distortional modes of deformation, which cannot be observed in sound and light waves. Our work further paves the way for encircling exceptional points in a tangible, purely elastic apparatus, for applications such as asymmetric mode switches. 21 Acknowledgments We are grateful to Profs. Nimrod Moiseyev, Alexei Mailybaev and Ankit Srivastava for fruitful dis- cussions. We thank Dr. Pernas-Salomón for his help regarding the hybrid matrix method. We ac- knowledge the support of the Israel Science Foundation, funded by the Israel Academy of Sciences and Humanities (Grant no. 1912/15), and the United States-Israel Binational Science Foundation (Grant no. 2014358), and Ministry of Science and Technology (Grant no. 880011). Appendix A The forthcoming formulation is an adaptation of the developments of Shmuel and Pernas-Salomón (2016) and Pernas-Salomón and Shmuel (2018) to the current problem. Here, the modified state vector -- consisting of quantities which are continuous across the interface between two adjacent layers -- is sm (x1, x2) = (u1 (x1, x2) Cu1, σ11 (x1, x2) Cσ11, u2 (x1, x2) Cu2, σ21 (x1, x2) Cσ21)T , (A.1) where the normalizing coefficients required to stabilize subsequent calculations are (cid:12)(cid:12)(cid:1)−1 , (cid:12)(cid:12)(cid:1)−1 , Cu1 = k2−1 , Cσ11 =(cid:0)¯λk2(cid:12)(cid:12) ¯kL Cu2 = k2−1 , Cσ21 =(cid:0)¯µk2(cid:12)(cid:12) ¯kS (cid:16) (cid:16) (cid:17) (cid:17) ¯µ =(cid:0)µ(a) + µ(b)(cid:1) /2, ¯λ =(cid:2)(cid:0)λ(a) + µ(a) + λ(b) + µ(b)(cid:1) /2 + ¯µ(cid:3) /2. sm (x1, x2) = Q (x1, x2) ·(cid:16) k(a) S + k(b) k(a) L + k(b) ¯kS = ¯kL = L+, A(n) A(n) S+, A(n) L−, A(n) S− (A.2) (A.3) (A.4) (cid:17)T , and The state vector can be written as follows /2, /2, S L 22 where Q (x1, x2) = (1/Cu1, 1/Cσ11, 1/Cu2, 1/Cσ21)· L 2 −(cid:0)λ(p) + 2µ(p)(cid:1) k2(p) µ(p)(cid:16) S x1(cid:17) −ik2 2µ(p)k(p) −ik(p) L x1, e−ik(p) S x1, e−ik(p) ik(p) −λ(p)k2 −ik2 2µ(p)k(p) eik(p) L k2 L x1, eik(p) L S  ·(cid:16) S k2 S − k2 k2(p) e−ik2x2. 2 2 −(cid:0)λ(p) + 2µ(p)(cid:1) k2(p) L L −ik(p) −λ(p)k2 −ik2 (cid:17) −2µ(p)k(p) L k2  (cid:17) −ik2 −2µ(p)k(p) ik(p) µ(p)(cid:16) S S k2 S − k2 k2(p) 2 Q (x1, x2) can also be defined as (cid:32) Q (x1, x2) = F1 (x1, x2) F2 (x1, x2) F3 (x1, x2) F4 (x1, x2) A1 (x1, x2) A2 (x1, x2) A3 (x1, x2) A4 (x1, x2) (cid:33) . (A.5) (A.6) Using this definition the hybrid matrix of the nth layer is given by H(n)(cid:0)h(p)(cid:1) = (cid:32) (cid:32) F1 (x0, x2) A1 (cid:0)x0 + h(p), x2 (cid:0)x0 + h(p), x2 A1 (x0, x2) F1 F4 (x0, x2) A4 (x0, x2) . (cid:1) (cid:33) (cid:0)x0 + h(p), x2 (cid:1) (cid:33)−1 (cid:0)x0 + h(p), x2  , (A.8) F3 (x0, x2) (cid:0)x0 + h(p), x2 (cid:1) A4 (cid:0)x0 + h(p), x2 (cid:1) F4 A3 (x0, x2) F2 (x0, x2) A2 (x0, x2) (cid:1) A2 (cid:0)x0 + h(p), x2 (cid:1) A3 (cid:1) F2 (cid:0)x0 + h(p), x2 (cid:1) F3   = H(n)(cid:0)h(p)(cid:1) · (A.7) Using the definition in Eq. (A.7), we relate the field variables which appear in the modified state vector at the ends x0 and x0 + h(p) of the nth layer via  where H(n)(cid:0)h(p)(cid:1) is the 4 × 4 hybrid matrix of the nth layer. In order to calculate the total hybrid (n)(cid:0)x0 + h(p), x2 (cid:1) Cu1 (cid:1) Cσ11 (n)(cid:0)x0 + h(p), x2 (n) (x0, x2) Cu1 (n) (x0, x2) Cσ11 (n)(cid:0)x0 + h(p), x2 (cid:1) Cu2 (cid:1) Cσ21 (n)(cid:0)x0 + h(p), x2 (n) (x0, x2) Cu2 (n) (x0, x2) Cσ21 u2 σ21 u1 σ11 u1 σ11 u2 σ21 matrix of n layers, denoted by H(1,n), a recursive algorithm is used in terms of the hybrid matrix of 23 the first n − 1 layers (H(1,n−1)) and the hybrid matrix of the nth layer H(n), namely, H(1,k) 22 = H(k) 22 + H(k) 21 ·(cid:104) H(1,k) 21 = H(k) 12 = H(1,k−1) H(1,k) 11 = H(1,k−1) H(1,k) 11 12 21 ·(cid:104) 11 · H(k) (cid:105)−1 · H(1,k−1) (cid:105)−1 · H(1,k−1) 11 ·(cid:104) 11 ·(cid:104) I2 − H(1,k−1) · H(k) I2 − H(1,k−1) · H(k) 11 11 21 22 22 22 22 I2 − H(1,k−1) · H(k) 12 + H(1,k−1) · H(k) I2 − H(1,k−1) · H(k) + H(1,k−1) 12 12 22 , · H(k) 12 , (cid:105)−1 · H(1,k−1) 22 · H(k) (cid:105)−1 · H(1,k−1) 11 21 , (A.9) · H(k) 12 , Where H(1,k) layers. The generalized eigenproblem presented in Eq. (8) yields the characteristic equation denote the 2 × 2 blocks of the hybrid matrix H(1,k) of the first k , H(1,k) , H(1,k) , H(1,k) 22 12 21 11 a1 + λa2 + λ2a3 + λ3a4 + λ4a5 = 0 (A.10) for λ = eikBh, where a1 = h13h24 − h14h23 = det H12, a2 = −h13h42h31 − h24h42h31 + h12h43h31 + h22h44h31 − h31 + h13h32h41 + h24h32h41 − h12h33h41 − h22h34h41 + h11h33h42 + h21h34h42 − h42 − h11h32h43 − h21h32h44, a3 = 1 + h13h31 + h24h31 − h14h23h42h31 + h13h24h42h31 + h14h22h43h31 − h12h24h43h31 − h13h22h44h31 + h12h23h44h31 − h11h33 − h21h34 + h14h23h32h41 − h13h24h32h41 − h14h22h33h41 + h12h24h33h41 + h13h22h34h41 − h12h23h34h41 + h13h42 + h24h42 + h14h21h33h42 − h11h24h33h42 − h13h21h34h42 + h11h23h34h42 − h12h43 − h14h21h32h43 + h11h24h32h43 + h12h21h34h43 − h11h22h34h43 − h22h44 + h13h21h32h44 − h11h23h32h44 − h12h21h33h44 + h11h22h33h44, a5 = a1, a4 = a2, (A.11) and hij are the components of the total hybrid matrix. The latter two equalities imply that λ−1 = e−ikBh is also a solution, as expected. Eq. (A.10) yields a closed-form expression for cos kBh, given in Eq. (9). Appendix B The matrices A0, A1, A2 and B appearing in Eq. (16) are given by 24 Fig. 11. real kBh as function of the real and imaginary parts of k2h for laminate (23) and f = 100 kHz, using the EPWE and Hybrid matrix method, depicted in purple and blue points, respectively. (cid:32) A0 = A11 0 0 0 A22 0 (cid:33) (cid:32) , A1 = 0 A21 1 A12 1 0 (cid:33) (cid:32) , A2 = A11 2 0 0 A22 2 (cid:33) (cid:32) , B = B11 0 0 B22 (cid:33) , (B.1) where the (G, G(cid:48)) component of each block is 1 G,G(cid:48) = − (G + kB) [λ (G − G(cid:48)) + µ (G − G(cid:48))] , 0 GG(cid:48) = − [(G + kB) (G(cid:48) + kB)] [λ (G − G(cid:48)) + 2µ (G − G(cid:48))] , A11 0 GG(cid:48) = − [(G + kB) (G(cid:48) + kB)] µ (G − G(cid:48)) , A22 A12 1 GG(cid:48) = A21 2 GG(cid:48) = −µ (G − G(cid:48)) , A11 2 GG(cid:48) = −λ (G − G(cid:48)) − 2µ (G − G(cid:48)) , A22 GG(cid:48) = B22 B11 G,G(cid:48) = −ρ (G − G(cid:48)) . (B.2) Appendix C This appendix demonstrate the applicability of the extended plane wave expansion method, by comparing its results with the exact results of the hybrid matrix method. Fig. 11 shows real kBh as function of the real and imaginary parts of k2h for laminate (23) and f = 100 kHz. Points calculated with the extended plane wave expansion method and hybrid matrix method are depicted 25 in purple and blue, respectively. We used 51 plane waves for the extended plane wave expansion method and points calculated using the hybrid matrix method differ from the results of the hybrid matrix by 5 × 10−2. Appendix D The components of the submatrices in Eq. 35 are φ3 ∗(m) −(cid:0)s⊥ ∗(m) −(cid:0)s⊥ ∗(l) −(cid:0)s⊥ ∗(l) −(cid:0)s⊥ T3 ∗(m)(cid:1)(cid:3) dx1, ∗(m)(cid:1)(cid:3) dx1, ∗(l)(cid:1)(cid:3) dx1, ∗(l)(cid:1)(cid:3) dx1, (B.1) Ms T Tmm = T1 0 0 = = = = 4ih ω 0 4ih ω (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ (cid:90) h (cid:2)s⊥ 0 4ih ω = = 0 0 0 Ms T φml Ms T ψml Ms φTlm Ms φφll Ms ψTlm Ms ψψll Is Tm = Is φl = Is ψl = T (cid:104)P2(cid:105)(m) (l)s⊥ φ1 T3 , ∗(m) + s⊥ φ2 (l)s⊥ T4 (l)s⊥ T1 ∗(m) + s⊥ φ4 (l)s⊥ T2 (l)s⊥ T3 ∗(m) + s⊥ ψ2 (l)s⊥ T4 ψ1 (l)s⊥ T1 ∗(m) + s⊥ ψ4 (l)s⊥ T2 ψ3 (m)s⊥ φ3 ∗(l) + s⊥ T2 (m)s⊥ φ4 (m)s⊥ φ1 ∗(l) + s⊥ T4 (m)s⊥ φ2 (cid:104)P2(cid:105)(l) φ , (m)s⊥ T1 ψ3 ∗(l) + s⊥ T2 (m)s⊥ ψ4 (m)s⊥ ψ1 ∗(l) + s⊥ T4 (m)s⊥ ψ2 T3 T3 (cid:104)P2(cid:105)(l) ψ , ∗(m) + s⊥ I1s⊥ I1s⊥ I1s⊥ I2s⊥ I2s⊥ I2s⊥ ∗(l) + s⊥ ∗(l) + s⊥ ∗(m) −(cid:0)s⊥ ∗(l) −(cid:0)s⊥ ∗(l) −(cid:0)s⊥ I3s⊥ I3s⊥ ψ4 ψ3 φ3 φ4 T4 φ1 ψ1 T2 ∗(m)(cid:1)(cid:3) dx1, ∗(l)(cid:1)(cid:3) dx1, ∗(l)(cid:1)(cid:3) dx1, ∗(m) + s⊥ I4s⊥ T1 I3s⊥ ∗(l) + s⊥ φ2 I4s⊥ ∗(l) + s⊥ I4s⊥ ψ2 where subscript numbers denote the component of s⊥, and subscript letters denote the potential from which this component is derived. Appendix E Fig. 12 shows s⊥ 4 (panel d, nor- malized by k¯λ) at the interface x2 = 0 as calculated via Eq. (35). Specifically, solid and dashed curves correspond to the laminate and homogeneous half-space, respectively, where the real and imaginary parts of each field are denoted by blue and purple, respectively. The laminate and homo- 3 (panel c, normalized by k ¯µ), and s⊥ 2 (panel b), s⊥ 1 (panel a), s⊥ 26 (a) (c) (b) (d) (a) s⊥ 1 (b) s⊥ 2 (c) s⊥ 3 (d) s⊥ 4 at the interface x2 = 0 between laminate (23) and the homoge- Fig. 12. neous half-space (27), for f = 100 kHz, θi = 15.72◦ an incident shear wave. The fields in the laminate (resp. homogeneous half-space) fields are depicted in continuous (dashed) curves, where the real (imaginary) part is depicted in blue (purple). geneous half-space that was used have properties (23) and (27), respectively, and the calculation was carried out for f = 100 kHz, θi = 15.72◦, and an incident shear wave. For this example, we chose NR = 12, which ensures all the propagating transmitted and reflected modes in the x2 direction are included; the rest of the transmitted modes are those to have the lowest absolute value of Imk2h. The solution obtained does not match the fields perfectly, however it is sufficient on average provides a good approximation. The solution yields the following terms for the energy balance (36), T12 (cid:104)P2(cid:105)(1) (cid:12)(cid:12)(cid:12)Rφ−1 (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(−1) (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(0) (cid:12)(cid:12)(cid:12)Rφ (cid:12)(cid:12)(cid:12)Rψ−2 (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(−2) ψ φ 0 T /I2 (cid:104)P2(cid:105)I = 0.049, /I2 (cid:104)P2(cid:105)I = −0.091, φ /I2 (cid:104)P2(cid:105)I = −0.391, /I2 (cid:104)P2(cid:105)I = 0, (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(−1) (cid:12)(cid:12)(cid:12)Rψ−1 (cid:12)(cid:12)(cid:12)Rψ (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(0) (cid:12)(cid:12)(cid:12)2 (cid:104)P2(cid:105)(1) (cid:12)(cid:12)(cid:12)Rψ ψ 1 0 /I2 (cid:104)P2(cid:105)I = −0.0133, ψ /I2 (cid:104)P2(cid:105)I = −0.437, ψ /I2 (cid:104)P2(cid:105)I = −0.007, (B.1) which sum to 0.988: a difference of only 0.12%. In this case, the transmitted mode refracts posi- tively with transmitted angle of θ = 1.12◦, however carries only a small fraction of the energy of the incident wave. 27 01.34.3-0.4-0.200.20.40.60.801.34.3-0.6-0.4-0.200.201.34.3-0.04-0.0200.020.0401.34.3-0.04-0.0200.020.040.060.08References V. Achilleos, G. Theocharis, O. Richoux, and V. Pagneux. Non-hermitian acoustic metamate- rials: Role of exceptional points in sound absorption. Phys. Rev. B, 95:144303, Apr 2017. doi: 10.1103/PhysRevB.95.144303. URL https://link.aps.org/doi/10.1103/ PhysRevB.95.144303. S D M Adams, R V Craster, and S Guenneau. Bloch waves in periodic multi-layered acoustic waveguides. Proc. R. Soc. London A, 464(2098):2669 -- 2692, 2008. Samuel D.M. Adams, Richard V. Craster, and Sebastien Guenneau. Guided and standing bloch waves in periodic elastic strips. Waves in Random and Complex Media, 19(2):321 -- 346, 2009. doi: 10.1080/17455030802541566. URL https://doi.org/10.1080/ 17455030802541566. Fateme Aghighi, Joshua Morris, and Alireza V. Amirkhizi. Low-frequency micro-structured me- chanical metamaterials. Mechanics of Materials, 130:65 -- 75, 2019. ISSN 0167-6636. doi: https: //doi.org/10.1016/j.mechmat.2018.12.008. URL http://www.sciencedirect.com/ science/article/pii/S0167663618306483. Biswajit Banerjee. An introduction to metamaterials and waves in composites. CRC Press, 2011. D Bigoni, S Guenneau, A B Movchan, and M Brun. Elastic metamaterials with inertial locally resonant structures: Application to lensing and localization. Phys. Rev. B, 87(17):174303, 2013. doi: 10.1103/PhysRevB.87.174303. URL https://link.aps.org/doi/10. 1103/PhysRevB.87.174303. G. Bordiga, L. Cabras, A. Piccolroaz, and D. Bigoni. Prestress tuning of negative refraction and wave channeling from flexural sources. Applied Physics Letters, 114(4):041901, 2019. doi: 10.1063/1.5084258. URL https://doi.org/10.1063/1.5084258. M Brun, S Guenneau, A B Movchan, and D Bigoni. Dynamics of structural interfaces: Filtering and focussing effects for elastic waves. J. Mech. Phys. Solids, 58(9):1212 -- 1224, 2010. Huanyang Chen, C. T. Chan, and Ping Sheng. Transformation optics and metamaterials. Nature Materials, 9:387 EP -- , 04 2010. URL https://doi.org/10.1038/nmat2743. Qianli Chen and Ahmed Elbanna. Emergent wave phenomena in coupled elastic bars: from Scientific Reports, 7(1): extreme attenuation to realization of elastodynamic switches. 16204, 2017. doi: 10.1038/s41598-017-16364-8. URL https://doi.org/10.1038/ s41598-017-16364-8. 28 Y Chen, G Hu, and G Huang. A hybrid elastic metamaterial with negative mass density and tunable bending stiffness. J. Mech. Phys. Solids, 105:179 -- 198, 2017. ISSN 0022-5096. doi: https://doi. org/10.1016/j.jmps.2017.05.009. URL http://www.sciencedirect.com/science/ article/pii/S0022509617301229. J Christensen, M Willatzen, V R Velasco, and M.-H. Lu. Parity-Time Synthetic Phononic Media. Phys. Rev. Lett., 116(20):207601, 2016. doi: 10.1103/PhysRevLett.116.207601. URL https: //link.aps.org/doi/10.1103/PhysRevLett.116.207601. D J Colquitt, M Brun, M Gei, A B Movchan, N V Movchan, and I S Jones. Transforma- J. Mech. Phys. Solids, 72:131 -- 143, ISSN 0022-5096. doi: http://dx.doi.org/10.1016/j.jmps.2014.07.014. URL http: tion elastodynamics and cloaking for flexural waves. 2014. //www.sciencedirect.com/science/article/pii/S0022509614001586. Richard V Craster and Sébastien Guenneau. Acoustic metamaterials: Negative refraction, imaging, lensing and cloaking, volume 166. Springer Science & Business Media, 2012. Steven A. Cummer, Johan Christensen, and Andrea Alù. Controlling sound with acoustic metama- terials. Nature Reviews Materials, 1:16001 EP -- , 02 2016. URL https://doi.org/10. 1038/natrevmats.2016.1. Pierre A Deymier. Acoustic metamaterials and phononic crystals, volume 173. Springer Science & Business Media, 2013. Kun Ding, Z Q Zhang, and C T Chan. Coalescence of exceptional points and phase dia- grams for one-dimensional PT-symmetric photonic crystals. Phys. Rev. B, 92(23):235310, 2015. doi: 10.1103/PhysRevB.92.235310. URL https://link.aps.org/doi/10. 1103/PhysRevB.92.235310. Jörg Doppler, Alexei A Mailybaev, Julian Böhm, Ulrich Kuhl, Adrian Girschik, Florian Libisch, Thomas J Milburn, Peter Rabl, Nimrod Moiseyev, and Stefan Rotter. Dynamically encircling an exceptional point for asymmetric mode switching. Nature, 537:76 EP -- , 2016. URL https: //doi.org/10.1038/nature18605. Ramy El-Ganainy, Konstantinos G Makris, Mercedeh Khajavikhan, Ziad H Musslimani, Stefan Rotter, and Demetrios N Christodoulides. Non-Hermitian physics and PT symmetry. Nature Physics, 14:11 EP -- , 2018. URL https://doi.org/10.1038/nphys4323. Tamar Goldzak, Alexei A Mailybaev, and Nimrod Moiseyev. Light Stops at Exceptional Points. Phys. Rev. Lett., 120(1):13901, 2018. doi: 10.1103/PhysRevLett.120.013901. URL https: //link.aps.org/doi/10.1103/PhysRevLett.120.013901. 29 K F Graff. Wave Motion in Elastic Solids. Dover Books on Physics Series. Dover Publica- tions, 1975. ISBN 9780486667454. URL https://books.google.co.il/books?id= 5cZFRwLuhdQC. Hossein Hodaei, Absar U Hassan, Steffen Wittek, Hipolito Garcia-Gracia, Ramy El-Ganainy, Demetrios N Christodoulides, and Mercedeh Khajavikhan. Enhanced sensitivity at higher-order exceptional points. Nature, 548(7666):187, 2017. Zhilin Hou and Badreddine Assouar. Tunable elastic parity-time symmetric structure based on the shunted piezoelectric materials. Journal of Applied Physics, 123(8):85101, 2018. doi: 10.1063/ 1.5009129. URL https://doi.org/10.1063/1.5009129. Zhilin Hou, Huiqin Ni, and Badreddine Assouar. Pt-symmetry for elastic negative refraction. Phys. Rev. Applied, 10(4):44071, 2018. doi: 10.1103/PhysRevApplied.10.044071. URL https: //link.aps.org/doi/10.1103/PhysRevApplied.10.044071. Young-Chung Hsue, Arthur J. Freeman, and Ben-Yuan Gu. Extended plane-wave expansion method in three-dimensional anisotropic photonic crystals. Phys. Rev. B, 72:195118, Nov 2005. doi: 10.1103/PhysRevB.72.195118. URL https://link.aps.org/doi/10. 1103/PhysRevB.72.195118. M I Hussein, M J Leamy, and M Ruzzene. Dynamics of Phononic Materials and Structures: Histor- ical Origins, Recent Progress, and Future Outlook. Appl. Mech. Rev., 66(4):40802, 2014. URL http://dx.doi.org/10.1115/1.4026911. L M Joseph and R V Craster. Reflection from a semi-infinite stack of layers using homogenization. Wave Motion, 54:145 -- 156, 2015. ISSN 0165-2125. doi: https://doi.org/10.1016/j.wavemoti. 2014.12.003. URL http://www.sciencedirect.com/science/article/pii/ S0165212514001747. Muamer Kadic, Graeme W. Milton, Martin van Hecke, and Martin Wegener. 3d metamaterials. Nature Reviews Physics, 1(3):198 -- 210, 2019. doi: 10.1038/s42254-018-0018-y. URL https: //doi.org/10.1038/s42254-018-0018-y. M S Kushwaha, P Halevi, L Dobrzynski, and B Djafari-Rouhani. Acoustic band structure of periodic elastic composites. Phys. Rev. Lett., 71(13):2022 -- 2025, 1993. Vincent Laude, Younes Achaoui, Sarah Benchabane, and Abdelkrim Khelif. Evanescent bloch waves and the complex band structure of phononic crystals. Phys. Rev. B, 80:092301, Sep 2009. doi: 10.1103/PhysRevB.80.092301. URL https://link.aps.org/doi/10. 1103/PhysRevB.80.092301. 30 Xiaoguai Li and Celia Reina. Simultaneous spatial and temporal coarse-graining: From atomistic models to continuum elastodynamics. Journal of the Mechanics and Physics of Solids, 130:118 -- 140, 2019. ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2019.05.004. URL http: //www.sciencedirect.com/science/article/pii/S0022509618310792. M J S Lowe. Matrix techniques for modeling ultrasonic waves in multilayered media. IEEE Trans. Ultrason. Ferroelectr. Freq. Control, 42(4):525 -- 542, 1995. ISSN 0885-3010. doi: 10.1109/58. 393096. Yan Lu and Ankit Srivastava. Level repulsion and band sorting in phononic crystals. Journal of the Mechanics and Physics of Solids, 111:100 -- 112, 2018. Jihong Ma, Di Zhou, Kai Sun, Xiaoming Mao, and Stefano Gonella. Edge modes and asymmetric wave transport in topological lattices: Experimental characterization at finite frequencies. Phys. Rev. Lett., 121:094301, Aug 2018. doi: 10.1103/PhysRevLett.121.094301. URL https:// link.aps.org/doi/10.1103/PhysRevLett.121.094301. P Markos and C M Soukoulis. Wave Propagation. From Electrons to Photonic Crystals and Left- Handed Materials. New York, Wiley, 2008. Aurélien Merkel, Vicent Romero-García, Jean-Philippe Groby, Jensen Li, and Johan Christensen. Unidirectional zero sonic reflection in passive PT -symmetric willis media. Phys. Rev. B, 98: 201102, Nov 2018. doi: 10.1103/PhysRevB.98.201102. URL https://link.aps.org/ doi/10.1103/PhysRevB.98.201102. G W Milton, M Briane, and J R Willis. On cloaking for elasticity and physical equations with a transformation invariant form. New J. Phys., 8(10):248, 2006. URL http://stacks.iop. org/1367-2630/8/i=10/a=248. Graeme W Milton and Ornella Mattei. Field patterns: a new mathematical object. Proc. R. Soc. London A Math. Phys. Eng. Sci., 473(2198), 2017. ISSN 1364-5021. doi: 10.1098/rspa.2016. 0819. URL http://rspa.royalsocietypublishing.org/content/473/2198/ 20160819. Nimrod Moiseyev. Non-Hermitian Quantum Mechanics. Cambridge University Press, 2011. doi: 10.1017/CBO9780511976186. Nimrod Moiseyev and Shmuel Friedland. Association of resonance states with the incom- plete spectrum of finite complex-scaled Hamiltonian matrices. Phys. Rev. A, 22(2):618 -- 624, 1980. doi: 10.1103/PhysRevA.22.618. URL https://link.aps.org/doi/10.1103/ PhysRevA.22.618. 31 Amir Ashkan Mokhtari, Yan Lu, and Ankit Srivastava. On the properties of phononic eigenvalue problems. Journal of the Mechanics and Physics of Solids, 2019. ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2019.07.005. URL http://www.sciencedirect.com/ science/article/pii/S0022509619304247. Lorenzo Morini, Yoann Eyzat, and Massimiliano Gei. Negative refraction in quasicrystalline Journal of the Mechanics and Physics of Solids, 124:282 -- ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2018.10.016. URL http: multilayered metamaterials. 298, 2019. //www.sciencedirect.com/science/article/pii/S0022509618306410. H Nassar, X C Xu, A N Norris, and G L Huang. Modulated phononic crystals: Non-reciprocal wave propagation and Willis materials. Journal of the Mechanics and Physics of Solids, 101: 10 -- 29, 2017. ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2017.01.010. URL http: //www.sciencedirect.com/science/article/pii/S0022509616308997. Sia Nemat-Nasser. Anti-plane shear waves in periodic elastic composites: band structure Proc. R. Soc. London A Math. Phys. Eng. Sci., 471 doi: 10.1098/rspa.2015.0152. URL http://rspa. and anomalous wave refraction. (2180), 2015. ISSN 1364-5021. royalsocietypublishing.org/content/471/2180/20150152. Sia Nemat-Nasser. Inherent negative refraction on acoustic branch of two dimensional phononic crystals. Mechanics of Materials, 132:1 -- 8, 2019. ISSN 0167-6636. doi: https://doi.org/ 10.1016/j.mechmat.2018.12.011. URL http://www.sciencedirect.com/science/ article/pii/S016766361830752X. William J. Parnell and Tom Shearer. Antiplane elastic wave cloaking using metamaterials, homog- enization and hyperelasticity. Wave Motion, 50(7):1140 -- 1152, 2013. ISSN 0165-2125. doi: https://doi.org/10.1016/j.wavemoti.2013.06.006. URL http://www.sciencedirect. com/science/article/pii/S0165212513001157. Advanced Modelling of Wave Propagation in Solids. J. B. Pendry. A chiral route to negative refraction. Science, 306(5700):1353 -- 1355, 2004. ISSN 0036-8075. doi: 10.1126/science.1104467. URL https://science.sciencemag.org/ content/306/5700/1353. R Pérez-Álvarez, René Pernas-Salomón, and VR Velasco. Relations between transfer matrices and numerical stability analysis to avoid the ω problem. SIAM Journal on Applied Mathematics, 75 (4):1403 -- 1423, 2015. 32 René Pernas-Salomón and Gal Shmuel. Dynamic homogenization of composite and locally res- onant flexural systems. ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2018.06.011. URL http://www.sciencedirect.com/ science/article/pii/S0022509618302503. J. Mech. Phys. Solids, 119:43 -- 59, 2018. A. S. Phani and M. I. Hussein, editors. Dynamics of lattice materials. Wiley, New York, 2017. A Srikantha Phani. On elastic waves and related phenomena in lattice materials and structures. AIP Advances, 1(4):41602, 2011. doi: 10.1063/1.3676167. URL https://doi.org/10. 1063/1.3676167. J Raney, N Nadkarni, C Daraio, D M Kochmann, J A Lewis, and K Bertoldi. Stable propagation of mechanical signals in soft media using stored elastic energy. Proc. Natl. Acad. Sci. U. S. A., 2016. Christian E Rüter, Konstantinos G Makris, Ramy El-Ganainy, Demetrios N Christodoulides, Mordechai Segev, and Detlef Kip. Observation of parity -- time symmetry in optics. Nature physics, 6(3):192, 2010. A.V. Shanin, K.S. Knyazeva, and A.I. Korolkov. Riemann surface of dispersion diagram of a mul- tilayer acoustical waveguide. Wave Motion, 83:148 -- 172, 2018. ISSN 0165-2125. doi: https: //doi.org/10.1016/j.wavemoti.2018.09.010. URL http://www.sciencedirect.com/ science/article/pii/S0165212518303871. Richard A Shelby, David R Smith, and Seldon Schultz. Experimental verification of a negative index of refraction. science, 292(5514):77 -- 79, 2001. Chengzhi Shi, Marc Dubois, Yun Chen, Lei Cheng, Hamidreza Ramezani, Yuan Wang, and Xiang Zhang. Accessing the exceptional points of parity-time symmetric acoustics. Nature communi- cations, 7:11110, 2016. G Shmuel and R Band. Universality of the frequency spectrum of laminates. J. Mech. Phys. Solids, 92:127 -- 136, 2016. ISSN 0022-5096. doi: http://dx.doi.org/10.1016/j.jmps.2016.04.001. G Shmuel and R Pernas-Salomón. Manipulating motions of elastomer films by electrostatically- controlled aperiodicity. Smart Mater. Struct., 25(12):125012, 2016. ISSN 1361665X. doi: 10.1088/0964-1726/25/12/125012. URL http://stacks.iop.org/0964-1726/25/ i=12/a=125012. M M Sigalas and E N Economou. Elastic and acoustic wave band structure. J. Sound Vib., 158(2): 377 -- 382, 1992. 33 D. R. Smith, Willie J. Padilla, D. C. Vier, S. C. Nemat-Nasser, and S. Schultz. Composite medium with simultaneously negative permeability and permittivity. Phys. Rev. Lett., 84:4184 -- 4187, May 2000. doi: 10.1103/PhysRevLett.84.4184. URL https://link.aps.org/doi/10. 1103/PhysRevLett.84.4184. A Srivastava. Metamaterial properties of periodic laminates. J. Mech. Phys. Solids, 96:252 -- 263, ISSN 0022-5096. doi: http://dx.doi.org/10.1016/j.jmps.2016.07.018. URL http:// 2016. www.sciencedirect.com/science/article/pii/S0022509616303933. A Srivastava and J R Willis. Evanescent wave boundary layers in metamaterials and sidestep- ping them through a variational approach. Proc. R. Soc. London A Math. Phys. Eng. Sci., 473(2200), 2017. ISSN 1364-5021. doi: 10.1098/rspa.2016.0765. URL http://rspa. royalsocietypublishing.org/content/473/2200/20160765. E L Tan. Generalized eigenproblem of hybrid matrix for Floquet wave propagation in one- dimensional phononic crystals with solids and fluids. Ultrasonics, 50(1):91 -- 98, 2010. ISSN 0041-624X. doi: http://dx.doi.org/10.1016/j.ultras.2009.09.007. URL http://www. sciencedirect.com/science/article/pii/S0041624X09001085. Daniel Torrent and José Sánchez-Dehesa. Multiple scattering formulation of two-dimensional acoustic and electromagnetic metamaterials. New Journal of Physics, 13(9):093018, sep 2011. URL https://doi.org/10.1088% 2F1367-2630%2F13%2F9%2F093018. 10.1088/1367-2630/13/9/093018. doi: Daniel Torrent, William J. Parnell, and Andrew N. Norris. Loss compensation in time-dependent elastic metamaterials. Phys. Rev. B, 97:014105, Jan 2018. doi: 10.1103/PhysRevB.97.014105. URL https://link.aps.org/doi/10.1103/PhysRevB.97.014105. G Trainiti and M Ruzzene. Non-reciprocal elastic wave propagation in spatiotemporal periodic structures. New Journal of Physics, 18(8):083047, aug 2016. doi: 10.1088/1367-2630/18/8/ 083047. URL https://doi.org/10.1088%2F1367-2630%2F18%2F8%2F083047. Martin Wegener. Metamaterials beyond optics. ISSN 0036-8075. doi: 10.1126/science.1246545. URL https://science.sciencemag.org/ content/342/6161/939. Science, 342(6161):939 -- 940, 2013. J R Willis. Negative refraction in a laminate. J. Mech. Phys. Solids, 97:10 -- 18, 2015. ISSN 0022-5096. doi: http://dx.doi.org/10.1016/j.jmps.2015.11.004. URL http://www. sciencedirect.com/science/article/pii/S0022509615302623. 34 John Willis. A study of obliquely propagating longitudinal shear waves in a periodic laminate. arXiv e-prints, art. arXiv:1310.6561, Oct 2013. Xun-Wei Xu, Yu-xi Liu, Chang-Pu Sun, and Yong Li. Mechanical PT symmetry in coupled op- tomechanical systems. Phys. Rev. A, 92:013852, Jul 2015. doi: 10.1103/PhysRevA.92.013852. URL https://link.aps.org/doi/10.1103/PhysRevA.92.013852. A J Zelhofer and D M Kochmann. On acoustic wave beaming in two-dimensional structural lattices. Int. J. Solids Struct., 115-116:248 -- 269, 2017. ISSN 0020-7683. doi: http://doi.org/10.1016/j. ijsolstr.2017.03.024. URL http://www.sciencedirect.com/science/article/ pii/S0020768317301336. Pu Zhang. Symmetry and degeneracy of phonon modes for periodic structures with glide symme- try. Journal of the Mechanics and Physics of Solids, 122:244 -- 261, 2019. ISSN 0022-5096. doi: https://doi.org/10.1016/j.jmps.2018.09.016. URL http://www.sciencedirect.com/ science/article/pii/S0022509618304605. 35
1910.13265	1	1910	2019-10-29T13:41:19	Improvement in corrosion resistance and biocompatibility of AZ31 magnesium alloy by NH+2 ions	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Magnesium alloys have been considered to be favorable biodegradable metallic materials used in orthopedic and cardiovascular applications. We introduce NH+2 to the AZ31 Mg alloy surface by ion implantation at the energy of 50 KeV with doses ranging from 1e16 ions/cm2 to 1e17 ions/cm2 to improve its corrosion resistance and biocompatibility. Surface morphology, mechanical properties, corrosion behavior and biocompatibility are studied in the experiments. The analysis confirms that the modified surface with smoothness and hydrophobicity significantly improves the corrosion resistance and biocompatibility while maintaining the mechanical property of the alloy.	physics.app-ph	physics	Improvement in corrosion resistance and biocompatibility of AZ31 magnesium alloy by NH+ 2 ions Xian Wei1,2, Zhicheng Li3, Pinduo Liu4, Shijian Li1, Xubiao Peng1, Rongping Deng5, and Qing Zhao1∗ 1School of Physics, Beijing Institute of Technology, Beijing 100081, China 2Department of science, Taiyuan Institute of Technology, Taiyuan 030008, China 3Department of Physics, Taiyuan Normal University, Taiyuan 030031, China 4School of Life Science, Beijing Institute of Technology, Beijing 100081, China and 5Science Division, Beloit College, 700 College Street, Beloit, Wisconsin 53511, USA Magnesium alloys have been considered to be favorable biodegradable metallic materials used in orthopedic and cardiovascular applications. We introduce NH+ 2 to the AZ31 Mg alloy surface by ion implantation at the en- ergy of 50 KeV with doses ranging from 1×1016 ions/cm2 to 1×1017 ions/cm2 to improve its corrosion resistance and biocompatibility. Surface morphology, mechanical properties, corrosion behavior and biocompatibility are studied in the experiments. The analysis confirms that the modified surface with smoothness and hydrophobicity significantly improves the corrosion resistance and biocompatibility while maintaining the mechanical property of the alloy. PACS numbers: Keywords: Magnesium alloy; Corrosion resistance; Biocompatibility; Ion implantation I. INTRODUCTION Magnesium and its alloys as biodegradable metals have received much attention due to their promising applications in orthopedic, cardiovascular and other medical fields [1 -- 5]. For example, biodegradable implants can provide mechani- cal support during the mending of injured tissue and can also be decomposed in the body fluid after cure rather than been removed by a second surgery [6, 7]. The magnesium ions released from the implanted material also offer beneficial ef- fects on bone growth [8 -- 10]. Moreover, magnesium and its al- loys have remarkable advantages over the Ti-based alloys and stainless steels in density and elastic modulus. The density of magnesium alloys range from 1.74 to 2.0 g/cm3 and the elas- tic modulus is from 41 to 45 GPa which are closer to those of human bones compared to the Ti-based alloys (4.4-4.5 g/cm3 and 110-117 GPa) and stainless steels (7.9-8.1 g/cm3 and 189- 205 GPa) [11, 12]. There are still a number of challenges in the development of magnesium alloys for medical applications. Since the intrin- sic standard corrosion potential of Mg (-2.37 VS CE ) is smaller than that of other biodegradable metals (Zn: -0.76 VS CE , Fe: -0.44 VS CE ) [13, 14], pure magnesium and most magnesium alloys have poor performance in corrosion resistance and pro- duce hydrogen gas during the degradation process [15]. Ad- ditionally, unlike applications in pyrotechnics, metallurgical industry, and automotive products, biomedical field requires magnesium and its alloys to possess not only controlled degra- dation rate but also excellent biocompatibility [16, 17]. There- fore, a lot of methods such as alloying, coating and surface treatment have been used to enhance both the corrosion resis- tance and the biocompatibility of magnesium and its alloys. Ion implantation is one of the effective surface treatment methods to improve the anticorrosion and biocompatibility of magnesium alloys through the surface modification. Ion im- plantation can form a functional surface by inserting a beam ∗The corresponding author Email: [email protected] of varied ionized particles; it also has the following outstand- ing advantages: 1) It does not change the size and shape of the samples, so it can be used in the last step in the product man- ufacturing process, especially for those small, complex and irregular parts (such as screws, nuts, etc.); 2) The implanted ions have high purity and controllable concentration; 3) The treatment process is clean and environmental friendly. Many types of ions, such as gaseous ions O2 [18, 19], N2 [20 -- 22], metallic ions Gd [23], Ce [24], and dual ions Zr-N[25], Cr- O [26], have been utilized in the ion implantation technology to effectively improve the corrosion resistance of magnesium and its alloys. Additionally, a class of organic functional ion implantation has been applied in other fields such as chemical modified electrode, graphene, and bioceramic [27 -- 30]. Zhao et al. [31] reported that Al2O3 ceramics bombarded with NH+ 2 ions demonstrated excellent performance in biocompatibility with living bones in the animal tests. As Li et al. [32] re- ported that the 3T3 mouse fibroblasts and human endothelial cells achieved better attachment and proliferation cultured on the surface of the polypropylene implanted by COOH+ ions. However, to the best of our knowledge, there is limited infor- mation about the organic functional ions in Mg alloy surface modification. In the consideration of the potential of ion im- plantation for anticorrosion and the less toxicity of organic molecules comparing to the metal ions, this paper reports a study of improving the corrosion resistance and the biocom- patibility of AZ31 Mg alloy by NH+ 2 implantation. The me- chanical performance of the alloy is measured by Nano inden- ter. The corrosion property is analyzed by both electrochemi- cal and immersion tests, and the biocompatibility of the mod- ified alloy is investigated by the cell viability assay in vitro. II. EXPERIMENTAL DETAILS A. Ion implantation and surface characterization The AZ31 Mg alloy (Mg with Al 3.12 wt.%, Zn 0.93 wt.%, and Mn 0.30 wt.%) was cut into blocks with dimensions of 10mm×10mm×2mm. The samples were ground with different grit sizes of SiC sandpapers (up to 2000 grit), ultrasonically cleaned in acetone, absolute ethanol for 10 min respectively, and dried in the air. NH+ 2 ion implantation was performed on an ion implanter equipped with gaseous NH3 as the ion source. The pressure in the target chamber was maintained around 10−3 Pa, and the current density of ion beam was less than 0.01 A/cm2. The NH+ 2 ions identified by mass spectrom- etry were accelerated and implanted into the target samples with doses of 1 × 1016, 5 × 1016, and 1 × 1017 ions/cm2 at an energy of 50 KeV. The elemental depth profiles and chemical compositions of the treated samples were determined by X-ray photoelectron spectroscopy with Al Kα X-rays as the radiation source (XPS, PHI Quantera II, Ulvac-Phi Inc.). The approximated sputter- ing rate for depth profiling analysis was 8.8 nm/min based on the reference standard. Atomic force microscopy (AFM, SPM-960, Shimadzu, Japan) was used to examine the surface morphology before and after the ion implantation. The wa- ter contact angle was measured by a contact angle goniometer (FTA 200, Dataphysics Inc., USA). The mechanical properties of the sample surface were determined using Nano Indenter (XP, MTS Systems Corporation, USA). B. Electrochemical tests The electrochemical tests were operated on the electro- chemical workstation of a three-electrode cell system with a platinum foil as the counter electrode, a saturated calomel electrode (SCE) as the reference electrode, and the sample as the working electrode. The exposed surface area of the sample to the Hank′s solution (NaCl 8.00 g/L, KCl 0.40 g/L, CaCl2 0.14 g/L, MgCl2·6H2O 0.10 g/L, MgSO4·7H2O 0.10 g/L, KH2PO4 0.06 g/L, Na2HPO4·2H2O 0.06 g/L, NaHCO3 0.35 g/L, glucose 1.00 g/L) was 1 cm2 at the temperature of 37◦C. The potentiodynamic polarization curves were obtained at a scanning rate of 1 mV/s. The corrosion potential (Ecorr) and the corrosion current density (Icorr) were measured from the polarization curve according to Tafel analysis. The elec- trochemical impedance spectroscopy (EIS) was performed in a range from 100 kHz to 100 mHz with a 5 mV amplitude perturbation. For each type of samples, at least three samples were tested. was calculated as follows: v = ∆m S t 2 (1) where v is the corrosion rate expressed in units of grams per square meter per hour (g/(m2 · h)). ∆m represents the mass loss. S is the surface area of the sample before immersion. t is the immersion time. D. Cytotoxicity tests Mouse MC3T3-E1 preosteoblasts (acquired from Ameri- can Type Culture Collection (ATCC)) were utilized to evalu- ate the cytotoxicity of untreated and NH+ 2 implanted AZ31 Mg alloy by indirect cell assay. Cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM, Gibco) with 10% fetal bovine serum (FBS) at 37◦C in a humidified atmosphere of 5% CO2. Prior to the cell viability experiment, samples were sterilized under ultraviolet radiation for 2 hrs. Extracts were produced by immersing the samples into serum-free DMEM in a humidified atmosphere of 5% CO2 for 72 hrs. The sur- face area of extraction medium ratios were set at 0.5 and 1.25 cm2/ml, respectively. Then the supernatant fluid was with- drawn, filtered, and refrigerated at 4 ◦C for the following as- say. MC3T3-E1 cells were seeded on 96-well culture plates with a density of 1 × 103 cells per well and incubated for 24 hrs to allow attachment. The medium was then replaced by the corresponding alloy extracts with 10% FBS. The DMEM medium with 10% FBS was set as control groups. After 1 and 3 days of incubation, 10 µL MTT was added into each well and then cells were incubated for 4 hrs. 100 µL dimethyl sul- foxide was added to dissolve the formed formazan crystals. The absorbance of each well was measured with a microplate reader at the wavelength 490 nm (Cytation3, Bio-Teh, USA). In the end, the cell viability was computed as follows: Viability = ODsample ODcontrol × 100% (2) III. RESULTS A. Surface characterization C. Immersion tests Immersion tests on the NH2-implanted and the untreated AZ31 Mg alloys were carried out in Hank′s solution accord- ing to ASTM-G31-72 [33]. The samples were submerged in the solution with the volume to area ratio of 20 mL/cm2 at a temperature of 37◦C using water bath. After immersion for 3, 7, 15, 31 days, the samples were taken out from the solution. The corrosion products on the samples were removed with chromic acid. The samples were gently rinsed with distilled water, and then dried in open air. The changes on the sur- face morphology were detected using scanning electron mi- croscopy (SEM, S-4800, Hitachi, Japan). For each immersion time node, the average corrosion rate for 3 parallel specimens The XPS depth profile of the sample implanted with the dose of 1 × 1017 NH+ 2 /cm2 is depicted in Fig.1 and the high- resolution XPS Mg 1s, O 1s, Al 2p, and N 1s spectra are shown in Fig. 2. As shown in Fig.1, the concentration of N atom increases first with the sputtering time and then slowly decreases to 0%, indicating the formation of a thin NH2- containing layer with a maximum concentration of approxi- mately 16% at a depth of nearly 141 nm. The existence of oxygen in the top layer results from the formation of surface oxides due to the non-high vacuum condition. The oxygen and nitrogen concentrations fall down to very low levels at the depth exceeding 202 nm. The high-resolution Mg 1s spec- trum in the Fig 2(a) shows that Mg is observed in the form of oxidized state (MgO/Mg(OH)2) in the top surface layer. 100 ion implantation [42, 43]. 3 ) % ( n o i t a r t n e c n o C m o t A 80 60 40 20 0 Mg O Al N 0 50 100 150 200 Depth (nm) 250 300 350 FIG. 1: XPS depth profiles of the implanted sample with dose of 1 × 1017 NH+ 2 /cm2 by XPS. With the longer sputtering time, the Mg 1s peak shifts toward smaller binding energy, implying that Mg gradually turns to the metallic state (Mg0). The binding energy of Al 2p at 74.4 eV suggests that Al2O3 is observed on the surface. The inten- sity of the Al peak shifts from the oxidized state to the metallic state with increasing in depth. The binding energy of N 1s in the spectrum is around at 398.4 eV indicating the presence of amino. This result confirms the successful grafting of or- ganic functional groups on the surface of AZ31 magnesium alloys. As shown in Fig.2(b), the oxygen peak represents the formation of MgO, Mg(OH)2, and Al2O3 in the top layer. The oxygen peak intensity decreases with sputtering time, which is in accordance with the gradually reduced oxygen concen- tration as shown in Fig. 1. The formation of the amino layer and the oxide layer (MgO, Mg(OH)2, and Al2O3) provides a barrier to resist the corrosion of magnesium alloy [34 -- 37]. The atomic force microscopy (AFM) images of the un- treated and treated samples are shown in Fig. 3, which re- veals the changes in the surface topography after implanted with different doses of NH+ 2 ions. The surface of the untreated sample is relatively rough with many particles. After the im- plantation with various doses of NH+ 2 ions, more compact and finer particles are observed on the surface. The number of pits is significantly reduced with the higher implantation dose. The root-mean-square (RMS) roughness values reduce with increasing doses as shown in Table I. This indicates that the surface becomes smoother after NH+ 2 ion implantation which may be attributed to the surface restructure at raised tempera- ture during surface modification process [38]. Moreover, as reported in literatures [39 -- 41], the smoother surface corre- sponds to the lower corrosion rate because the electrode po- tential has small difference between the concave and convex. Since a relatively smooth surface of sample is obtained by sur- face modification through NH+ 2 ions, the sample is expected to be better resistant to corrosion. Fig. 4 shows the variation in water contact angles of AZ31 Mg alloys before and after being treated by NH+ 2 ions. AZ31 Mg alloy is a type of hydrophilic material with a water contact angle of 47.95◦. With the higher implantation doses, the wa- ter contact angle of the treated sample increases. The higher contact angle corresponds to the lower wettability, which indi- cates that more hydrophobic surface is obtained after the NH+ 2 B. Mechanical property The results of the hardness and the elastic tests are shown in Fig. 5. The indentation depth is much larger than the im- plantation depth as revealed in the XPS depth profile results, indicating that the data of hardness and the data of elastic modulus are not only based on the implanted layer but also the substrate layer. Both the hardness and the elastic modu- lus of the untreated sample maintain a plateau with increas- ing depth. While the hardness and elastic modulus of the treated samples exhibit diverse trend with the displacement into the surface. As to the implanted sample with dose of 1 × 1017 NH+ 2 /cm2, the maximum hardness reaches 5.4 GPa at 35.7 nm under the surface and then decreases to a plateau value of 1.2 GPa at approximately 1017.7 nm. The maximum value of elastic modulus is 67.3 GPa at the topmost layer and then decreases to a constant value of 44.5 GPa. Similar trend can be observed for the samples with doses of 1 × 1016 and /cm2, respectively. Both the hardness and the 5 × 1016 NH+ 2 elastic modulus of the treated samples decrease gradually to constant values which are the values of the substrate with in- creased indentation depth. Thus, the mechanical performance of the implanted layer is better than the substrate, or at least comparable to the latter [44, 45]. C. Corrosion performance The potentiodynamic polarization curves of different sam- ples soaked in Hank′s solution at 37◦C are exhibited in Fig. 6. The electrochemical parameters are analyzed to charac- terize the corrosion nature as listed in Table I. In contrast to untreated AZ31 magnesium alloy, the samples implanted with NH+ 2 all achieve lower corrosion current density (Icorr). The Icorr values for the treated samples are shifted towards neg- ative direction as increasing implantation dosage. This indi- cates that the surface layer containing more NH+ 2 ions may be more beneficial to restrain the transfer of Cl− during the po- larization process [46 -- 48]. Therefore the degradation of mag- nesium alloy could be suppressed by NH+ 2 ions implantation. The EIS experiment is conducted to analyze the corrosion property of electrode system. In general, the impedance spec- tra are described either by a Nyquist (complex plane) plot in which the opposite of imaginary part of impedance (Zim) versus the real part or by a Bode plot where the modulus of the impedance and the phase angle versus frequency, respec- tively [49 -- 52]. Fig. 7 shows the EIS data of unimplanted and NH2-implanted samples in Hank′s solution at 37◦C. As shown in Fig. 7(a), two capacitive loops are emerged at high fre- quency due to the formation of magnesium hydroxide/oxide and the dissolution of Mg through cracks [53]. The capacitive loops are distinguishable for the treated samples with dose of 5 ×1016 and 1 ×1017 ions/cm2, whereas almost overlapping for the treated sample of 1 × 1016 ions/cm2 and untreated sample [54]. The capacitive loops of the implanted samples are sig- nificantly larger than that of the untreated sample. And the di- (a) Mg 1s Mg0 (b) O 1s Mg2+ ) . u . a ( y t i s n e t n I 23.0min 19.5min 19.5min 16.0min 12.5min 9.0min 5.5min 2.0min 1304 1306 1308 1310 1302 Binding Energy (eV) Al0 4 23.0min 19.5min 16.0min O2+ 12.5min 9.0min 5.5min 2.0min 526 528 530 532 534 536 538 540 Binding Energy (eV) ) . u . a ( y t i s n e t n I 1300 (c) Al 2p ) . u . a ( y t i s n e t n I 70 (d) N 1s ) . u . a ( y t i s n e t n I 23.0min 19.5min 16.0min 12.5min 9.0min 5.5min 2.0min 80 82 396 23.0min 19.5min 16.0min 12.5min 9.0min 5.5min 2.0min 406 408 + 2 NH 400 398 404 Binding Energy(eV) 402 Al3+ 74 72 Binding Energy (eV) 78 76 FIG. 2: High-resolution XPS spectra of the implanted sample with dose of 1 × 1017 NH+ 2 /cm2 at different sputtering time. TABLE I: Comparison of AZ31 with and without implantation in terms of surface roughness. Doses (ions/cm2)RMS roughness (nm)Ecorr (V)icorr (µA/cm2) 0 1 × 1016 5 × 1016 1 × 1017 35.80 ± 1.03 29.68 ± 4.31 17.20 ± 1.46 12.80 ± 0.44 -1.03 -1.03 -1.05 -1.01 869.72 441.34 206.00 125.23 ameter of the capacitive semicircle is the largest for implanted sample with a dose of 1 × 1017 NH+ 2 /cm2. Since the larger is the capacitive loop, the better is the corrosion resistance. The Nyquist plot implies that the anti-corrosion performance of AZ31 Mg alloy improves after NH+ 2 ion implantation. The Bode impedance plots of the untreated and the im- planted sample are displayed in Fig. 7 (b). The impedances of the implanted samples are higher at all frequency range com- pared to that of the untreated samples. As to the sample with high implantation dose (1 × 1017 NH+ 2 /cm2), the impedance is 8 times larger than that of the untreated sample at the low fre- quency of 100 mHz, and the impedance is increased evidently by 2 times compared with that of the untreated sample at the high frequency of 100 kHz. As reported in literatures [38, 55 -- 59], the higher impedances at low and high frequencies reflect the slower process of charge transfer and the poor ability of the electrolyte to penetrate. So the Bode impedance result implies that surface protection is enhanced after NH+ 2 ion im- plantation. The Bode phase angle evolution of the untreated and the implanted sample is shown in Fig. 7 (d). The maxi- mum phase angles of treated samples increase with increasing implantation dose, and are all larger than that of the untreated sample. More capacitive behavior is demonstrated by larger phase angle, which suggests the occurrence of a more stable and denser layer to retard the electrolyte penetration to the substrate [60, 61]. Hence, the difficulties in charge transfer and electrolyte penetration confirm that ion implantation ef- fectively improves the corrosion resistance. To further investigate the enhanced corrosion resistance, the electrical equivalent circuit is utilized to fit the EIS spectra of the untreated sample and the implanted sample, as shown in Fig. 7 (c). The equivalent circuit displays two layers, a porous outer layer and a barrier-like inner layer. CPE1 is the constant phase element of the outer porous layer and R1 is the corre- 5 FIG. 3: Surface morphology of the samples observed by AFM: (a) untreated, (b) implanted sample with dose of 1×1016 ions/cm2, (c) implanted sample with dose of 5 × 1016 ions/cm2, and (d) implanted sample with dose of 1 × 1017 ions/cm2. ner layer of the implanted sample can further reinforce the corrosion resistance with the corrosion evolution. These EIS results suggest that the samples modified by NH+ 2 ion improve the anti-corrosion property of AZ31 Mg alloys and are in ac- cordance with the polarization curve. The immersion test is conducted to further study the corro- sion resistance behavior. The corrosion rate is calculated by the mass loss method with sample soaked in the Hank′s so- lution for 3, 7, 15, 31 days at 37◦, respectively. As shown in Fig. 8, the corrosion rate decreases with increasing immersion time. The corrosion rates of NH2-implanted samples have smaller values than that of the untreated sample. Fig. 9 ex- hibits the corrosion morphologies of the NH2-implanted and the untreated samples after immersed into the Hank′s solution for 3, 7, 15, 31 days at 37◦C, respectively. Group (a) is the surface morphology of the untreated sample while Group (b) is the surface morphology of the NH2-implanted sample with dose of 1 × 1016 ions/cm2. As shown in Group (a) of Fig. 9, cracks are obviously observed after immersion for 3 days. The general trend of crack spacing goes bigger along with longer immersion time. The untreated sample even suffers from se- vere corrosion pits after immersion for 31 days. On the con- trary, some tiny cracks are observed on the NH2-implanted sample after immersion for 3 and 7 days. Although the cor- rosion cracks grow larger, the obvious corrosion pit does not appear on the treated surface after immersion for 31 days in FIG. 4: Contact angles for water on AZ31 Mg alloy. sponding resistance. The CPE2 stands for the constant phase element in the inner layer in parallel connection with the re- sistance R2. Rs represents the solution resistance between the working electrode and the reference electrode. The fitted pa- rameters of the electrical equivalent circuit are listed in Table II. As disclosed in Table II, R1 and R2 increase while CPE1 and CPE2 decrease after samples implanted by NH+ 2 ion, indi- cating that the corrosion rates are retarded in Hank′s solution. As to the sample of high implantation dose (1×1017 ions/cm2), it achieves the highest resistance, approximately 10 times in R1 and R2 larger than that of the untreated sample. The resis- tance R2 of the inner layer is dramatically increased which is attributed to the formation of MgO, Al2O3, and amino based on the results of the XPS depth profile. Hence, the stable in- 6 4 2 0 0 ) a P G ( s s e n d r a H Untreated 1×10 16ions/cm 2 5×10 16ions/cm 2 1×10 17ions/cm 2 500 1000 Depth (nm) 1500 2000 (a) 100 ) a P G ( s u l u d o M 80 60 40 20 0 6 Untreated 1×10 16ions/cm 2 5×10 16ions/cm 2 1×10 17ions/cm 2 500 1000 Depth (nm) 1500 2000 (b) FIG. 5: The hardness (a) and the elastic modulus values (b) acquired from the untreated and NH+ 2 -implanted samples. TABLE II: Electrochemical parameters of AZ31 with and without implantation obtained by equivalent circuit simulation. Doses (ions/cm2)Rs(Ωcm2)CPE1(Ω−2cm2S −n) n1 R1(Ωcm2)CPE2(Ω−2cm−2S −n) n2 R2(Ωcm2) 0 1 × 1016 5 × 1016 1 × 1017 11.83 15.67 35.52 21.93 3.93 × 10−5 3.64 × 10−5 2.77 × 10−5 1.55 × 10−5 0.98 19.74 0.98 50.78 0.96 159.00 0.99 197.10 7.42 × 10−4 6.64 × 10−4 5.55 × 10−4 5.09 × 10−4 0.54 24.59 0.67 34.11 0.59 207.40 0.54 262.20 -0.6 E C S -0.8 . s v V / l a i t n e t o P -1.0 -1.2 -1.4 10-6 Untreated 1×1016 ions/cm2 5×1016 ions/cm2 1×1017 ions/cm2 10-5 10-4 10-3 10-2 Current Density /A.cm-2 FIG. 6: Potentiodynamic polarization curves of untreated and NH+ 2 implanted AZ31 in Hank′s solution. Group (b). Therefore, the corrosion rate of the AZ31 Mg alloy is effectively decreased after the NH+ 2 ion implantation. D. In vitro cytotoxicity studies In order to evaluate the vitro cytotoxicity of the samples, the viability of MC3T3-E1 cells with different extraction medium ratios (0.50 and 1.25 cm2/ml) after incubation for 3 day is measured, as shown in Fig. 10. The cell viability of the con- trol group is considered as 100%. As known in ISO 10993- 5, cell viability is larger than 80% corresponding to Grade 0 or Grade 1 of cytotoxicity of materials, which indicates that this material can be used as biomaterials [62, 63]. For the extraction medium ratio of 0.5 cm2/ml, cells of the untreated samples exhibit viabilities below 80%, indicating obvious cy- totoxic effects according to the ISO 10993-5. However, cell viabilities in NH+ 2 -AZ31 extracts are all above 80% and higher than that in the untreated AZ31 extract during the whole in- cubation period. As to the extraction medium ratio of 1.25 cm2/ml (Fig. 10(b)), the cells display roughly the same via- bility (about 80%) for different samples after incubation for 1 day. However, after 3 days of incubation, cells grown in the extract from NH+ 2 -implanted samples present significantly in- creased activity compared to the untreated sample, indicating that the cytotoxic effect of the implanted sample on MC3T3- E1 preosteoblasts is not developed, though the reduced cell viability is observed in the first day of incubation. This in- dicates that the implanted samples with the smoother surface and the reduced corrosion rate show the relatively stable and biofriendly effect on cell growth [64]. Overall, the cell via- bility of samples with NH+ 2 ion implantation is significantly higher than that of the untreated samples after the incubation for 3 days, indicating the better biocompatibility of the NH+ 2 - implanted AZ31 Mg alloy [65]. IV. DISCUSSION A. Corrosion behavior According to the XPS spectra (Fig. 1-2), NH+ 2 ions have been successfully implanted into the sample surface with the formation of the ammonia and the oxide layers. A relatively smooth and compact surface of treated sample is observed by AFM (Fig. 3). As shown in Fig. 4, a larger water contact angle of the treated sample suggests that a hydrophobic surface is Untreated 5×10 16ions/cm 2 1×10 16ions/cm 2 1×10 17ions/cm 2 ) 2 m c . ( m Z i Untreated 1×10 16ions/cm 2 ) 2 m c . ( m Z i Rs Zre ( .cm2) Zre( .cm2) (a) 7 Untreated 1×10 16ions/cm 2 5×10 16ions/cm 2 1×10 17ions/cm 2 Frequency(Hz) (b) ) 2 m c . ( e c n a d e p m I g e d / e l g n A e s a h P CPE1 R1 (c) CPE2 R2 Frequency (Hz) (d) FIG. 7: EIS measurements (scatter plot) and model fit (solid lines) of samples: (a) Nyquist plot, (b) Bode plot of impedance versus frequency, (c) Equivalent circuit of samples with and without implantation, and (d) Bode plot of phase angle versus frequency. 0.24 0.23 0.22 0.21 0.20 0.19 0.18 ) ) h . 2 m ( / g ( e t a r n o i s o r r o C 0.17 3 days Untreated 1×10 16ions/cm 2 5×10 16ions/cm 2 1×10 17ions/cm 2 7 days 15 days 31 days Immersion time (days) FIG. 8: The corrosion rate of samples with and without implantation during different immersion time. obtained after the NH+ 2 ions implantation. The modulated smooth surface with hydrophobicity signif- icantly improves the corrosion resistance of AZ31 Mg alloy which can be proved by the electrochemical tests (Fig. 6-7) and the immersion test (Fig. 8-9) in Hank′s solution at the temperature of 37◦C. The degradation process is generally ac- companied by electrochemical reactions. Magnesium is one of the active metals that can quickly dissolve (Eq. (3)) at the anode electrode during immersion [66, 67]. Mg2+ reacts with H2O to generate hydrogen gas at the cathode electrode (Eq. (4)). The Mg(OH)2 layer forms with consuming the H+ and producing the OH− (Eq. (5)), creating an alkaline environ- ment. Mg → Mg2+ + 2e− H2O + 2e− → H2 + 2OH − Mg2+ + 2OH − → Mg(OH)2 (3) (4) (5) The Mg(OH)2 can be corroded by Hank′s solution which con- tains abundant Cl−. According to the penetration/dissolution mechanism as reported in [68], Cl− disperses and interacts in the hydroxide, then penetrates through the hydroxides to reach the substrate, which results in the dissolution of the substrate. The thinner Mg(OH)2 layer of the untreated sample is easily attacked by Cl− leading to the poor corrosion resistance of the material. However, for samples implanted by NH+ 2 , the stable products including the ammonia layer and oxide layer (MgO/Al2O3) are formed as the strong barriers against the corrosion to the substrate. These layers can delay the elec- trolytes to Mg substrate. Therefore, the corrosion rate is de- creased after the NH+ 2 ion implantation. 8 FIG. 9: Surface morphology of samples in different immersion time periods. Group (a) is the bare sample, and group (b) is implanted sample with dose of 1 × 1016 ions/cm2. % ( y t i l i b a i V ) l l e C 160 140 120 100 80 60 40 20 0 Untreated 5×1016ions/cm 2 1×1016ions/cm 2 1×1017ions/cm 2 1 3 Culture Time (day) (a) 160 140 120 100 80 60 40 20 0 Untreated 5×1016ions/cm 2 1×1016ions/cm 2 1×1017ions/cm 2 cal as a functional group, which is generally found in organic molecules, is less toxic in comparison with the metal ions. In contrary, the untreated sample presents a lower cell viability compared to the implanted ones, mainly resulting from the higher concentration of the released metal ions. Therefore, the biocompatibility is enhanced through grafting the -NH2 amidogen radical onto the inorganic biomaterial (AZ31). 1 3 Culture Time (day) (b) V. CONCLUSIONS FIG. 10: In vitro viability of MC3T3-E1 cells cultured in unim- planted and implanted AZ31 extraction mediums for 1 and 3 days with different extraction medium ratios: (a) 0.50 cm2/ml and (b) 1.25 cm2/ml. B. Cytotoxicity evaluation Although the AZ31 magnesium alloy is a candidate of bio- material because of its mechanical property and biodegrad- ability, improvements in chemical stability in living tissues, biological properties including wettability, adhesion and bio- compatibility are required so that it could have strong bondage with organic bone tissue. The -NH2 and -NH amidogen rad- icals are helpful in simulating the activity of living tissues. In order to evaluate the biocompatibility of NH+ 2 implanted AZ31, we have conducted the toxicity assay which is a vital index for rapidly screening the biocompatibility of biometal materials. The toxicity of implants is mainly affected by the amount of released ions during degradation. In practice, cy- totoxic effect occurs if the released ion concentration is be- yond the tolerance limit. Moreover the corrosion process con- tributes to the local alkalization and hydrogen evolution which are harmful to cell viability. As shown in Fig. 10, samples implanted by NH+ 2 ions do not induce toxicity to MC3T3-E1 cells after cultured for 3 days. This indicates that the decreased corrosion rate leads to the limited alkalization and hydrogen evolution and delays the release of ions. Additionally, the -NH2 amidogen radi- In this study, AZ31 magnesium alloy is implanted by differ- ent doses of NH+ 2 ions at energy of 50 KeV. The degradation behavior and the cytotoxicity of the ion implanted samples have been investigated. The results from potentiodynamic po- larization, electrochemical impedance spectroscopy, and im- mersion tests demonstrate that the implanted samples show the improved performance in corrosion resistance. This is at- tributed to the formation of a stable and protective layer com- posed of oxide and amino with stable chemical bonds as a bar- rier to prevent the material from being corroded by biological liquid. Furthermore, the modification by NH+ 2 ions improves the bioactivity of the material surface. In the toxicity experi- ment, higher viability of implanted samples is observed from the extraction medium ratio of 1.25 cm2/ml after incubating for 3 days, which confirms the desired biocompatibility of the treated samples in vitro. Our data indicates that NH+ 2 ion im- plantation is a favorable method of improving corrosion re- sistance and cytocompatibility of the AZ31 magnesium alloy. The method of implanting simple organic functional groups on Mg alloy surface may provide a new perspective for the surface treatment for biomaterials. Acknowledgments This work is supported by National Science Foundation (NSF) of China with the Grant No.11675014. Additional sup- port is provided by Ministry of Science and Technology of China (2013Y Q030595 − 3). 9 [1] R. Erbel, C. Di Mario, J. Bartunek, J. Bonnier, B. de Bruyne, F. R. Eberli, P. Erne, M. Haude, B. Heublein, M. Horrigan, C. Il- sley, D. Bose, J. Koolen, T. F. Luscher, N. Weissman, R. Waks- man, Temporary scaffolding of coronary arteries with bioab- sorbable magnesium stents: a prospective, non-randomised multicentre trial, Lancet 369 (2007) 1869 -- 1875. [2] R. Waksman, R. Pakala, P. K. Kuchulakanti, R. Baffour, D. Hellinga, R. Seabron, F. O. Tio, E. Wittchow, S. Hartwig, C. Harder, R. Rohde, B. Heublein, A. Andreae, K. H. Wald- mann, A. Haverich, Safety and efficacy of bioabsorbable mag- nesium alloy stents in porcine coronary arteries, Catheter Car- dio Inte 68 (2006) 607 -- 617. [3] F. Witte, J. Fischer, J. Nellesen, H. A. Crostack, V. Kaese, A. Pisch, F. Beckmann, H. Windhagen, In vitro and in vivo corrosion measurements of magnesium alloys, Biomaterials 27 (2006) 1013 -- 1018. [4] F. Witte, V. Kaese, H. Haferkamp, E. Switzer, A. Meyer- Lindenberg, C. J. Wirth, H. Windhagen, In vivo corrosion of four magnesium alloys and the associated bone response, Bio- materials 26 (2005) 3557 -- 3563. [5] P. Zartner, R. Cesnjevar, H. Singer, M. Weyand, First success- ful implantation of a biodegradable metal stent into the left pul- monary artery of a preterm baby, Catheter Cardio Inte 66 (2005) 590 -- 594. [6] W. H. Jin, G. M. Wang, Z. J. Lin, H. Q. Feng, W. Li, X. Peng, A. M. Qasim, P. K. Chu, Corrosion resistance and cytocompat- ibility of tantalum-surface-functionalized biomedical zk60 mg alloy, Corros Sci 114 (2017) 45 -- 56. [7] F. Witte, N. Hort, C. Vogt, S. Cohen, K. U. Kainer, R. Willumeit, F. Feyerabend, Degradable biomaterials based on magnesium corrosion, Curr Opin Solid St M 12 (2008) 63 -- 72. [8] X. N. Gu, Y. F. Zheng, Y. Cheng, S. P. Zhong, T. F. Xi, In vitro corrosion and biocompatibility of binary magnesium al- loys, Biomaterials 30 (2009) 484 -- 498. [9] Y. Zhao, M. I. Jamesh, W. K. Li, G. S. Wu, C. X. Wang, Y. F. Zheng, K. W. K. Yeung, P. K. Chu, Enhanced antimicro- bial properties, cytocompatibility, and corrosion resistance of plasma-modified biodegradable magnesium alloys, Acta Bio- mater 10 (2014) 544 -- 556. [10] F. Feyerabend, J. Fischer, J. Holtz, F. Witte, R. Willumeit, H. Drucker, C. Vogt, N. Hort, Evaluation of short-term effects of rare earth and other elements used in magnesium alloys on primary cells and cell lines, Acta Biomater 6 (2010) 1834 -- 1842. [11] M. P. Staiger, A. M. Pietak, J. Huadmai, G. Dias, Magnesium and its alloys as orthopedic biomaterials: A review, Biomateri- als 27 (2006) 1728 -- 1734. [12] M. Ali, M. Hussein, N. Al-Aqeeli, Magnesium-based compos- ites and alloys for medical applications: A review of mechani- cal and corrosion properties, Journal of Alloys and Compounds (2019). [13] P. Li, C. Schille, E. Schweizer, E. Kimmerle-Muller, F. Rupp, A. Heiss, C. Legner, U. E. Klotz, J. Geis-Gerstorfer, L. Schei- deler, Selection of extraction medium influences cytotoxicity of zinc and its alloys, Acta Biomater (2019). [14] Y. Zheng, X. Xu, Z. Xu, J. Wang, H. Cai, Development of Zn- Based Degradable Metallic Biomaterials, Metallic Biomaterials in: Directions and Technologies, John Wiley Sons, New, 2017. [15] Y. J. Chen, Z. G. Xu, C. Smith, J. Sankar, Recent advances on the development of magnesium alloys for biodegradable im- plants, Acta Biomater 10 (2014) 4561 -- 4573. [16] Y. Li, C. Wen, D. Mushahary, R. Sravanthi, N. Harishankar, G. Pande, P. Hodgson, Mg-zr-sr alloys as biodegradable im- plant materials, Acta Biomater 8 (2012) 3177 -- 3188. [17] K. Kainer, F. V. Buch, The current state of technology and potential for further development of magnesium applications, Magnesium-Alloys and Technology, Weinheim: Wiley-VCH Verlag, 2004. [18] G. S. Wu, K. Feng, A. Shanaghi, Y. Zhao, R. Z. Xu, G. Y. Yuan, P. K. Chu, Effects of surface alloying on electrochemical corro- sion behavior of oxygen-plasma-modified biomedical magne- sium alloy, Surf Coat Tech 206 (2012) 3186 -- 3195. [19] G. J. Wan, M. F. Maltz, H. Sun, P. P. Li, N. Huang, Corro- sion properties of oxygen plasma immersion ion implantation treated magnesium, Surf Coat Tech 201 (2007) 8267 -- 8272. [20] G. Wu, X. Zeng, S. Yao, H. Han, Ion implanted az31 magne- sium alloy, Materials Science Forum 546-549 (2007) 551 -- 554. [21] S. Fukumoto, A. Yamamoto, M. Terasawa, T. Mitamura, H. Tsubakino, Microstructures and corrosion resistance of mag- nesium implanted with nitrogen ions, Materials Transactions 42 (2001) 1232 -- 1236. [22] G. S. Wu, K. J. Ding, X. Q. Zeng, X. M. Wang, S. S. Yao, Improving corrosion resistance of titanium-coated magnesium alloy by modifying surface characteristics of magnesium alloy prior to titanium coating deposition, Scripta Mater 61 (2009) 269 -- 272. [23] X. W. Tao, Z. Z. Wang, X. B. Zhang, Z. X. Ba, Y. M. Wang, Nanomechanical and corrosion properties of zk60 magnesium alloy improved by gd ion implantation, Surf Rev Lett 21 (2014). [24] X. M. Wang, X. Q. Zeng, S. S. Yao, G. S. Wu, Y. J. Lai, The corrosion behavior of ce-implanted magnesium alloys, Mater Charact 59 (2008) 618 -- 623. [25] M. Q. Cheng, Y. Q. Qiao, Q. Wang, H. Qin, X. L. Zhang, X. Y. Liu, Dual ions implantation of zirconium and nitrogen into magnesium alloys for enhanced corrosion resistance, antimi- crobial activity and biocompatibility, Colloid Surface B 148 (2016) 200 -- 210. [26] R. Z. Xu, G. S. Wu, X. B. Yang, T. Hu, Q. Y. Lu, P. K. Chu, Controllable degradation of biomedical magnesium by chromium and oxygen dual ion implantation, Mater Lett 65 (2011) 2171 -- 2173. [27] M. X. Zhang, C. M. Guo, J. B. Hu, Indium tin oxide electrode 2 ion implantation technique for determination modified by a nh+ of daunorubicin, J Electrochem Soc 160 (2013) H1 -- H5. [28] D. M. Gao, Y. Sun, Q. Y. Zhao, J. B. Hu, Q. L. Li, Determina- tion of hemoglobin at a novel nh2/ito ion implantation modified electrode, Microchim Acta 160 (2008) 241 -- 246. [29] D. J. Li, L. F. Niu, Effects of cooh+ ion implantation on hemo- compatibility of polypropylene, Sci China Ser E 45 (2002) 666 -- 670. [30] M. X. Guo, M. S. Li, X. Q. Liu, M. L. Zhao, D. J. Li, D. S. Geng, X. L. Sun, H. Q. Gu, N-containing functional groups in- duced superior cytocompatible and hemocompatible graphene by nh2 ion implantation, J Mater Sci-Mater M 24 (2013) 2741 -- 2748. [31] Q. Zhao, G. J. Zhai, D. H. L. Ng, X. Z. Zhang, Z. Q. Chen, Surface modification of al2o3 bioceramic by nh2+ ion implan- tation, Biomaterials 20 (1999) 595 -- 599. [32] D. J. Li, L. F. Niu, Cell attachment of polypropylene surface- modified by cooh+ ion implantation, Nucl Instrum Meth B 192 (2002) 393 -- 401. [33] ASTM-G31-72. Standard practice for laboratory immersion corrosion testing of metals, Annual book of ASTM standards., 10 American Society for Testing and Materials, Philadephia, Penn- sylvania, U S A, 2004. [34] W. H. Jin, G. S. Wu, H. Q. Feng, W. H. Wang, X. M. Zhang, P. K. Chu, Improvement of corrosion resistance and biocom- patibility of rare-earth we43 magnesium alloy by neodymium self-ion implantation, Corros Sci 94 (2015) 142 -- 155. [35] M. I. Jamesh, G. S. Wu, Y. Zhao, W. H. Jin, D. R. McKenzie, M. M. M. Bilek, P. K. Chu, Effects of zirconium and nitrogen plasma immersion ion implantation on the electrochemical cor- rosion behavior of mg-y-re alloy in simulated body fluid and cell culture medium, Corros Sci 86 (2014) 239 -- 251. [36] C. M. Wang, J. Shen, X. K. Zhang, B. Duan, J. X. Sang, In vitro degradation and cytocompatibility of a silane/mg(oh)2 compos- ite coating on az31 alloy by spin coating, J Alloy Compd 714 (2017) 186 -- 193. [37] H. M. Wong, Y. Zhao, V. Tam, S. L. Wu, P. K. Chu, Y. F. Zheng, M. K. T. To, F. K. L. Leung, K. D. K. Luk, K. M. C. Che- ung, K. W. K. Yeung, In vivo stimulation of bone formation by aluminum and oxygen plasma surface-modified magnesium implants, Biomaterials 34 (2013) 9863 -- 9876. [38] R. Z. Xu, X. B. Yang, P. H. Li, K. W. Suen, S. Wu, P. K. Chu, Eelectrochemical properties and corrosion resistance of carbon- ion-implanted magnesium, Corros Sci 82 (2014) 173 -- 179. [39] W. Li, D. Y. Li, Influence of surface morphology on corrosion and electronic behavior, Acta Mater 54 (2006) 445 -- 452. [40] T. Hong, M. Nagumo, Effect of surface roughness on early stages of pitting corrosion of type 301 stainless steel, Corros Sci 39 (1997) 1665 -- 1672. Corrosion resistance improvement of mg alloy az31 by combin- ing bilayer amorphous dlc: H/sinx film with n+ ions implanta- tion, Journal of Alloys and Compounds 762 (2018) 171 -- 183. [53] P. S. Correa, C. F. Malfatti, D. S. Azambuja, Corrosion be- havior study of az91 magnesium alloy coated with methyltri- ethoxysilane doped with cerium ions, Prog Org Coat 72 (2011) 739 -- 747. [54] M. Nabizadeh, A. A. Sarabi, H. E. Mohammadloo, Compara- tive investigation of cu ion and adipic acid addition on electro- chemical and microstructure characteristics of vanadium con- version coating on az31 mg alloy, Surf Coat Tech 357 (2019) 1 -- 11. [55] Y. Chen, X. H. Wang, J. Li, J. L. Lu, F. S. Wang, Long-term anticorrosion behaviour of polyaniline on mild steel, Corros Sci 49 (2007) 3052 -- 3063. [56] A. Amirudin, D. Thierry, Application of electrochemical impedance spectroscopy to study the degradation of polymer- coated metals, Prog Org Coat 26 (1995) 1 -- 28. [57] J. N. Murray, Electrochemical test methods for evaluating or- ganic coatings on metals: an update. part iii: Multiple test pa- rameter measurements, Prog Org Coat 31 (1997) 375 -- 391. [58] A. Shi, S. Koka, J. Ullett, Performance evaluation on the weathering resistance of two usaf coating systems (standard 85285 topcoat versus fluorinated apc topcoat) via electrochemi- cal impedance spectroscopy, Prog Org Coat 52 (2005) 196 -- 209. [59] J. H. Park, G. D. Lee, A. Nishikata, T. Tsuru, Anticorrosive behavior of hydroxyapatite as an environmentally friendly pig- ment, Corros Sci 44 (2002) 1087 -- 1095. [41] K. Sasaki, G. T. Burstein, The generation of surface rough- ness during slurry erosion-corrosion and its effect on the pitting potential, Corros Sci 38 (1996) 2111 -- 2120. [60] Z. B. Wang, H. X. Hu, C. B. Liu, Y. G. Zheng, The effect of fluoride ions on the corrosion behavior of pure titanium in 0.05 m sulfuric acid, Electrochim Acta 135 (2014) 526 -- 535. [42] S. A. Salman, M. Okido, Self-assembled monolayers formed on az31 mg alloy, J Phys Chem Solids 73 (2012) 863 -- 866. [43] Y. H. Wang, W. Wang, L. Zhong, J. Wang, Q. L. Jiang, X. Y. Guo, Super-hydrophobic surface on pure magnesium substrate by wet chemical method, Appl Surf Sci 256 (2010) 3837 -- 3840. [44] R. W. Y. Poon, K. W. K. Yeung, X. Y. Liu, P. K. Chu, C. Y. Chung, W. W. Lu, K. M. C. Cheung, D. Chan, Carbon plasma immersion ion implantation of nickel-titanium shape memory alloys, Biomaterials 26 (2005) 2265 -- 2272. [45] T. T. Zhao, Y. Li, Y. Liu, X. Q. Zhao, Nano-hardness, wear resistance and pseudoelasticity of hafnium implanted niti shape memory alloy, J Mech Behav Biomed 13 (2012) 174 -- 184. [46] I. Nakatsugawa, R. Martin, E. J. Knystautas, Improving corro- sion resistance of az91d magnesium alloy by nitrogen ion im- plantation, Corrosion 52 (1996) 921 -- 926. [47] S. Akavipat, E. B. Hale, C. E. Habermann, P. L. Hagans, Effects of iron implantation on the aqueous corrosion of magnesium, Mater Sci Eng 69 (1985) 311 -- 316. [48] E. Leitao, M. Barbosa, M. F. Dasilva, J. C. Soares, J. P. Muller, Electrochemical studies of magnesium implanted with high- doses of light-ions, Nucl Instrum Meth B 39 (1989) 559 -- 562. [49] P. Girault, J. L. Grosseau-Poussard, J. F. Dinhut, L. Marechal, Influence of a chromium ion implantation on the passive be- haviour of nickel in artificial sea-water: An eis and xps study, Nucl Instrum Meth B 174 (2001) 439 -- 452. [50] J. Macdonald, Chapters 1-4: Impedance spectroscopy- Emphasising solid Materials and systems, John Wiley Sons, New York, 1987. [51] Y. Y. Chang, D. Y. Wang, Corrosion behavior of electroless nickel-coated aisi 304 stainless steel enhanced by titanium ion implantation, Surf Coat Tech 200 (2005) 2187 -- 2191. [52] H. Zhu, T. Zhao, Q. Wei, N. Liu, L. Ma, Z. Hu, Y. Wang, Z. Yu, [61] S. L. de Assis, S. Wolynec, I. Costa, Corrosion characterization of titanium alloys by electrochemical techniques, Electrochim Acta 51 (2006) 1815 -- 1819. [62] D. Bian, W. R. Zhou, J. X. Deng, Y. Liu, W. T. Li, X. Chu, P. Xiu, H. Cai, Y. H. Kou, B. G. Jiang, Y. F. Zheng, Devel- opment of magnesium-based biodegradable metals with dietary trace element germanium as orthopaedic implant applications, Acta Biomater 64 (2017) 421 -- 436. [63] ISO 10993-5:2009(E). Biological evaluation of medical de- vices, Part 5: Tests for in vitro cytotoxicity, International Or- ganization for Standardization, 2009. [64] J. D. Sun, Y. Zhu, L. Meng, P. Chen, T. T. Shi, X. Y. Liu, Y. F. Zheng, Electrophoretic deposition of colloidal particles on mg with cytocompatibility, antibacterial performance, and corrosion resistance, Acta Biomater 45 (2016) 387 -- 398. [65] S. Bagherifard, D. J. Hickey, S. Fintova, F. Pastorek, I. Fernandez-Pariente, M. Bandini, T. J. Webster, M. Guagliano, Effects of nanofeatures induced by severe shot peening (ssp) on mechanical, corrosion and cytocompatibility properties of mag- nesium alloy az31, Acta Biomater 66 (2018) 93 -- 108. [66] G. L. Song, A. Atrens, Understanding magnesium corrosion - a framework for improved alloy performance, Adv Eng Mater 5 (2003) 837 -- 858. [67] G. L. Song, A. Atrens, M. Dargusch, Influence of microstruc- ture on the corrosion of diecast az91d, Corros Sci 41 (1999) 249 -- 273. [68] H. P. Duan, C. W. Yan, F. H. Wang, Effect of electrolyte ad- ditives on performance of plasma electrolytic oxidation films formed on magnesium alloy az91d, Electrochim Acta 52 (2007) 3785 -- 3793.
1910.11896	2	1910	2019-10-29T01:56:02	Energy-efficient ultrafast SOT-MRAMs based on low-resistivity spin Hall metal Au0.25Pt0.75	[ "physics.app-ph", "cond-mat.mes-hall" ]	Many key electronic technologies (e.g., large-scale computing, machine learning, and superconducting electronics) require new memories that are fast, reliable, energy-efficient, and of low-impedance at the same time, which has remained a challenge. Non-volatile magnetoresistive random access memories (MRAMs) driven by spin-orbit torques (SOTs) have promise to be faster and more energy-efficient than conventional semiconductor and spin-transfer-torque magnetic memories. This work reports that the spin Hall effect of low-resistivity Au0.25Pt0.75 thin films enables ultrafast antidamping-torque switching of SOT-MRAM devices for current pulse widths as short as 200 ps. If combined with industrial-quality lithography and already-demonstrated interfacial engineering, our results show that an optimized MRAM cell based on Au0.25Pt0.75 can have energy-efficient, ultrafast, and reliable switching, e.g. a write energy of < 1 fJ (< 50 fJ) for write error rate of 50% (<1e-5) for 1 ns pulses. The antidamping torque switching of the Au0.25Pt0.75 devices is 10 times faster than expected from a rigid macrospin model, most likely because of the fast micromagnetics due to the enhanced non-uniformity within the free layer. These results demonstrate the feasibility of Au0.25Pt0.75-based SOT-MRAMs as a candidate for ultrafast, reliable, energy-efficient, low-impedance, and unlimited-endurance memory.	physics.app-ph	physics	Article type: Full Paper Title: Energy-efficient ultrafast SOT-MRAMs based on low-resistivity spin Hall metal Au0.25Pt0.75 Authors: Lijun Zhu, Lujun Zhu, Shengjie Shi, D. C. Ralph, and R. A. Buhrman Dr. Lijun Zhu, Dr. S. Shi, Prof. R. A. Buhrman Cornell University, Ithaca, New York 14850, USA Email: [email protected] Prof. Lujun Zhu College of Physics and Information Technology, Shaanxi Normal University, Xi'an 710062, China Prof. D. C. Ralph Cornell University, Ithaca, New York 14850, USA Kavli Institute at Cornell, Ithaca, New York 14850, USA Abstract: Many key electronic learning, and superconducting electronics) require new memories that are fast, reliable, energy-efficient, and of low- impedance at the same time, which has remained a challenge. Non-volatile magnetoresistive random access memories (MRAMs) driven by spin-orbit torques (SOTs) have promise to be faster and more energy-efficient than conventional semiconductor and spin-transfer-torque magnetic memories. This work reports that the spin Hall effect of low-resistivity Au0.25Pt0.75 thin films enables ultrafast antidamping-torque switching of SOT- MRAM devices for current pulse widths as short as 200 ps. If combined with industrial-quality lithography and already-demonstrated interfacial engineering, our results show that an optimized MRAM cell based on Au0.25Pt0.75 can have energy-efficient, ultrafast, and reliable switching, e.g. a write energy of < 1 fJ (< 50 fJ) for write error rate of 50% (<10-5) for 1 ns pulses. The antidamping torque switching of the Au0.25Pt0.75 devices is 10 times faster than expected from a rigid macrospin model, most likely because of the fast micromagnetics due to the enhanced non-uniformity within the free layer. These results demonstrate the feasibility of Au0.25Pt0.75-based SOT-MRAMs as a candidate for ultrafast, reliable, energy-efficient, low-impedance, and unlimited-endurance memory. Key words: Spin Hall effect, Spin-orbit torque, magnetoresistive random access memory, Write error rate large-scale computing, machine technologies (e.g., 1. Introduction Many key electronic technologies, e.g., large-scale computing, machine learning, and superconducting electronics, would benefit from the development of new fast, non-volatile, and energy-efficient memories.[1-3] While the conventional non-volatile 2-terminal spin-transfer-torque (STT) magnetoresistive random access memory (MRAM) is attractive for its good scalability and high thermal stability[4,5] during fast sub-ns write,[6] the required high write current density can exert severe stress on the magnetic tunnel junction (MTJ) and induce wear-out and breakdown of the MTJ barrier,[7] leading ultimately to degradation of the memory cell. Meanwhile, the shared read/write path can lead to write upon read errors. An alternative, 3-terminal spin-orbit- torque (SOT) MRAM [8,9] has the potential to mitigate these issues. In a SOT-MRAM, the spin current generated by the spin Hall effect (SHE)[10-13] of a heavy metal layer switches the magnetic free layer of a MTJ (see Figure 1a). The non-volatile SOT-MRAMs can have long data retention, zero standby power, and fast and reliable write.[7,14-17] SOT-MRAMs based on a spin Hall metal that combines a giant spin Hall ratio (θSH) with a relatively low resistivity (ρxx) can also have unlimited endurance due to the suppression of Joule heating induced bursting 1 and migration of the write line [2] as well as low values of write impedance that is compatible with superconducting circuits in cryogenic computing systems.[1] Recent harmonic response measurements [18,19] have indicated that the Au0.25Pt0.75 alloy can be a particularly compelling spin Hall metal for high-performance SOT-MRAM application due to the combination of a relatively 𝑗 low resistivity (ρxx ≈ 80 µΩ cm) with a giant antidamping SOT efficiency (𝜉DL ≈ 0.3-0.35). In this work, we show that prototype SOT-MRAM devices based on Au0.25Pt0.75 can achieve highly energy-efficient, ultrafast (down to 200 ps), and reliable switching. The antidamping torque switching of the Au0.25Pt0.75 devices is 10 times faster than expected from a rigid macrospin model, most likely because of the enhanced non-uniformity within the free layer. 2. Results and Discussion 2.1 Characterization of the SOT-MRAM Devices As shown in Figure 1b-d, the SOT-MRAM devices were lithographically patterned from sputter-deposited multilayer stacks consisting of Si/SiO2/Ta 1/Au0.25Pt0.75 5/Hf 0.5/ Fe0.6Co0.2B0.2 1.4/Hf 0.1/MgO 1.6/Fe0.6Co0.2B0.2 4/Pt 3/Ru 4 (numbers are layer thicknesses in nm), and were post annealed at 240 oC for 1 hour. The 0.1 nm- thick Hf spacer inserted at the Fe0.6Co0.2B0.2/MgO interface is to reduce the demagnetization field (4πMeff) of the free layer by enhancing the interfacial perpendicular magnetic anisotropy.[15] Low value of 4πMeff reduces the critical current for antidamping switching.[20,21] Ferromagnetic resonance measurements on the unpatterned films yield magnetic damping α = 0.027 ± 0.001 and 4πMeff = 0.460 ± 0.003 T for the magnetic free layer (see Figure S1). As determined by vibrating sample magnetometry, the saturation magnetization (Ms) of the Fe0.6Co0.2B0.2 layers is 1240 emu/cm3.[22,23] All measurements were performed at room temperature. Figure 1b is a top-view scanning electron microscopy (SEM) image of one of the SOT-MRAM devices, showing an elliptical (190×45 nm2) MTJ with the long axis transverse to the spin Hall channel and thus to the write-current flow. The spin Hall channel is 300 nm wide in the center where the pillar is located (Figure 1b and Figure S2) and 1.2 µm long as measured by the cross-sectional transmission electron microscopy (TEM) image (Figure 1c). For the sake of fabrication simplicity, we make electrical contact to the two ends of the channel using two micron-size trapezoid-shaped MTJ pillars (marked as "via" regions in Figure 1b,c, see Supporting Information for more details about the dimensions of the "via" and an estimate that each "via" contributes an effective series resistance of approximately 200 Ω). The channel resistance (Rch), including the "via" resistance and the series resistance from the two "vias" was 850 Ω, much lower than that of previously reported SOT- MRAMs from our laboratory (2.5-4.2 kΩ for W,[24] Ta,[9] Pt/Hf multilayers [23]) due to the reduced channel length and the relatively low ρxx of Au0.25Pt0.75.[18] Figure 1e,f show the major and minor magnetic switching loops of a representative device for in-plane magnetic fields applied along the long axis of the MTJ pillar. The major loop indicates a coercivity (Hc) of 450 Oe for the 4 nm-thick Fe0.6Co0.2B0.2 reference layer due to the shape anisotropy of the elliptical MTJ pillar. The minor loop is artificially centered after subtraction of the dipole field (Hdipole ≈150 Oe) from the 4 nm reference layer. The minor loop indicates an Hc of 15 Oe for the 1.4 nm free layer and a tunnel magnetoresistance ratio of 20% for the MTJ. 2.2 Direct Current Switching Figure 1g shows the characteristic switching behavior of the same SOT-MRAM device as the dc write current in the spin Hall channel is ramped quasi-statically (with an in-plane field equal to -Hdipole applied along the long axis of the MTJ pillar to compensate the dipole field from the reference layer). The MTJs show abrupt switching at write currents of ≈75 μA. Since thermal fluctuations assist the reversal of a nanoscale MTJ during slow current ramps,[21,25,26] we carried out ramp rate measurements (Figure 1h). Within the macrospin model, the switching current Ic should scale with the ramp rate (𝐼) following [20] 𝐼c = 𝐼c0 (1 + 1 𝛥 ln 𝑡0𝛥𝐼 𝐼𝑐0 ), (1) where Ic0 is the critical switching current in absence of thermal fluctuations, Δ the stability factor equal to the magnetic energy barrier for reversal between the P and AP states normalized by the thermal energy kBT, and 𝑡0 2 the thermal attempt time which we assume to be 1 ns. By fitting the data to Equation (1), we obtain Ic0 = 312 ± 11 μA for P→AP and 356 ± 14 μA for AP→P switching, and Δ ≈ 28 ± 2. These results were consistently reproduced by other devices. For practical application, Δ can be increased significantly even for sub-100-nm devices by optimizing the shape anisotropy during pillar etching process and by introducing tensile strain anisotropy.[27] Considering a parallel resistor model,[23] the current shunted into the Fe0.6Co0.2B0.2 free layer and Hf spacers (ρPt/Hf ≈ 80 μΩ cm, ρFeCoB≈ρHf≈130 μΩ cm) can be estimated to be ≈ 0.2Ic0. The critical switching density in the Au0.25Pt0.75 spin Hall channel is, therefore, jc0 = (1.86 ± 0.08)×107 A/cm2 for P→AP switching (2.12 ±0.07)×107 A/cm2 for AP→P switching, which are a factor of 2 lower than that for devices with pure Pt channels [14] in which α and 4πMeff are even smaller (Table 1). Within the simple macrospin model, jc0 for antidamping torque switching of an in-plane magnetized MTJ can be estimated by [21,26] 𝑗 jc0 = (2e/ћ)Aμ0Mstα(Hc+4πMeff/2)/𝜉DL , (2) to be 0.30 ± 0.07, which is consistent with our previous harmonic response measurements on where the factor A ≈ exp(-dHf/λs,Hf) denotes the attenuation of spin current by the Hf spacer layer in between the spin Hall channel and the magnetic free layer of the MTJ. With the Hf thickness dHf = 0.5 nm and the spin diffusion length λs,Hf = 0.9 ± 0.2 nm,[28] we determine A ≈ 0.57 ± 0.08 for the SOT-MRAMs. From Equation (2) 𝑗 we estimate 𝜉DL 𝑗 Au0.25Pt0.75/Co bilayers without a Hf spacer (𝜉DL 𝑗 significantly higher than those previously obtained in W devices (𝜉DL 𝑗 Pt0.85Hf0.15 devices (𝜉DL 𝑗 account (note that A was assumed to be unity in previous reports [14,15,17,23] when calculating 𝜉DL ). The SOT 𝑗 ≈ 0.29, ρxx≈140 µΩ cm [23]), but the lower- efficiency is similar to [Pt 0.6/Hf 0.2]6/Pt 0.6 multilayers (𝜉DL resistivity Au0.25Pt0.75 (ρxx ≈ 85 µΩ cm) is more favorable for applications that require unlimited endurance [2] and low device impedance.[1] Au0.25Pt0.75 is also thermally stable as indicated by the constant ρxx and θSH upon annealing to 400 oC.[19] ≈ 0.30 is ≈ 0.12),[14] and ≈ 0.23)[17] when the spin current attenuation by the Hf spacer layers (A ≤1) is taken into 𝑗 ≈ 0.3-0.35).[18] As compared in Table I, the value of 𝜉DL 𝑗 ≈ -0.20),[24] Pt devices (𝜉DL 2.3. Ultrafast and reliable pulse current switching We characterized the performance of the MRAMs in the short-pulse regime using a measurement method similar to that described in Refs. [15,17]. Figure 2a shows the switching phase diagram in the pulse width (τ) regime of 0.2-6 ns for the two cases AP→P and P→AP, respectively. Each data point is the statistical switching probability result of 1000 switching attempts. In determining the current values plotted in Figure 2, we have taken into account the impedance discontinuity between the 50 Ω cable and the MRAM channel,[17,29] so that the currents quoted denote the real pulse magnitudes within the channel. We find that the low ρxx and giant θSH of Au0.25Pt0.75 [18] allow the MRAM device to be switched many millions of times in the sub-ns pulse regime with no indication of degradation in the MgO barrier or the spin Hall channel.[2] For 200 ps pulses, the write current (I) for 50% switching probability are 3 mA (AP→P) and 3.27 mA (P→AP), and for 400 ps pulses both are ≈ 2 mA. The write energy (Ewrite = I2Rchτ) of the Au0.25Pt0.75 device at the current corresponding to 50% switching probability is plotted as a function of τ in Figure S3. Ewrite is as low as 1 pJ, 1.4 pJ, and 2 pJ for 1 ns, 400 ps, and 200 ps switching, respectively. This is encouraging as the values of α, 4πMeff, and channel dimensions could all be reduced further by additional optimization so that the write current and energy can be decreased significantly (see below). While as discussed below it is apparent that our devices do not reverse as a rigid single domain when driven by strong SOT pulses, we can still parameterize a time scale (τ0) characteristic of the switching process from fits of the 50% switching probability points to the macrospin model prediction [21] I = I∞ (1+ τ0/τ), (3) where I∞ denotes the critical switching current at infinite pulse width. As shown in Figure 2b, we find τ0 = 1.52 ± 0.02 ns and I∞ = 0.441 ± 0.005 mA for P→AP switching and τ0 =0.86 ± 0.01 ns and I∞ = 0.617 ± 0.005 mA for AP→P switching. The write current density (j∞) of 2.1 (3.0)×107 A/cm2 for P→AP (AP→P) switching is higher than the zero-temperature dc switching current density from the ramp rate. This difference represents initial evidence that the SOT-induced magnetic reversal in the short-pulse regime is not well-described by macrospin dynamics, because I∞ and Ic0 should be close in the case that a macrospin moment is excited by the antidamping 3 spin torque.[21] It has been a consistent observation that, in the short pulse regime, the spin-torque switching of the in- plane magnetized SOT-MRAMs (τ0 < 2 ns)[2,14-16] and metallic spin valves (τ0 ≈ 1 ns)[26,30] are much faster than the prediction of the macrospin model. We find that the Au0.25Pt0.75 devices are more than a fact of 10 faster and more energy-efficient than that expected for a rigid macrospin. As indicated by Bedau et al.,[26] the characteristic time for antidamping torque switching of a macrospin nanomagnet can be estimated as τ0 ≈ (4πMeffαγ)-1 (4) where γ is the gyromagnetic ratio. With the experimental values of α and 4πMeff of the actual Au0.25Pt0.75 devices, Eq. (4) yields τ0 ≈18 ns for a macrospin reversal process, much slower than our measurements. Because the switching can be quite fast, our Au0.25Pt0.75 devices are much more energy efficient than expected by a rigid macrospin model in short pulse regime. For example, for the pulse width of 200 ps (1 ns), the required switching current for the Au0.25Pt0.75 device is 6I∞ (2I∞), markedly smaller than 90I∞ (20I∞) predicted by the macrospin simulation (see Figure 2b). Understanding the switching mechanism of the in-plane devices are the key to develop ultrafast memory for technological applications. Here we attribute the observed ultrafast switching mainly to the enhanced the non- uniform micromagnetic dynamics within the free layer of our devices. As have been suggested by previous efforts,[31-34] the antidamping torque switching of the in-plane magnetized free layer is achieved via a fast evolution of non-uniform micromagnetic dynamics rather than via a coherent macrospin reversal within the free layer. The magnetic non-uniformity of the free layer should enhance the micromagnetic dynamics and speed up the switching. As schematically shown in Figure 2d, the SOT-MRAMs fabricated in our group have substantial tapering (see Ref. [23] for TEM imaging of the tapering) in the MTJ free layer that results from the ion-milling process, which should result in spatially non-uniform SOTs and dipole field within the magnetic free layer. There is also strong interfacial Dzyaloshinskii-Moriya interaction (DMI)[35] and magnetic roughness (variations of thickness and interfacial magnetic anisotropy field)[22] at the Pt (alloy)/FM interfaces, which should enhance the magnetic non-uniformity.[36] This explains the fact that our Pt (alloy) based in-plane SOT-MRAMs (τ0 ≈ 1 ns) with enhanced tapering and DMI are typically faster than the W devices [2] and the TaB devices (τ0 ≈ 3.3-3.4 ns as fitted in Figure S4)[37,38] where both the tapering of free layer and the interfacial DMI are relatively weak.[39] Another factor that could assist the evolution of non-uniform dynamics is the current-induced effective transverse field (Heff), which is the sum of the fieldlike SOT effective field (HFL) and Oersted field (HOe) in the SOT-MRAM geometry. Early micromagnetic simulations [14,31] show that Heff, if parallel to the spin polarization of the spin Hall current, can speed up the non-uniform dynamics and thus the switching of the free layer. As indicated by the dc switching phase diagrams (Hc vs I) in Figure 2e, for our Au0.25Pt0.75 devices a positive (negative) charge current induced Heff reduce Hc for AP→P (P→AP) switching. This indicates that Heff is parallel to the spin polarization of the spin current from the Au0.25Pt0.75 channel and therefore could play an assisting role in antidamping torque switching of the free layer. However, the magnetic non-uniformity of the free layer turns out to be more critical than Heff in determining the switching speed. As we compare in Figure 2e,f, the device discussed above (Heff / I ≈ 115 Oe/mA, I∞=0.44 mA, denoted as Device A) is not as fast as a control device (denoted as Device B, the stack is Au0.25Pt0.75 4/FeCoB 1.6/MgO 2/FeCoB 4) that has two times smaller effective field (Heff / I ≈24 Oe/mA, I∞=1.09 mA). This difference in the τ0 values is likely suggestive of a less significant non-uniformity in the free layer of Device A than in the Device B. A thin Hf layer has been consistently found to reduce the interfacial spin-orbit coupling (thus most likely interfacial DMI) at heavy metal/ferromagnetic interfaces.[19] Finally, Heff in the short strong pulse region can exceed the value of Hc (Hc= 15 Oe for Device A, 40 Oe for Device A), which might be reminiscent of a switching driven directly by Heff. However, the fact that Device B is faster than Device A reaffirms that it is the antidamping torque rather than Heff which dominates the switching process. This conclusion is also supported by the rhombehedral shape of the bistable region (P/AP) in the dc phase diagram (Figure 2e). We speculate that magnetization switching by an effective field that is collinear with the magnetization seems to be slow because the excitation of the dynamics likely requires accumulation of thermal fluctuation for a nonzero initial torque. In any case, the very short characteristic switching time of our devices is a very positive observation for application and motivates further study on the switching mechanisms in 3-terminal SOT-MRAMs. For memory, SOT reversal must be both fast and highly reliable. While the limited-statistics switching probability for pulse current and duration sweeps (for instance, 1000 events in Figure 2a) are routinely used to report the existence of high-speed switching,[4,16,17] they do not convey the statistics of the write error rate 4 (WER) -- information that is critical for applications. We have tested the reliability of our Au0.25Pt0.75 devices by measuring WER statistics during up to 105 switching attempts. As shown in Figure 2c, WERs for the pulse duration of 1 ns scale down quickly as the write current increases, extrapolation of which indicates WERs of < 10-5 at 4 mA (2.2×108 A/cm2) for both the P→AP and the AP→P switching. Further decreases in the required write currents of Au0.25Pt0.75 SOT-MRAM devices can be expected from straightforward additional optimization of the stack materials and device dimensions. Interface engineering has already been demonstrated to significantly reduce both α and 4πMeff of the Fe0.6Co0.2B0.2 free layer.[19,20,22] For example, our optimized MRAMs based on a spin Hall channel of [Pt 0.6/Hf 0.2]6 multilayers [23] or W [14] achieved α of ≈0.011 and 4πMeff of ≈ 0.2 T, both of which are more than a factor of 2 less compared to our present Au0.25Pt0.75 devices. As shown in Fig. S5 we have found that a Pt 0.5/Hf 0.25 bilayer can effectively suppress α contributed by the interfacial spin-orbit coupling via spin memory loss [19] and two-magnon scattering,[22] despite that the 0.5 nm-thick Hf single-layer spacer that was inserted at the bottom of the 1.4 nm Fe0.6Co0.2B0.2 free layer appears ineffective in doing so probably because Hf does not wet Au surface and forms a discontinuous layer. The write current and the write power can be further reduced by additional factors of 3 and >18 by using more aggressive industry-level lithography to narrow the spin Hall channel from 300 nm to below 100 nm [27,37] and to shorten it from 1.2 µm to 200 nm,[7] and by replacing the magnetic stacks in the "via" regions (see Figure 1c and Figure S2) with highly conductive materials. These approaches in combination would lower the zero temperature switching current (density) of Au0.25Pt0.75 SOT-MRAM devices to Ic0 < 30 μA (jc0 < 4.8×106 A/cm2) and I∞ < 60 μA (j∞ < 9.6×106 A/cm2). The write energy for 50% switching probability for 1 ns will be < 1fJ. Assuming similar switching dynamics, for reliable switching with 1 ns pulse and the WER of <10-5, the write current (density) would scale to < 0.35 mA (5.6×107 A/cm2) and the write energy to < 50 fJ. 3. Conclusion We have demonstrated that Au0.25Pt0.75 is a particularly compelling spin Hall metal that can enable very energy- efficient and ultrafast switching of in-plane-magnetized SOT-MRAMs due to the combination of a giant spin 𝑗 Hall effect (𝜉DL ≈ 0.30) and a low resistivity (ρxx ≈ 80 µΩ cm). We have demonstrated switching of prototypical SOT-MRAM structures with 50% probability using I ≈ 3 mA and Ewrite = 2 pJ for 200 ps current pulses, and write error rates < 10-5 at I = 4 mA and Ewrite = 14 pJ for 1 ns pulses. We extrapolate that further reductions of α and 4πMeff of the free layer (as already demonstrated in other SOT-MRAM structures) along with improved lithography to reduce the dimension of the spin Hall channel can enable 1 ns switching of SOT-MRAM devices with write energy < 50 fJ and WER of < 10-5. The relatively low channel resistance due to the low ρxx of Au0.25Pt0.75 is beneficial for decreasing write energies, achieving unlimited endurance, and also for matching the impedance of superconducting circuits in cryogenic computation systems. We find that the current-induced SOT switches the Au0.25Pt0.75-based MRAMs much faster than expected from a rigid macrospin model, most likely due to the rapid micromagnetics within the free layer that is enhanced by the spatial non-uniformities in the free- layer magnetization that may be induced by DMI, interfacial magnetic roughness, and/or tapering in the MTJ free layer. If combined with the strain and voltage gating architectures proposed in the industry,[27,37,38] the Au0.25Pt0.75-based in-plane SOT-MRAMs can be also very dense. The non-volatile MRAM also have long data retention and zero standby power. These results indicate that the Au0.25Pt0.75-based SOT-MRAM is a good candidate for ultrafast, energy-efficient, low-impedance, unlimited-endurance memory for large scale computing systems, machine-learning systems, and superconducting electronics. Experimental section Sample growth and device fabrication: All of the samples are sputter deposited at room temperature with an argon pressure of 2 mTorr and a base pressure of ~ 1×10-8 Torr. A highly resistive oxidized Si substrate (ρxx > 1010 µΩ cm) was used to avoid current shunting into the substrate during the direct current and the pulse current switching measurements. The 1 nm Ta layer at the bottom was used to improve the smoothness and adhesion of the Au0.25Pt0.75. The top bilayers of Pt 3 nm/Ru 4 nm were used to protect the multilayers during device fabrication. The multilayer samples are patterned into 3-terminal MRAM devices schematically shown in Figure S2 with a three-step procedure. First, we defined the spin Hall channel using DUV lithography (ASML) and ion beam etching and measured the channel size to be 300×600 nm2 by atomic force microscopy (Veeco Icon). We then defined the elliptical MTJ nanopillars with different aspect ratios and μm-size "via" pillars (as vertical 5 connector between the bottom channel to top contact) onto the spin Hall channel with e-beam lithography (JEOL JBX-6300FS) and ion beam etching, and isolated the pillars with 80 nm thick SiO2 deposited by an e-beam evaporator. Finally, contacts of Ti 5 nm/Pt 50 nm were sputter-deposited on the top of the MTJ pillars and "via" pillars for electrical measurements. Measurements: For the dc switching measurements of the MRAM devices, a lock-in amplifier was used to read the differential resistance of the magnetic tunnel junctions with a 0.1 V oscillatory voltage applied onto MTJ pillars series-connected to 10 MΩ resistor (read current ≈ 1 μA). A Keithley 2400 source-meter was used to source write current into the spin Hall channels. For the short pulse measurement, the pulse was generated using a picosecond pulse generator and the MTJ resistance was measured with a NI-DAQ (voltmeter) and a Keithley 2450 (current source). A vibrating sample magnetometer was used to determine the sample magnetization. Flip- chip ferromagnetic resonance was used to determine the magnetic damping constant and the effective demagnetization field of the Fe0.6Co0.2B0.2 free layer on large-area unpattern chips by sweeping an in-plane magnetic field at each fixed microwave frequency (see Figure S1). The MRAM devices were characterized by cross-sectional scanning transmission electron microscopy (TEM) imaging in a spherical-aberration-corrected (Cs-corrected) 300-kV FEI Titan G2 microscope. Acknowledgements This work was supported in part by the Office of Naval Research (N00014-15-1-2449) and by the NSF MRSEC program (DMR-1719875) through the Cornell Center for Materials Research. The devices were fabricated, in part, at the Cornell NanoScale Facility, an NNCI member supported by NSF Grant No. ECCS-1542081. The TEM measurements performed at Shaanxi Normal University were supported by the Science and Technology Program of Shaanxi Province (Grant No. 2019JQ-433) and the Fundamental Research Funds for the Central Universities (Grant No. GK201903024). Supporting information: Supporting Information is available from the Wiley Online Library or from the author. Conflict of Interest The authors declare no conflict of interest. References [1] D. S. Holmes, A. L. Ripple, and M. A. Manheimer, IEEE Trans. Appl. Supercond. 2013, 23, 1701610. [2] Y. Shiokawaa, E. Komura, Y. Ishitani, A. Tsumita, K. Suda, Y. Kakinuma, and T. Sasaki, AIP Adv. 2019, 9, 035236. [3] T. Endoh, 2019 Silicon Nanoelectronics Workshop (SNW), 2019, DOI: 10.23919/SNW.2019.8782898. [4] A. D. Kent and D. C. Worledge, Nat. Nanotechnol. 2015, 10,187. [5] K. Watanabe, B. Jinnai, S. Fukami, H. Sato, H. Ohno, Nat. Commun. 2018, 9, 663. [6] G. Jan, L. Thomas, S. Le, Y.-J. Lee, H. Liu, J. Zhu, J. Iwata-Harms, S. Patel, R.-Y. Tong, S. Serrano-Guisan, D. Shen, R. He, J. Haq, J. Teng, V. Lam, R. Annapragada, Y.-J. Wang, T. Zhong, T. Torng, and P.-K. Wang, 2016 IEEE Symposium on VLSI Technology, 2016, DOI: 10.1109/VLSIT.2016.7573362. [7] K. Garello, F. Yasin, S. Couet, L. Souriau, J. Swerts, S. Rao, S. Van Beek, W. Kim, E. Liu, S. Kundu, D. Tsvetanova, N. Jossart, K. Croes, E. Grimaldi, M. Baumgartner, D. Crotti, A. Furnémont, P. Gambardella, G.S. Kar, 2018 IEEE Symposium on VLSI Circuits, 2018, DOI: 10.1109/VLSIC.2018.8502269. [8] I. M. Miron, G. Gaudin, S. Auffret, B. Rodmacq, A. Schuhl, S. Pizzini, J. Vogel,P. Gambardella, Nat. Mater. 2010, 9, 230 -- 234. [9] L. Liu, C.-F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, R. A. Buhrman, Science 2012, 336, 555. [10] M. I. Dyakonov and V. I. Perel, Phys. Lett. A 1971, 35, 459. [11] J. Hirsch, Phys. Rev. Lett. 1999, 83, 1834. [12] S. Zhang, Phys. Rev. Lett. 2000, 85, 393. [13] L. Zhu, L. Zhu, M. Sui, D. C. Ralph, and R. A. Buhrman, Sci. Adv. 2019, 5, eaav8025. 6 [14] S. V. Aradhya, G. E. Rowlands, J. Oh, D. C. Ralph, R. A. Buhrman, Nano. Lett. 2016, 16, 5987 -- 5992. [15] S. Shi, Y. Ou, S.V. Aradhya, D. C. Ralph, R. A. Buhrman, Phys. Rev. Appl. 2018, 9, 011002. [16] M. Cubukcu, O. Boulle, N. Mikuszeit, C. Hamelin, T. Brächer, N. Lamard, M.-C. Cyrille, L. Buda- Prejbeanu, K. Garello, I. M. Miron, O. Klein, G. de Loubens,V. V. Naletov, J. Langer, B. Ocker, P. Gambardella, and G. Gaudin, IEEE Trans. Magn. 2018, 54, 9300204. [17] M.-H. Nguyen, S. Shi, G. E. Rowlands, S. V. Aradhya, C. L. Jermain, D. C. Ralph, R. A. Buhrman, Appl. Phys. Lett. 2016,112, 062404. [18] L. Zhu, D. C. Ralph, R. A. Buhrman, Phys. Rev. Appl. 2018, 10, 031001. [19] L. Zhu, D. C. Ralph, R. A. Buhrman, Phys. Rev. Lett. 2019, 122, 077201. [20] M.-H. Nguyen, C.-F. Pai, K. X. Nguyen, D. A. Muller, D. C. Ralph, and R. A. Buhrman, Appl. Phys. Lett. 2015, 106, 222402. [21] J. Z. Sun, Phys. Rev. B 2000, 62, 570. [22] L. Zhu, D. C. Ralph, R. A. Buhrman, Phys. Rev. Lett. 2019,123, 057203. [23] L. Zhu, L. Zhu, S. Shi, M. Sui, D.C. Ralph, R.A. Buhrman, Phys. Rev. Appl. 2019, 11, 061004. [24] C.-F. Pai, L. Liu, Y. Li, H. W. Tseng, D. C. Ralph, R. A. Buhrman, Appl. Phys. Lett. 2012, 101, 122404. [25] E. B. Myers, F. J. Albert, J. C. Sankey, E. Bonet, R. A. Buhrman, D. C. Ralph, Phys. Rev. Lett. 2002, 89, 196801. [26] D. Bedau, H. Liu, J. Z. Sun, J. A. Katine, E. E. Fullerton, S. Mangin, and A. D. Kent, Appl. Phys. Lett. 2010, 97, 262502. [27] Y. Ohsawa, H. Yoda, S. Shirotori, M. Shimizu, B. Altansargai, H. Sugiyama, N. Shimomura, K. Koi, Y. Kato, S. Oikawa, T. Inokuchi, and A. Kurobe, 2019 Electron Devices Technology and Manufacturing Conference (EDTM), 2019, DOI: 10.1109/EDTM.2019.8731241. [28] Y. Ou, C. F. Pai, S. Shi, D. C. Ralph, R. A. Buhrman, Phys. Rev. B 2016, 94, 140414(R). [29] C. Wang, Y.-T. Cui, J. A. Katine, R. A. Buhrman, and D. C. Ralph, Nat. Phys. 2011, 7, 496. [30]P. M. Braganca, I. N. Krivorotov, O. Ozatay, A. G. F. Garcia, N. C. Emley, J. C. Sankey, D. C. Ralph, and R. A. Buhrma，Appl. Phys. Lett. 2005, 87, 112507. [31] G. E. Rowlands, S. V. Aradhya, S. Shi, E. H. Yandel, J. Oh, D. C. Ralph, and R. A. Buhrman, Appl. Phys. Lett. 2017,110, 122402. [32] P. M. Braganca, O. Ozatay, A. G. F. Garcia, O. J. Lee, D. C. Ralph, and R. A. Buhrman, Phys. Rev. B 2008, 77, 144423. [33] S. Emori, U. Bauer, S.-M. Ahn, E. Martinez and G. S. D. Beach, Nat. Mater. 2013, 12, 611. [34] M. Baumgartner, K. Garello, J. Mendil, C. O. Avci, E. Grimaldi, C. Murer, J. Feng, M. Gabureac, C. Stamm, Y. Acremann, S. Finizio, S. Wintz, J. Raabe and P. Gambardella, Nat. Nanotechnol. 2017, 12, 980. [35] L. J. Zhu, K. Sobotkiewich, X. Ma, X. Li, D. C. Ralph, R. A. Buhrman, Adv. Funct. Mater. 2019, 29, 1805822. [36] M. Cubukcu, J. Sampaio, K. Bouzehouane, D. Apalkov, A. V. Khvalkovskiy, V. Cros, and N. Reyren, Phys. Rev. B 2016,93, 020401(R). [37] T. Inokuchi, H. Yoda, K. Koi, N. Shimomura, Y. Ohsawa, Y. Kato, S. Shirotori, M. Shimizu, H. Sugiyama, S. Oikawa, B. Altansargai, and A. Kurobe, Appl. Phys. Lett. 2019,114, 192404. [38] S. Shirotori, H. Yoda, Y. Ohsawa, N. Shimomura, T. Inokuchi, Y. Kato, Y.Kamiguchi, K. Koi, K. Ikegami, H. Sugiyama, M. Shimizu, B. Altansargai, S. Oikawa, M. Ishikawa, IEEE. Trans. Magn. 2017, 53, 3401104. [39] X. Ma, G. Yu, C. Tang, X. Li, C. He, J. Shi, K. L. Wang, and X. Li, Phys. Rev. Lett. 2018, 120, 157204. 7 Figure 1. a) Schematic of a SOT-MRAM with an in-plane MTJ and a spin Hall channel. b) Top-view SEM image, c) Cross-sectional TEM image (dark field), d) Sample stack, e) Magnetic major loop, and f) Magnetic minor loop of the SOT-MRAM devices. g) DC switching loop, h) DC switching currents vs current ramp rate for P→AP (red circles) and AP→P (blue squares) switching as a function of current ramp rate. In b), the two yellow dashed lines indicates the area of the spin Hall channel; in h) the solid lines represents the best fit of the data to Equation (1). 8 Figure 2. Fast pulse switching of the Au0.25Pt0.75-based MRAM device. (a) Pulse switching phase diagrams (the color scale represents the switching probability for 1000 events), (b) Rescaled write current for AP→P (red) and for P→AP switching (black) plotted as a function of pulse width (50% switching probability), (c) The write error rates (WERs) with 1 ns pulse (105 events) plotted as a function of write current, and (d) Schematic depict of non- uniformities within the free layer. (e) DC phase diagrams and (f) Rescaled write current (P→AP switching, 50% switching probability) for two devices with different strength of current-induced effective field (Heff /I ≈ 115 Oe/mA for Device A and 24 Oe/mA for Device B). In (a), the green dots indicate the 50% switching probability points; in (b), the dashed gray lines denotes the macrospin reversal with a critical switching time of 18 ns as calculated using Equation (4); in (b) and (f) the solid lines represent the best fits to Equation (3) of the 50% switching probability points. Table 1. Comparison of the parameters of the SOT-MRAM devices calculated using Equation (2), indicating the effectiveness of Au0.25Pt0.75 in generating dampinglike spin-orbit torque. The factor A ≈ exp(-dHf/λs,Hf) denotes the attenuation of spin current by the Hf spacer layer (the thickness dHf and the spin diffusion length λs,Hf) in between the spin Hall channel and the magnetic free layer of the magnetic tunnel junction. Rch is the resistance of the spin 𝑗 Hall channel, jc0 critical dc switching current density, α magnetic damping, 4πMeff the demagnetization field, 𝜉DL the dampinglike spin-orbit torque efficiency per unit current density. Structure Au0.25Pt0.75 5/Hf 0.5/Fe0.6Co0.2B0.2 1.4/Hf 0.1 [Pt 0.6/Hf 0.2]5/Pt 0.6/Fe0.6Co0.2B0.2 1.6 W 4.4/Hf 0.25/ Fe0.6Co0.2B0.2 1.8/Hf 0.1 [Pt 0.6/Hf 0.2]6/Pt 0.6/Hf 0.25/Fe0.6Co0.2B0.2 1.6/Hf 0.1 0.76±0.05 0.76±0.05 0.46±0.09 0.46±0.09 1.05 Pt0.85Hf0.15 6/Hf 0.7/ Fe0.6Co0.2B0.2 1.4 Pt 5/Hf 0.7/Fe0.6Co0.2B0.2 1.6 Ta 6.2/Fe0.4Co0.4B0.2 1.6 3 1 9 A Rch (kΩ) 0.57±0.08 0.85 4.3 3.8 3.6 2.5 1 jc0 (107 A/cm2) 2.0 1.0 0.36 0.54 1.4 4.0 3.7 α 4πMeff (T) 0.027 0.460 0.017 0.553 0.011 0.197 0.012 0.211 0.017 0.362 0.018 0.4165 0.021 0.76 𝑗 𝜉DL Refs 0.30 This work 0.29 0.23 -0.20 0.23 0.12 -0.12 [23] [23] [15] [17] [14] [9] Supporting Information for Energy-efficient ultrafast SOT-MRAMs based on low-resistivity spin Hall metal Au0.25Pt0.75 Lijun Zhu1, Lujun Zhu2, Shengjie Shi1, D. C. Ralph1,3, and R. A. Buhrman1 1. Cornell University, Ithaca, New York 14850, USA 2. College of Physics and Information Technology, Shaanxi Normal University, Xi'an 710062, China 3. Kavli Institute at Cornell, Ithaca, New York 14850, USA * [email protected] Section 1. Ferromagnetic resonance measurement Section 2. Write impedance of the SOT-MRAMs Section 3. Calculated write power at 50% switching probability Section 4. Literature data on the pulse switching of SOT-MRAMs Section 5. Interfacial engineering of the damping Section 1. Ferromagnetic resonance measurement Figure S1. Ferromagnetic resonance measurement on the free layer of Device A. a) FMR linewidth ΔH vs the resonance frequency f, b) frequency vs FMR resonance field Hr. The solid lines in (a) and (b) represent the best fits to ΔH = ΔH0 + (2π/γ)αf, and f = (γ/2π)√𝐻𝑟(𝐻𝑟 + 4𝜋𝑀eff), respectively. ΔH0 and γ are the inhomogeneous broadening of the FMR linewidth and the gyromagnetic ratio, respectively. 10 Section 2. Write impedance of the SOT-MRAMs The measured write impedance of our SOT-MRAM devices includes the resistance of the channel and the two "vias". The width of the spin Hall channel varies gradually from ≈10 µm (at two ends, see Figure S2) to 300 nm (at the center, see Figure 1b). Due to the spreading effect, the current flow is not uniform in the channel, especially at the two thick ends. As shown in Figure S2, the "vias" are trapezoid-shaped with the width narrowing gradually from 8 µm at the end far from the center of the channel to 1 µm close to the center of the channel. If we assume a uniform current flow in the "vias" during write of the device, the total resistance of the "vias" can be estimated to be ≈170 Ω by comparing the areas of the vias (15 µm2) and the MTJ pillar (190×45 nm2). However, because of the spreading effect, the current is mostly flowing at the narrow end of vias rather than flowing uniformly in the "via" region, which makes the electrically effective area of each "via" much less than 15 µm2. A control Pt-based MRAM device with a similar device dimensions shows that the measured "channel resistance" is reduced by 400 Ω when the magnetic stack in "via" regions were etched away and filled with highly conductive Ti/Pt. Therefore, we can conclude that the resistance contribution of the two "via" regions should be approximately 400 Ω for our devices. Figure S2. Schematic of the 3-terminal MRAM device. a) Dimensions of the "via" region; b) Spreading effect of the current flow; c) Side-view of the MRAM devices, indicating that the two μm-size "via" pillars are consist of the same magnetic stack as the nanopillar of the magnetic tunnel junction (MTJ) and contribute to the measured channel resistance of the device. 11 Section 3. Calculated write power at 50% switching probability Figure S3. Write energy (50% switching probability) of a typical SOT-MRAM device based on Au0.25Pt0.75. Section 4. Literature data on the pulse switching of SOT-MRAMs Figure S4. Write pulse width dependence of switching current density for 50% probability for in-plane magnetized SOT-MRAM based on a spin Hall channel of W (red dots, ref. 2) and TaB (blue circles, ref. 37). Fitting the data to Equation (3) in the maintext yields the critical switching time (τ0) of 3.4 ns for the W device and 3.3 ns for the TaB device. 12 Section 5. Interfacial engineering of the damping Figure S5. The FMR linewidth vs frequency for AuPt 4/FeCoB 1.6 and AuPt 4/Pt 0.5/Hf 0.25/FeCoB 1.6, indicating a substantial reduction of damping due to the insertion of the Pt 0.5/Hf 0.25 bilayer spacer. 13
1712.05178	2	1712	2017-12-15T11:30:26	Photostable single-photon emission from self-assembled nanocrystals of polycyclic aromatic hydrocarbons	[ "physics.app-ph", "cond-mat.mes-hall", "cond-mat.mtrl-sci", "physics.optics", "quant-ph" ]	Quantum technologies could largely benefit from the control of quantum emitters in sub-micrometric size crystals. These are naturally prone to the integration in hybrid devices, including heterostructures and complex photonic devices. Currently available quantum emitters sculpted in nanocrystals suffer from spectral instability, preventing their use as single photon sources e.g., for most quantum optics operations. In this work we report on unprecedented performances of single-photon emission from organic nanocrystals (average size of hundreds \SI{}{\nano\meter}), made of anthracene (Ac) and doped with dibenzoterrylene (DBT) molecules. The source has hours-long photostability with respect to frequency and intensity, both at room and at cryogenic temperature. When cooled down to \SI{3}{K}, the 00-zero phonon line shows linewidth values (\SI{50}{MHz}) close to the lifetime-limit. Such optical properties in a nanocrystalline environment make the proposed organic nanocrystals a unique single-photon source for integrated photonic quantum technologies.	physics.app-ph	physics	Photostable single-photon emission from self-assembled nanocrystals of polycyclic aromatic hydrocarbons Sofia Pazzagli,∗,†,‡ Pietro Lombardi,‡,¶ Daniele Martella,¶ Maja Colautti,¶ Bruno Tiribilli,§ Francesco Saverio Cataliotti,†,‡,¶,(cid:107) and Costanza Toninelli∗,‡,¶,(cid:107) †Dipartimento di Fisica ed Astronomia, Università di Firenze, Via Sansone 1, I-50019 Sesto F.no, Firenze, Italy ‡CNR-INO, Istituto Nazionale di Ottica, Via Carrara 1, 50019 Sesto F.no, Firenze, Italy ¶LENS and Università di Firenze, Via Carrara 1, 50019 Sesto F.no, Firenze, Italy §CNR-ISC Istituto dei Sistemi Complessi, via Madonna del Piano 10, I-50019 Sesto F.no, Firenze, Italy (cid:107)QSTAR, Largo Fermi 2, I-50125 Firenze, Italy E-mail: [email protected]; [email protected] Abstract Quantum technologies could largely benefit from the control of quantum emitters in sub-micrometric size crystals. These are naturally prone to the integration in hybrid devices, including heterostructures and complex photonic devices. Currently available quantum emitters sculpted in nanocrystals suffer from spectral instability, preventing their use as single photon sources e.g., for most quantum optics operations. In this work we report on unprecedented performances of single-photon emission from organic nanocrystals (average size of hundreds nm), made of anthracene (Ac) and doped with 1 dibenzoterrylene (DBT) molecules. The source has hours-long photostability with re- spect to frequency and intensity, both at room and at cryogenic temperature. When cooled down to 3 K, the 00-zero phonon line shows linewidth values (50 MHz) close to the lifetime-limit. Such optical properties in a nanocrystalline environment make the proposed organic nanocrystals a unique single-photon source for integrated photonic quantum technologies. The ubiquitous deployment of nanocrystals (NCs) in photonics stems from the impressive tunability of their physical and chemical properties, combined with the nano-positioning op- portunities offered by support-free colloids and from the possibility of mass-production at low costs.1 These features have also promoted NCs as efficient biological markers for imaging,2,3 color filters in liquid crystal displays,4 as well as functionalizing elements in light emitting and light harvesting devices.5 On the other hand, advanced nanophotonic applications are emerg- ing based on the generation, manipulation and detection of single photons.6,7 Indeed, lever- aging single-photon statistics and quantum coherence for sub-diffraction imaging,8 quantum cryptography,9 simulation,10 enhanced precision measurements and information processing11 have become roadmap targets for the next 10-20 years.12 Single-photon sources based on quantum emitters hold promise for these applications because of their on-demand opera- tion.13–17 However, despite great efforts in the last years to attain controllable sources by coupling solid-state emitters to nanophotonic structures, each platform privileges either the freedom in the device design18–23 or the quality of single-photon emission.24,25 Deterministic positioning and control of quantum emitters remains elusive for epitaxial quantum dots,26–29 color centers in bulk diamond30,31 and organic molecules in crystalline matrices.32–35 On the other hand, versatile approaches based on today-available NCs present important short- comings with respect to single-photon applications. Photoinduced charge rearrangements in the passivation layer and in the environment of inorganic semiconductors quantum-dot NCs36,37 lead to spectral instability of the exciton line,38 hindering basic quantum optics operations with the emitted photons. Moreover, intermittence in the photoluminescence,39 2 named blinking, seriously affects the average fluorescence quantum yield and hence the pho- ton state purity. Although important results have been obtained by improving synthesis protocols40 or introducing perovskite materials,41,42 the emitter photostability in time or frequency is still below expectations. Notably, similar issues characterize the emission of color centres in nanodiamonds, including those which possess superb optical properties in bulk such as the widely studied negatively charged silicon vacancy,16,43 or chromium-related defects.44 Hence, despite the wealth of materials and protocols, there still are fundamental limitations for the use of NCs in single-photon applications. We here propose and report on self-assembled and support-free organic NCs (hundreds nm in size) of anthracene doped with single fluorescent dibenzoterrylene molecules (DBT:Ac). We demonstrate that the remarkable features of the bulk system,45–47 belonging to the family of Polycyclic Aromatic Hydrocarbons (PAH), are preserved in a nanocrystalline environment. In particular, DBT:Ac NCs exhibit bright and photostable single-photon emission at room temperature that is spectrally stable and almost lifetime-limited (50 MHz) at cryogenic tem- peratures. The combination of such properties is unique and opens the way to the use of organic NCs for quantum technologies and for single-photon applications in general. Results and discussion We adapted a simple, cost-effective and well-established reprecipitation method48–51 to grow Ac NCs doped with controlled concentration of DBT molecules (for details see the Exper- imental Section). In this procedure, a dilute solution of the compounds prepared with a water-soluble solvent (acetone, in our case) is injected into sonicating water where it divides into many droplets. The solvent gradually dissolves and correspondingly the concentration in the micro-doplets becomes super-saturated until the compounds, which instead are not water-soluble, are reprecipitated in the form of NCs. The size and shape of the resulting NCs can ideally be controlled by varying the thermodynamic conditions.52 3 For morphological and optical characterization, a drop of the suspension of NCs in water is deposited on a coverglass substrate and dried in desiccator. Typical scanning electron microscopy (SEM) and atomic force microscopy (AFM) images are displayed in Figures 1a and 1b, respectively. For some NCs it is possible to identify peculiar features of crystalline Ac - such as the hexagonal-like morphology - while others exhibit a round-like shape, possibly due to a few nm acetone-rich solvent cage. By analysing the AFM images of 92 NCs (Figure 1d), we deduce an average equivalent diameter of (113± 64) nm and an average thickness of (65 ± 13) nm, compatible with a platelet-like shape. Such values and shape are particularly promising for the coupling to evanescent fields in proximity to surfaces.35 The crystalline Figure 1: Morphology of DBT:Ac NCs grown via reprecipitation: Typical SEM (a) and AFM topography (b) images. (c) Cross section showing the NC thickness profile along the white line in panel (b). (d) Statistical analysis on AFM images yielding a NC equivalent diameter and standard deviation of (113 ± 64)nm and an average thickness of (65 ± 13)nm. (e) Normalized XRD pattern in which only the peaks from the (001) plane and higher order reflections are well resolved, due to the NCs' platelet-like morphology (with c-axis perpendicular to the substrate). 4 nature suggested by the clear-cut edges and flat surfaces (see Figure 1c) is verified by X-ray diffraction (XRD) measurements. The XRD pattern shown in Figure 1e exhibits a strong diffracted peak at 9.17◦ - that corresponds to the (001) plane - and other equivalent periodic peaks corresponding to the (002), (003), and (004) planes, matching the crystallographic data for an anthracene monoclinic system.53 This also reveals that the Ac NCs, once deposited on the substrate, are mainly iso-oriented with the c-axis perpendicular to the substrate. Let us note that the transition dipole moment of a DBT molecule in the main insertion site of an Ac crystals is mostly oriented along the b-axis,54 and thus results parallel to the substrate. The sub-micrometric size of the crystalline matrix may compromise the optical properties of DBT molecules embedded therein, due to strain within the crystal and imperfections at the interfaces. Indeed, besides the case of quantum dots, it was reported for other nanocrystalline systems that such effects determine fluorescence instability and linewidth broadening.43,55 We thus perform single molecules microscopy and spectroscopy on DBT:Ac NCs with a home- built epifluorescence scanning confocal microscope - described in details in the Experimental Section - that allows for both room and cryogenic temperature investigation. At room temperature (RT), the sample is illuminated with a 767 nm continuous wave (CW) diode laser to pump DBT molecules into the vibrational band of the first singlet electronic excited state (see the simplified Jablonski diagram in Figure 2a). After a fast (ps- timescale) non radiative relaxation process to the lowest level of the vibrational manifold, molecules decay to the electronic ground state (singlet). The resulting red-shifted fluo- rescence light around 785 nm is detected with an electron multiplied charge couple device (EMCCD). Typical white light and wide-field fluorescence images are compared in Figure 2b, showing that more than 90% of the Ac NCs are successfully doped with DBT. To prove that the detected fluorescence stems from individual DBT molecules, single isolated crystals are illuminated in confocal mode with an excitation intensity of 15 kW cm−2 (well below saturation) and the correlation between photon arrival times is measured with the Hanbury Brown-Twiss (HBT) setup (see the Experimental Section). Figure 2c shows the histogram of 5 Figure 2: Photophysics of the NCs at RT:(a) Off-resonant pumping scheme employed for single molecule microscopy at RT. (b) Comparison between white light and fluorescence wide-field EMCCD images of the same region demonstrating that about 90% of the Ac NCs are successfully doped with DBT. (c) Measured photon anti-bunching (g2(0) = 0.05 from the fit, red solid line) from the emission of a single NC without any background correction (black dots). (d) Time-resolved measurement of the fluorescence decay from a single DBT:Ac NC (black line). A single exponential fit (red line) yields an excited state lifetime of (4.2±0.1)ns, as a free fitting parameter with the relative standard error. The inset shows the distribution of the excited-state lifetime collected from 40 NCs. The red curve is a Gaussian fit centered at 4.2 ns with FWHM = 0.4 ns. (e) Saturation measurement performed on a single DBT:Ac NC (black dots). The fitted red curve yields a saturation intensity of Is = (80 ± 6) kW cm−2 and a maximum detected count rate R∞ = (1.50 ± 0.02) Mcps. In the inset, a typical BFP image of a single NC. the observed coincidences from a single NC, featuring a strong antibunching dip. The exper- imental data is fitted at short time delays τ with the function g2(τ ) = 1 − b · exp(−τ/∆t), where ∆t accounts for the excitation and spontaneous emission rates47 and b, the dip depth, is found to be (95 ± 1)%. Among 40 analyzed NCs, 73% of them displays an antibunching 6 dip larger than 50%, demonstrating that the proposed recipe is reliable to grow individual Ac NCs in 2/3 of cases doped with single DBT molecules. The purity of this system, i.e. the second-order correlation function at zero time delay g2(0), can be as low as 0.05 ± 0.01 without any background correction. To gain further information on the emitter properties, we study the relaxation dynam- ics by means of time-correlated single-photon counting (TCSPC) measurements, collecting photons emitted after Ti:Sa-pulsed excitation (average intensity equal to 20 kW cm−2) with a single-photon avalanche diode (SPAD). Figure 2d shows a typical measured fluorescence decay curve from which the lifetime τf of the excited state can be derived via a single ex- ponential fit in the presence of a constant background. The fit (red curve) yields an excited state lifetime of (4.2 ± 0.1)ns. Repeating the measurement on 40 NCs we obtain the dis- tribution for the excited-state lifetime shown in the inset of Figure 2d, which can be fitted with a Gaussian centered at 4.2 ns with full width at half maximum (FWHM) of 0.4 ns, in agreement with previous studies on the bulk system.45,56,57 The brightness of the NC-based single-photon source is quantified by studying the sat- uration behavior of the system, non-resonantly pumped with the 767 nm-CW laser. Mea- surements are performed at different excitation intensities, scanning the sample under the confocal laser spot in the small region where the NC is located and detecting the red-shifted fluorescence with a single SPAD. From the obtained fluorescence maps the mean value within an area around the brightest pixel is extracted and corrected for the background counts, which is the mean value within an area out of the NC and is linear with the laser power. Data are plotted as a function of the laser intensity I (black dots in Figure 2e) and fitted with the function describing the saturation of the photon detection rate R(I):58 R(I) = R∞ I I + Is (1) with Is the saturation intensity and R∞ the maximum detected count rate. For the molecule 7 reported in Figure 2e the fit-procedure yields as free fitting parameters with the relative standard errors Is = (80±6) kW cm−2 and a maximum detected count rate R∞ = (1.50±0.02) Mcps. These can be considered as typical values. Accounting for the quantum efficiency of the SPAD, ηdet = 50%, the measured R∞ corresponds to a collected photon rate of 3 MHz at the detector. Moreover, comparing this value with the theoretical one of 240 MHz related to the measured lifetime through the relation R∞ (cid:39) (1/τf ) and assuming unitary quantum yield, we estimate a total collection efficiency of our setup at RT to be around 1%, ascribed to the limited numerical aperture of the optics and their transmission combined with the molecule emission profile.34 In order to determine the alignment of DBT molecules within the Ac NCs, the emission from single molecules is detected by imaging the objective back focal plane (BFP) from which the angular radiation pattern can be deduced.59 A typical BFP image is shown in the inset of Figure 2e, where the emission pattern features two side lobes facing each other beyond the critical angle, corresponding to the coupling between the evanescent wave in air with the propagative wave in the coverglass. The geometry and the direction of the two lobes confirms a horizontally aligned molecule, compatible with the XRD observations. To conclude on the observed photophysical properties of the DBT:Ac NCs at RT, let us note that the repeated excitation of the same molecule to study its saturation behavior is a qualitative proof of the stability of its fluorescence. After several hours of measurements at RT, though, some molecules start exhibiting fluorescence blinking behavior, typically before they stop to fluoresce completely. This so-called photobleaching is most probably due to chemical reactions of the dye molecule with ambient oxygen,60 a process that is more likely to occur in conjunction with the sublimation of Ac at RT. For sub-micrometric crystals we observe that sublimation at RT takes place on a time-scale of about one day but it is completely suppressed when covering the sample with a thin layer of a water-soluble polymer, such as poly(vinyl alcohol). At cryogenic temperatures, highly doped DBT:Ac NCs are studied under resonant ex- 8 Figure 3: Photophysics of the NCs at 2.9 K: (a) Resonant excitation spectrum of a single DBT molecule in an Ac NC. The pumping scheme is sketched in the top-inset. Data (black circles) are fitted with a Lorentzian profile (red curve) and an average over four consecutive measurements yields a FWHM = (51 ± 10) MHz. The bottom-inset shows the linewidth distribution of 35 molecules. (b) 2D plot of the excitation spectrum in time, in a frequency range where two molecules within the same NC are excited. The difference of the two peak central frequencies is plotted as white circles in the map, while the relative distributions are shown as histograms in the top panel. (c) Saturation curve (red circles) and power broadening (blue circles) of the ZPL are displayed with the theoretical fits (solid lines), yielding a maximum number of detected photons R∞ = (16.8 ± 0.4) kcps. (d)-top Excitation spectrum collected from a single NC within a frequency range of 800 GHz around 784.6 nm. (d)-bottom Inhomogeneous distribution of the ZPLs collected from 20 NCs. citation of the so-called 00-Zero Phonon Line (ZPL) (S0,ν0(cid:105) → S1,ν0(cid:105)). In this pumping scheme, sketched as an inset of Figure 3a, single DBT molecules can be addressed spectrally one at a time by tuning the frequency of a narrow-band laser and exploiting the inhomoge- neous distribution of the molecular resonances. In fact, depending on the host matrix, the ZPLs of PAH molecules can be distributed over a frequency range that can be smaller than 1 GHz in unstressed sublimated crystals and as high as 10 THz in polymers or amorphous materials.61–63 We found that a DBT concentration about six orders of magnitude higher 9 than the one proposed for RT characterization allows to spectrally select single molecules within our experimental full range of about 800 GHz around 784.6 nm. This will be discussed further on in the manuscript. Figure 3a shows the excitation spectrum of a single DBT molecule illuminated in confocal mode at 0.3 W cm−2 (below saturation), recording the red-shifted fluorescence as a function of the laser frequency (black circles). The spectral line is fitted with a Lorentzian profile (red curve) yielding a FWHM of 51 MHz, with an uncertainty of 10 MHz given by the standard deviation of four consecutive measurements of the same spectrum. Repeating this procedure on 35 molecules in different NCs leads to the distribution displayed in the inset of Figure 3a, with a low-width cutoff consistent with the lifetime-limited value of 40 MHz. The presence of molecules with broader linewidth can be explained in terms of the reduced size of the NC, which provides a less homogeneous environment for DBT molecules than that of bulk Ac. Moreover, interface effects on fluorescence stability and linewidth broadening are more likely to occur. However, let us note that the observed linewidth distribution is narrower than that measured by Gmeiner et al.,64 confirming the high crystallinity of the Ac NC grown via reprecipitation. Spectral diffusion has so far hindered the deployment of traditional (inorganic) NCs for narrow-band applications, such as single-photon sources for quantum technologies. We hence carefully analyze the spectral stability of the molecule transition frequency. In Figure 3b, the NC fluorescence counts detected from a SPAD are displayed in a 2D color map, obtained repeatedly scanning for 1 hour the excitation frequency of the pump laser over 10 GHz. The excitation of two different molecules can be recognized. The mean values and standard deviations for the two molecule ZPL frequencies over all measurements are (65 ± 6)MHz for peak 1 and (59 ± 4)MHz for peak 2. However, the common-mode fluctuation of the peak central frequencies is a clear indication of the non-negligible contribution given by the pump laser instability (the laser diode is thermally stabilized but not referred to any absolute frequency standard). To get rid of this contribution and highlight possible spectral 10 diffusion we analyze the distribution of the the two peak central frequencies difference (see the top panel in Figure 3b), plotted as white circles in the map. The maximum variation of such differential value is 17 MHz, which is well within the molecule linewidth and suggests negligible spectral diffusion for DBT:Ac NCs at 3 K. The same analysis has been carried out on couples of molecules in 8 different NCs. We observed sizable fluctuations only in one case where the ZPL central frequency over time exhibited a standard deviation of 54 MHz. As a term for comparison, we remind here that aromatic molecules in polymers or other amorphous hosts present linewidths as large as few GHz, accompanied by large spectral jumps (of the order of tens of GHz).63,65,66 Also, when molecules are embedded in a poor crystalline environment, a broadening of both linewidth and inhomogeneous distributions is observed, and spectral jumps, even if in a narrow frequency range of tens of MHz, are more likely to occur.64 Figure 3c shows in logarithmic scale a typical saturation profile of a single molecule and its line broadening at low temperatures, obtained by measuring the excitation spectrum for several pump powers and plotting the detected count rates at resonance (blue circles) and the FWHMs (red circles) as a function of the excitation intensity. Detected counts are fitted with equation 1, providing a saturation intensity Is = (0.73 ± 0.03)W cm−2 and a maximum number of detected photons R∞ = (16.8± 0.4)kcps (free fitting parameters with the relative standard errors), or equivalently R∞ = 33.6 kcps accounting for the detection efficiency ηdet. This count rate is compatible with the collection efficiency of our experimental setup for low temperature measurements of about 0.3 × 10−3 , mainly due to the the orientation of the emissive dipole and the low numerical aperture of the collecting optics. The power broadening of the homogeneous spectral line γhom(I) fits perfectly with the expected saturation law (blue line in Figure 3c) given by the equation:67 (cid:18) (cid:19)1/2 I Is γhom(I) = γhom(0) 1 + 11 (2) which assumes negligible spectral diffusion, as previously demonstrated. The inhomogeneous broadening of DBT molecules in Ac NCs is studied tuning the exci- tation laser over the available frequency range of about 800 GHz. In the top of Figure 3d a typical excitation spectrum collected at 2.9 K from a single NC is displayed, where we can distinguish about 80 peaks, each corresponding to a single molecule. The same measurement performed simultaneously on 20 NCs illuminated in wide-field for two orthogonal polariza- tion of the laser pump allows to estimate the inhomogeneous distribution of the ZPLs of DBT molecules in Ac NCs. The result of this analysis is plotted in the histogram on the bottom of Figure 3d. We deduce a mean value of the transition frequency equal to 785.1 nm with a standard deviation of 0.4 nm, which is in agreement with the inhomogeneous broadening measured for other dyes in crystalline systems.58 Finally, we observe that DBT:Ac NCs are ideal for the deterministic integratation into nanophotonic devices, opening new perspectives on the use of molecules in the development of real-world quantum technologies. Conclusions In this work we demonstrate organic nanocrystals doped with quantum emitters, performing as efficient, photostable and scalable single-photon sources, at both room and cryogenic tem- peratures. In particular DBT:Ac crystals are presented, with an average size of few-hundreds nanometer. The growth procedure is based on reprecipitation, an inexpensive method that is adapted for a precise tuning of DBT concentration. Atomic force microscopy shows that the crystals grown under our experimental conditions present an average thickness of about 60 nm and an average size of 100 nm. The reported values can be controlled and reduced by varying the reprecipitation conditions, such as water temperature, droplet size, injected so- lution concentration and addition of surfactants. X-ray diffraction confirms the crystallinity of the nanoparticles and their platelet-like morphology. At room temperature, single DBT molecules in NCs show a maximum detected count rate of 1.5 MHz, a multi-photon prob- 12 ability lower than 5% and a well defined dipole orientation. At 2.9 K, the vast majority of molecules exhibits linewidths close to the lifetime-limited value and a relative narrow inhomogeneous distribution of 180 GHz around 785 nm. Accurate investigation on their pho- tostability demonstrates that each NC embeds several molecules with stable fluorescence lines, with no signs of blinking or spectral diffusion on time scales of hours. These results may be extended to different molecular host-guest systems, functionalization protocols and purposes, making active organic nanocrystals a new toolbox for the integration of quantum emitters in photonic and optoelectronic circuits, as well as in complex hybrid devices. Experimental Section DBT:Ac NCs growth protocol. The DBT:Ac NCs growth procedure consists in inject- ing 250 µL of a mixture 1 : 106 of 1mM DBT-toluene and 5mM Ac-acetone solutions into 5 mL water. While continuously sonicating the system for 30 min, solvents dissolved in water and DBT:Ac crystals are formed in aqueous suspension. Solvents and Ac are purchased from Sigma Aldrich, water is deionized by a Milli-Q Advantage A10 System (18.2 mΩ cm at 25 ◦C) and DBT is purchased from Mercachem. Morphological characterization. Crystals size is evaluated by scanning electron mi- croscopy (SEM, Phenom Pro, PhenomWorld) and atomic force microscopy (AFM, Pico SPM from Molecular Imaging in AC mode equipped with a silicon probe NSG01 (NT-MDT) with 210 kHz resonant frequency). XRD measurements were performed at CRIST, the Crystal- lographic Centre of the University of Florence (Italy), with a XRD Bruker New D8 on a sample made of few µL of suspension desiccated on a silicon-low background sample holder (Bruker AXS). Optical setup. The optical characterization of DBT molecules within the sub-um Ac crystalline matrix was performed with a versatile home-built scanning fluorescence confo- cal microscope. The setup is equipped with a closed cycle Helium cryostat (Cryostation 13 by Montana Instruments), capable of cooling samples down to 2.9 K. Molecules can be ex- cited at 767 nm either by a continuous wave laser (CW, Toptica DL110-DFB) and a pulsed Ti:Sapphire (200 fs pulse width, 81.2 MHz repetition rate) laser. Alternatively, at cryogenic temperature, resonant excitation is performed with a narrowband fiber-coupled CW laser (Toptica, LD-0785-0080-DFB-1) centered at 784.6 nm, whose frequency can be scanned con- tinuously over a range of 800 GHz. All laser sources are linearly polarized to allow optimal coupling to single DBT transition by means of a half-wave plate in the excitation path. The laser intensities reported in the main text are calculated from the power measured at the objective entrance divided by the area of the confocal spot measured on the bare substrate (in both cases larger than the diffraction limited spot). For low temperature measurements, the excitation light is focused onto the sample by a long working distance air objective (Mi- tutoyo 100× Plan Apochromat, NA = 0.7, WD = 6 mm) and can be scanned over the sample through a telecentric system and a dual axis galvo-mirror. For room temperature measure- ments, a high-NA oil immersion objective (Zeiss Plan Apochromat, 100×, NA= 1.4) is used to focus light on the sample which is mounted on a piezoelectric nanopositioner (NanoCube by Physik Instrumente). The Stokes-shifted fluorescence is collected by the same microscope objective used in excitation, separated from the excitation light through a dichroic mirror (Semrock FF776-Di01) and a longpass filter (Semrock RazorEdge 785RS-25) and detected by either an EM-CCD camera (Andor iXon 885, 1004×1002 pixels, pixel size 8 µm×8 µm) or two single-photon avalanche diodes (τ-SPAD-50 Single Photon Counting Modules by Pico- Quant). SPADs can be used independently or in a Hanbury Brown-Twiss (HBT) configura- tion, using a time-correlated single-photon counting (TCSPC) card (PicoHarp, PicoQuant). A converging lens can be inserted in the excitation path to switch between confocal and wide-field illumination while a converging lens is added in the detection path before the EMCCD camera to study the wave-vector distribution of the light emitted by single DBT molecules via BFP imaging. 14 Notes The authors declare no competing financial interest. Acknowledgement The authors would like to thank S. Ciattini, L. Chelazzi (CRIST) for helping with XRD measurements, M. Mamusa for dynamic light scattering experiments, F. Intonti for the mi- croinfiltration setup, D. S. Wiersma for access to clean room facilities, M. Bellini and C. Corsi for Ti:sapphire operation, K.G. Schädler and F.H.L. Koppens for helpful feedback on the NCs properties and useful discussions about integration in hybrid devices. This work benefited from the COST Action MP1403 (Nanoscale Quantum Optics). The authors acknowledge financial support from the Fondazione Cassa di Risparmio di Firenze (GRANCASSA) and MIUR program Q-Sec Ground Space Communications. References (1) Kovalenko, M. V. et al. Prospects of Nanoscience with Nanocrystals. ACS Nano 2015, 9, 1012–1057. (2) Zheng, X. T.; Ananthanarayanan, A.; Luo, K. Q.; Chen, P. Glowing graphene quantum dots and carbon dots: properties, syntheses, and biological applications. Small 2015, 11, 1620–1636. (3) Lyu, Y.; Xie, C.; Chechetka, S. A.; Miyako, E.; Pu, K. Semiconducting polymer nanobioconjugates for targeted photothermal activation of neurons. Journal of the American Chemical Society 2016, 138, 9049–9052. (4) Kim, T.-H.; Jun, S.; Cho, K.-S.; Choi, B. L.; Jang, E. Bright and stable quantum dots and their applications in full-color displays. MRS Bulletin 2013, 38, 712–720. 15 (5) Talapin, D. V.; Lee, J.-S.; Kovalenko, M. V.; Shevchenko, E. V. Prospects of Colloidal Nanocrystals for Electronic and Optoelectronic Applications. Chemical Reviews 2010, 110, 389–458. (6) O'Brien, J. L.; Furusawa, A.; Vučković, J. Photonic quantum technologies. Nature Photonics 2009, 3, 687–695. (7) Aharonovich, I.; Englund, D.; Toth, M. Solid-state single-photon emitters. Nature Photonics 2016, 10, 631–641. (8) Gatto Monticone, D.; Katamadze, K.; Traina, P.; Moreva, E.; Forneris, J.; Ruo- Berchera, I.; Olivero, P.; Degiovanni, I. P.; Brida, G.; Genovese, M. Beating the Abbe Diffraction Limit in Confocal Microscopy via Nonclassical Photon Statistics. Phys. Rev. Lett. 2014, 113, 143602. (9) Sangouard, N.; Zbinden, H. What are single photons good for? Journal of Modern Optics 2012, 59, 1458–1464. (10) Tillmann, M.; Tan, S.-H.; Stoeckl, S. E.; Sanders, B. C.; de Guise, H.; Heilmann, R.; Nolte, S.; Szameit, A.; Walther, P. Generalized Multiphoton Quantum Interference. Phys. Rev. X 2015, 5, 041015. (11) Knill, E.; Laflamme, R.; Milburn, G. J. A scheme for efficient quantum computation with linear optics. nature 2001, 409, 46–52. (12) QuantumManifesto, available at http://qurope.eu/manifesto, 2016. (13) Lounis, B.; Orrit, M. Single-photon sources. Reports On Progress In Physics 2005, 68, 1129–1179. (14) Chu, X.-L.; Götzinger, S.; Sandoghdar, V. A single molecule as a high-fidelity photon gun for producing intensity-squeezed light. Nat Photon 2017, 11, 58–62. 16 (15) Loredo, J. C.; Broome, M. A.; Hilaire, P.; Gazzano, O.; Sagnes, I.; Lemaitre, A.; Almeida, M. P.; Senellart, P.; White, A. G. Boson Sampling with Single-Photon Fock States from a Bright Solid-State Source. Physical Review Letters 2017, 118, 130503. (16) Sipahigil, A.; Jahnke, K. D.; Rogers, L. J.; Teraji, T.; Isoya, J.; Zibrov, A. S.; Jelezko, F.; Lukin, M. D. Indistinguishable photons from separated silicon-vacancy centers in dia- mond. Physical Review Letters 2014, 113, 113602. (17) Lettow, R.; Rezus, Y. L. A.; Renn, A.; Zumofen, G.; Ikonen, E.; Götzinger, S.; San- doghdar, V. Quantum Interference of Tunably Indistinguishable Photons from Remote Organic Molecules. Phys. Rev. Lett. 2010, 104, 123605. (18) Bermudez-Urena, E.; Gonzalez-Ballestero, C.; Geiselmann, M.; Marty, R.; Radko, I. P.; Holmgaard, T.; Alaverdyan, Y.; Moreno, E.; Garcia-Vidal, F. J.; Bozhevolnyi, S. I.; Quidant, R. Coupling of individual quantum emitters to channel plasmons. Nat Commun 2015, 6. (19) Schroeder, T.; Schell, A. W.; Kewes, G.; Aichele, T.; Benson, O. Fiber-Integrated Diamond-Based Single Photon Source. Nano Lett. 2011, 11, 198–202. (20) Liebermeister, L.; Petersen, F.; Münchow, A. v.; Burchardt, D.; Hermelbracht, J.; Tashima, T.; Schell, A. W.; Benson, O.; Meinhardt, T.; Krueger, A.; Stiebeiner, A.; Rauschenbeutel, A.; Weinfurter, H.; Weber, M. Tapered fiber coupling of single pho- tons emitted by a deterministically positioned single nitrogen vacancy center. Applied Physics Letters 2014, 104, 031101. (21) Riedrich-Möller, J.; Arend, C.; Pauly, C.; Mücklich, F.; Fischer, M.; Gsell, S.; Schreck, M.; Becher, C. Deterministic Coupling of a Single Silicon-Vacancy Color Cen- ter to a Photonic Crystal Cavity in Diamond. Nano Letters 2014, 14, 5281–5287. (22) Schell, A. W.; Kaschke, J.; Fischer, J.; Henze, R.; Wolters, J.; Wegener, M.; Benson, O. 17 Three-dimensional quantum photonic elements based on single nitrogen vacancy-centres in laser-written microstructures. Scientific reports 2013, 3. (23) Shi, Q.; Sontheimer, B.; Nikolay, N.; Schell, A. W.; Fischer, J.; Naber, A.; Benson, O.; Wegener, M. Wiring up pre-characterized single-photon emitters by laser lithography. Scientific Reports 2016, 6, 31135. (24) Somaschi, N.; Giesz, V.; De Santis, L.; Loredo, J.; Almeida, M. P.; Hornecker, G.; Portalupi, S. L.; Grange, T.; Antón, C.; Demory, J. Near-optimal single-photon sources in the solid state. Nature Photonics 2016, 10, 340–345. (25) Sapienza, L.; Davanço, M.; Badolato, A.; Srinivasan, K. Nanoscale optical positx- ioning of single quantum dots for bright and pure single-photon emission. Nature Communications 2015, 6, 7833. (26) Arcari, M.; Söllner, I.; Javadi, A.; Lindskov Hansen, S.; Mahmoodian, S.; Liu, J.; Thyrrestrup, H.; Lee, E.; Song, J.; Stobbe, S.; Lodahl, P. Near-Unity Coupling Effi- ciency of a Quantum Emitter to a Photonic Crystal Waveguide. Physical Review Letters 2014, 113, 093603. (27) Daveau, R. S.; Balram, K. C.; Pregnolato, T.; Liu, J.; Lee, E. H.; Song, J. D.; Verma, V.; Mirin, R.; Nam, S. W.; Midolo, L.; Stobbe, S.; Srinivasan, K.; Lodahl, P. Efficient fiber- coupled single-photon source based on quantum dots in a photonic-crystal waveguide. Optica 2017, 4, 178–184. (28) Zadeh, I. E.; Elshaari, A. W.; Jöns, K. D.; Fognini, A.; Dalacu, D.; Poole, P. J.; Reimer, M. E.; Zwiller, V. Deterministic Integration of Single Photon Sources in Silicon Based Photonic Circuits. Nano Letters 2016, 16, 2289–2294. (29) Davanço, M.; Liu, J.; Sapienza, L.; Zhang, C.-Z.; De Miranda Cardoso, J. V. c.; Verma, V.; Mirin, R.; Nam, S. W.; Liu, L.; Srinivasan, K. Heterogeneous integra- 18 tion for on-chip quantum photonic circuits with single quantum dot devices. Nature Communications 2017, 8, 889. (30) Hausmann, B. J. M.; Shields, B.; Quan, Q.; Maletinsky, P.; McCutcheon, M.; Choy, J. T.; Babinec, T. M.; Kubanek, A.; Yacoby, A.; Lukin, M. D.; Lončar, M. Integrated Diamond Networks for Quantum Nanophotonics. Nano Letters 2012, 12, 1578–1582. (31) Mouradian, S. L.; Schröder, T.; Poitras, C. B.; Li, L.; Goldstein, J.; Chen, E. H.; Walsh, M.; Cardenas, J.; Markham, M. L.; Twitchen, D. J.; Lipson, M.; Englund, D. Scalable Integration of Long-Lived Quantum Memories into a Photonic Circuit. Phys. Rev. X 2015, 5, 031009. (32) Türschmann, P.; Rotenberg, N.; Renger, J.; Harder, I.; Lohse, O.; Utikal, T.; Götzinger, S.; Sandoghdar, V. Chip-Based All-Optical Control of Single Molecules Co- herently Coupled to a Nanoguide. Nano Letters 2017, 17, 4941–4945. (33) Lombardi, P.; Ovvyan, A.; Pazzagli, S.; Mazzamuto, G.; Kewes, G.; Neitzke, O.; Gruh- ler, N.; Benson, O.; Pernice, W.; Cataliotti, F. Photostable molecules on chip: inte- grated single photon sources for quantum technologies. ACS Photonics 2017, (34) Checcucci, S.; Lombardi, P.; Rizvi, S.; Sgrignuoli, F.; Gruhler, N.; Dieleman, F. B.; Cataliotti, F. S.; Pernice, W. H.; Agio, M.; Toninelli, C. Beaming light from a quantum emitter with a planar optical antenna. Light: Science & Applications 2016, 6, e16245. (35) Skoff, S. M.; Papencordt, D.; Schauffert, H.; Bayer, B. C.; Rauschenbeutel, A. An optical nanofiber-based interface for single molecules. arXiv preprint arXiv:1604.04259 2016, (36) Pisanello, F.; Leménager, G.; Martiradonna, L.; Carbone, L.; Vezzoli, S.; Desfonds, P.; Cozzoli, P. D.; Hermier, J.-P.; Giacobino, E.; Cingolani, R.; De Vittorio, M.; Bra- 19 mati, A. Non-Blinking Single-Photon Generation with Anisotropic Colloidal Nanocrys- tals: Towards Room-Temperature, Efficient, Colloidal Quantum Sources. Adv. Mater. 2013, 25, 1974–1980. (37) Liu, J.; Konthasinghe, K.; Davanço, M.; Lawall, J.; Anant, V.; Verma, V.; Mirin, R.; Nam, S. W.; Song, J. D.; Ma, B. Direct observation of nanofabrication influence on the optical properties of single self-assembled InAs/GaAs quantum dots. arXiv preprint arXiv:1710.09667 2017, (38) Empedocles, S. A.; Bawendi, M. G. Quantum-confined stark effect in single CdSe nanocrystallite quantum dots. Science 1997, 278, 2114–2117. (39) Efros, A. L.; Nesbitt, D. J. Origin and control of blinking in quantum dots. Nat Nano 2016, 11, 661–671. (40) Chandrasekaran, V.; Tessier, M. D.; Dupont, D.; Geiregat, P.; Hens, Z.; Brainis, E. Nearly Blinking-Free, High-Purity Single-Photon Emission by Colloidal InP/ZnSe Quantum Dots. Nano Lett. 2017, –. (41) Park, Y.-S.; Guo, S.; Makarov, N. S.; Klimov, V. I. Room Temperature Single-Photon Emission from Individual Perovskite Quantum Dots. ACS Nano 2015, 9, 10386–10393. (42) Raino, G.; Nedelcu, G.; Protesescu, L.; Bodnarchuk, M. I.; Kovalenko, M. V.; Mahrt, R. F.; Stöferle, T. Single Cesium Lead Halide Perovskite Nanocrystals at Low Temperature: Fast Single-Photon Emission, Reduced Blinking, and Exciton Fine Struc- ture. ACS Nano 2016, 10, 2485–2490. (43) Jantzen, U.; Kurz, A. B.; Rudnicki, D. S.; Schäfermeier, C.; Jahnke, K. D.; Ander- sen, U. L.; Davydov, V. A.; Agafonov, V. N.; Kubanek, A.; Rogers, L. J.; Jelezko, F. Nanodiamonds carrying silicon-vacancy quantum emitters with almost lifetime-limited linewidths. New Journal of Physics 2016, 18, 073036. 20 (44) Tran, T. T.; Kianinia, M.; Bray, K.; Kim, S.; Xu, Z.-Q.; Gentle, A.; Sontheimer, B.; Bradac, C.; Aharanovich, I. Nanodiamonds with photostable, sub-gigahertz linewidths quantum emitters. arXiv preprint arXiv:1705.06810 2017, (45) Toninelli, C.; Early, K.; Bremi, J.; Renn, A.; Sandoghdar, V. Near-infrared single- photons from aligned molecules in ultrathin crystalline films at room temperature. Optics Express 2010, 18, 6577–6582. (46) Nicolet, A. A.; Hofmann, C.; Kol'chenko, M. A.; Kozankiewicz, B.; Orrit, M. Single dibenzoterrylene molecules in an anthracene crystal: Spectroscopy and photophysics. ChemPhysChem 2007, 8, 1215–1220. (47) Trebbia, J.-B.; Ruf, H.; Tamarat, P.; Lounis, B. Efficient generation of near infra-red single photons from the zero-phonon line of a single molecule. Optics Express 2009, 17, 23986–23991. (48) Horn, D.; Rieger, J. Organic nanoparticles in the aqueous phase - Theory, experiment, and use. Angewandte Chemie International Edition 2001, 40, 4330–4361. (49) Kasai, H.; Nalwa, H. S.; Oikawa, H.; Okada, S.; Matsuda, H.; Minami, N.; Kakuta, A.; Ono, K.; Mukoh, A.; Nakanishi, H. A novel preparation method of organic microcrys- tals. Japanese Journal of Applied Physics 1992, 31, L1132. (50) Kang, P.; Chen, C.; Hao, L.; Zhu, C.; Hu, Y.; Chen, Z. A novel sonication route to prepare anthracene nanoparticles. Materials Research Bulletin 2004, 39, 545–551. (51) Baba, K.; Kasai, H.; Nishida, K.; Nakanishi, H. Nanocrystal; InTech, 2011. (52) Chung, H.-R.; Kwon, E.; Oikawa, H.; Kasai, H.; Nakanishi, H. Effect of solvent on organic nanocrystal growth using the reprecipitation method. Journal of crystal growth 2006, 294, 459–463. 21 (53) Brock, C. P.; Dunitz, J. Temperature dependence of thermal motion in crystalline anthracene. Acta Crystallographica Section B: Structural Science 1990, 46, 795–806. (54) Nicolet, A. A.; Hofmann, C.; Kol'chenko, M. A.; Kozankiewicz, B.; Orrit, M. Single dibenzoterrylene molecules in an anthracene crystal: Main insertion sites. ChemPhysChem 2007, 8, 1929–1936. (55) Meltzer, R.; Yen, W.; Zheng, H.; Feofilov, S.; Dejneka, M.; Tissue, B.; Yuan, H. Ev- idence for long-range interactions between rare-earth impurity ions in nanocrystals embedded in amorphous matrices with the two-level systems of the matrix. Physical Review B 2001, 64, 100201. (56) Mazzamuto, G.; Tabani, A.; Pazzagli, S.; Rizvi, S.; Reserbat-Plantey, A.; Schädler, K.; Navickaite, G.; Gaudreau, L.; Cataliotti, F.; Koppens, F. Single-molecule study for a graphene-based nano-position sensor. New Journal of Physics 2014, 16, 113007. (57) Polisseni, C.; Major, K. D.; Boissier, S.; Grandi, S.; Clark, A. S.; Hinds, E. Stable, single-photon emitter in a thin organic crystal for application to quantum-photonic devices. Optics Express 2016, 24, 5615–5627. (58) Moerner, W. E.; Fromm, D. P. Methods of single-molecule fluorescence spectroscopy and microscopy. Review of Scientific Instruments 2003, 74, 3597–3619. (59) Lieb, M. A.; Zavislan, J. M.; Novotny, L. Single-molecule orientations determined by direct emission pattern imaging. JOSA B 2004, 21, 1210–1215. (60) Kozankiewicz, B.; Orrit, M. Single-molecule photophysics, from cryogenic to ambient conditions. Chem. Soc. Rev. 2014, 43, 1029–1043. (61) Veerman, J.; Garcia-Parajo, M.; Kuipers, L.; Van Hulst, N. Single molecule mapping of the optical field distribution of probes for near-field microscopy. Journal of Microscopy 1999, 194, 477–482. 22 (62) Kramer, A.; Segura, J.-M.; Hunkeler, A.; Renn, A.; Hecht, B. A cryogenic scanning near-field optical microscope with shear-force gapwidth control. Review of scientific instruments 2002, 73, 2937–2941. (63) Kozankiewicz, B.; Bernard, J.; Orrit, M. Single molecule lines and spectral hole burning of terrylene in different matrices. The Journal of chemical physics 1994, 101, 9377–9383. (64) Gmeiner, B.; Maser, A.; Utikal, T.; Götzinger, S.; Sandoghdar, V. Spectroscopy and mi- croscopy of single molecules in nanoscopic channels: spectral behavior vs. confinement depth. Physical Chemistry Chemical Physics 2016, 18, 19588–19594. (65) Walser, A.; Zumofen, G.; Renn, A.; Götzinger, S.; Sandoghdar, V. Spectral dynamics and spatial localization of single molecules in a polymer. Molecular Physics 2009, 107, 1897–1909. (66) Boiron, A.-M.; Tamarat, P.; Lounis, B.; Brown, R.; Orrit, M. Are the spectral trails of single molecules consistent with the standard two-level system model of glasses at low temperatures? Chemical physics 1999, 247, 119–132. (67) Ambrose, W.; Basché, T.; Moerner, W. Detection and spectroscopy of single pentacene molecules in ap-terphenyl crystal by means of fluorescence excitation. The Journal of Chemical Physics 1991, 95, 7150–7163. 23 Graphical TOC Entry 24
1911.12644	1	1911	2019-11-28T10:59:59	Low-Voltage Printed, All-Polymer Integrated Circuits Employing a Low-leakage and High-Yield Polymer Dielectric	[ "physics.app-ph" ]	In the path toward the integration of organic field effect transistors (OFETs) and logic circuits into low-cost and mass produced consumer products, all-organic devices based on printed semiconductors are one of the best options to meet stringent costs requirements. Within this framework, it is still challenging to achieve low voltage operation, as required by the use of thin film batteries and energy harvesters, for which a high capacitance and reliable organic dielectric is required. Here, the development of a parylene-C based dielectric bilayer compatible with top-gate architectures for low-voltage OFETs and logic circuits is presented. The polymer bilayer dielectric allowed the high yield fabrication of all-polymer, bendable, transparent p- and n-type OFETs operating below 2 V, with low leakage, uniform performances and high yield. Such result is a key enabler for the reliable realization of complementary logic circuits that can operate already at a voltage bias of 2 V, such as well-balanced inverters, ring-oscillators and D-Flip-Flops.	physics.app-ph	physics	DOI: 10.1002/ ((please add manuscript number)) Article type: Full Paper Low-Voltage Printed, All-Polymer Integrated Circuits Employing a Low-leakage and High- Yield Polymer Dielectric Elena Stucchi, Giorgio Dell'Erba, Paolo Colpani, Yun-Hi Kim and Mario Caironi* E. Stucchi, Dr. G. Dell'Erba, P. Colpani, Dr. M. Caironi Center for Nano Science and Technology @PoliMi, Istituto Italiano di Tecnologia, via Pascoli 70/3, 20133 Milano, Italy E-mail: [email protected] E. Stucchi Dipartimento di Fisica, Politecnico di Milano, P.za Leonardo da Vinci 32, 20133 Milano, Italy Prof. Y.-H. Kim Department of Chemistry and Research Institute of Green Energy Convergence Technology (RIGET), Gyeongsang National University, Jinju 660-701, Republic of Korea Keywords: organic field-effect transistors, printed integrated circuits, parylene, inkjet printing, flexible electronics In the path toward the integration of organic field effect transistors (OFETs) and logic circuits into low-cost and mass produced consumer products, all-organic devices based on printed semiconductors are one of the best options to meet stringent costs requirements. Within this framework, it is still challenging to achieve low voltage operation, as required by the use of thin film batteries and energy harvesters, for which a high capacitance and reliable organic dielectric is required. Here, the development of a parylene-C based dielectric bilayer compatible with top-gate architectures for low- voltage OFETs and logic circuits is presented. The polymer bilayer dielectric allowed the high yield fabrication of all-polymer, bendable, transparent p- and n-type OFETs operating below 2 V, with low 1 leakage, uniform performances and high yield. Such result is a key enabler for the reliable realization of complementary logic circuits that can operate already at a voltage bias of 2 V, such as well- balanced inverters, ring-oscillators and D-Flip-Flops. 1. Introduction Organic field effect transistors (OFETs) have been extensively studied in the last decades, and are currently considered as the building block of the next generation of thin film electronics.[1-3] This technology allows for the development of large area, inexpensive, transparent electronic devices by means of cost-effective manufacturing techniques. Organic materials can be processed from solution, mostly by means of printing and coating techniques, which are low-temperature processes, enable mass production and minimize the by-products.[4] Being these manufacturing techniques compatible with flexible, plastic and other low-cost substrates, there is a clear potential for the integration of additional electronic functionalities into mass-produced, consumer products.[5-7] In order to meet very stringent costs constraints, which do not allow integration of conventional electronic chips, all- organic circuits fabricated by means of scalable techniques are one of the best options. One of the main limitations hindering the adoption of all-organic printed circuits in real application is related to their operating voltage.[8] In order to grant their portability and easy integration, these circuits need in fact to be powered by thin film batteries[9, 10] and/or energy harvesters, such as plastic solar cells,[11, 12] thus requiring maximum operation voltages of a few volts and low power consumption, while keeping reasonably high values of accumulated charge density and of current flowing into the circuits. Efficient low voltage operation can be achieved by acting on the capacitance of the dielectric layer, which should be as high as possible and at the same time guarantee optimal charge accumulation and transport at the semiconductor-dielectric interface of both holes and electrons, in order to enable complementary architectures to drastically reduce power consumption.[13, 14] Many efforts have been recently devoted to the development of suitable polymer materials for gate dielectric applications, aiming at the achievement of the highest possible gate capacitance.[13] Two 2 main strategies may be followed, either increasing the dielectric constant of the employed material or decreasing its thickness. The integration of high-k materials as dielectrics is not straightforward, as the energetic disorder at the interface might interfere with charge transport inside the semiconductor layer.[15, 16] Multilayer structures combining low-k and high-k materials have been therefore introduced.[17, 18] However, high-k materials typically show dielectric relaxations occurring at low frequency,[19, 20] possibly limiting the maximum operation frequency of OFETs. Such limit is particularly severe in electrolyte gated transistors,[21] where huge capacitances are achieved at the expense of the switching speed because of ions motion, even in recent solid-state electrolytes.[22] To avoid such limitations, an obvious choice is to downscale a low-k dielectric polymer. However, this strategy typically leads to very high leakage currents or complete breakdown of devices. A possible solution to the latter is the adoption of poly(chloro-p-xylene)-C (Parylene-C), a low-k, semicrystalline, thermoplastic polymer, which is commonly deposited by means of chemical vapor deposition (CVD);[23] this is not a solution-based printing technique, but it has already been shown to be large- area compatible and industrially scalable, so it would not be a limiting factor during a possible upscale of the manufacturing. This dielectric grants a low energetic disorder at the semiconductor-dielectric interface[24] and presents some appealing characteristics, such as its chemical inertness, flexibility, and the ability to form conformal, pinhole-free coatings.[24, 25] For such reasons, parylene has been largely investigated as a dielectric for OFETs, although most of the reports deal with bottom-gated structures,[26-33] where the dielectric is deposited on an inert surface and not on the active material. Staggered, top-gate structures are however very important towards the development of logic circuits as they allow for optimal charge injection,[34] thanks to the possibility of reducing the contact resistance through gate to source/drain electrodes overlap.[35] Moreover, they allow for controlled interfaces between the semiconductor and the dielectric layer,[36, 37] and for a partial self- encapsulation effect thanks to the upper gate dielectric and the gate electrode which improves environmental stability of devices.[38] In the context of OFETs based on a printed semiconductor, only one example of top-gated device employing parylene as dielectric has been reported, with 3 opaque evaporated metallic contacts and operating voltages in the order of 30 V, thus not compatible with low voltage operation.[39] Here we exploit a thin parylene layer for the development of a high capacitance polymer bilayer dielectric, comprising an ultra-thin poly(methyl methacrylate) (PMMA), compatible with top-gate OFETs. Our bilayer dielectric achieves areal capacitances in the order of 20 nF cm-2 with leakage currents below 1 nA cm-2. It enables the robust fabrication of low-voltage, transparent, fully- polymeric, and complementary OFETs, where all the layers are inkjet printed but the parylene layer, based on poly([N,N'-bis(2-octyldodecyl)-naphthalene-1,4,5,8-bis(dicarboximide)-2,6-diyl]-alt-5,5'- (2,2'-bitthiophene)) (P(NDI2OD-T2) and poly[2,5-bis(7-decylnonadecyl)pyrrolo[3,4-c]pyrrole- 1,4(2H,5H)-dione-(E)-1,2-di(2,2′-bithiophen-5-yl)ethene] (29-DPP-TVT) as n- and p-type semiconductors, respectively. The ideal dielectric characteristics allow the transistors to operate with sub-10 V voltages and low leakage and to be fabricated with a 99 % yield and good uniformity, as probed on a 100 transistors array for each OFET polarity. The reliable process was key to enable the integration of OFETs into logic circuits of different level of complexity operating down to 2 V, from well-balanced complementary inverters to 7-stages ring oscillators with stage delays as low as 1.14 ms, and D-Flip-Flops, fundamental building blocks of registers, counters and timers. Our results, exploiting scalable deposition processes, therefore offer a viable path for the cost-effective integration of electronic functionalities into consumer products. 2. Results and discussion For the realization of high capacitance all-polymer dielectrics, we opted for a bilayer dielectric combining a thin parylene film on top of an ultra-thin solution-processed layer of poly(methyl methacrylate) (PMMA), thus combining the excellent dielectric properties of parylene[40] and the optimal interface formed by PMMA with a broad range of polymer semiconductors. The PMMA 4 layer is in fact important to achieve optimal n-type devices (Figure S1), as it protects the electron transporting semiconductor from the degrading effect of the chlorine atoms present in the parylene.[39] The overall dielectric properties of the bilayer dielectric have been investigated in order to find the best combination to be integrated into printed OFETs and circuits. For this analysis, a MIM (metal- insulator-metal) structure has been employed, as shown in Figure 1a, with the dielectric bilayer deposited between two metallic electrodes. The PMMA layer is 20 nm thick, while for the parylene layer we considered different thicknesses; for the analysis presented here, three different capacitors have been examined, with parylene layers that are 120, 200 and 310 nm thick. The capacitance as a function of frequency for the three above mentioned capacitors has been measured and is shown in Figure 1b. The areal capacitance at 1 kHz is 11.54 nF cm-2 for the 310 nm thick parylene layer, and it increases to 16.69 and 19.00 nF cm-2 for the 200 nm and 120 nm thick layers, respectively. The average leakage current at 10 V amounts to 10 nA cm-2 for the 120 nm thick parylene layer, while the 200 and 310 nm thick layers show average leakage currents of 4.6 and 2.25 nA cm-2, respectively (Figure 1c). Bilayers employing parylene layers thinner than ~120 nm, despite leading to an increase in capacitance, typically resulted in defective capacitors or short circuits. Following the capacitors characterization, we therefore decided to adopt bilayers with parylene films with thicknesses of about 100 nm for the printed transistors, so that a good compromise could be reached between low operating voltage, favored by the high capacitance value, and low leakages, together with a high yield and uniform performances. 5 Figure 1. (a) Schematic diagram of the MIM structure. (b) Capacitance as a function of the frequency and (c) leakage current density of the capacitors. All the values shown here have been obtained as average over 3 identical devices. The dielectric bilayer has been integrated into p- and n-type bottom-contact/top-gated OFETs (Figure 2a). These devices have been fabricated exclusively by means of additive processes, which are cost- effective and industrially scalable. First, poly(3,4-ethylenedioxithiophene):polystyrene sulphonate (PEDOT:PSS) source and drain contacts have been inkjet-printed (Figure 2b) onto 125 µm thick polyethylene naphthalate (PEN) substrates, defining a channel length (L) of 65 µm and width (W) of 1000 µm. Small pads of silver nanoparticles-based ink have been printed in correspondence of the points where electrical contacts are created during the characterization of transistors and circuits; it should be pointed out that these pads are only needed to facilitate external electrical probing of the devices, which can operate also without them, in a fully transparent configuration. Onto the contacts for the n-type transistors, a polyethylenimine (PEI)-based injection layer has been printed (Figure 2c), leading to enhanced electrons injection thanks to a reduced electrode work-function and interface doping.[41, 42] P(NDI2OD-T2) has been inkjet printed for n-type devices, while 29-DPP-TVT for the p-type ones (Figure 2d,e). The dielectric stack has been subsequently deposited on top of the semiconductor: the 20 nm thick PMMA layer has been spin-coated, while parylene has been deposited 6 by means of chemical vapor deposition (CVD), thus obtaining a dielectric stack with an overall thickness lower than 150 nm. Lastly, PEDOT:PSS gate electrodes have been inkjet-printed on top of the dielectric (Figure 2f). Figure 2. Scheme of the printed OFET (a) and optical micrographs of the fabrication steps: (b) PEDOT:PSS source and drain electrodes, together with the silver pads, inkjet printed on the PEN substrate, (c) PEI interlayer inkjet printed on the contacts of the n-type transistors, (d) and (e) inkjet printed P(NDI2OD-T2) and 29-DPP-TVT semiconductors for n- and p-type devices, respectively, (f) PEDOT:PSS gate electrode, inkjet printed after the deposition of the dielectric bilayer. The electrical characterization of the resulting OFETs is presented in Figure 3. As it can be seen from the transfer curves (Figure 3a,d), both p- and n-type transistors show proper current modulation at voltages as low as 1 V. Output curves (Figure 3b,e) are rather ideal in the case of p-type devices, while a residual S-shaped characteristic is visible for n-type devices, likely owing to contact effects, though not precluding a correct low voltage operation of the devices. The presented electrical data have been extracted from an array of 100 transistors for each polarity, in order to address both the performances and the uniformity of the printed OFETs. The average 7 transfer curves with their standard deviation, for an applied source-drain voltage VDS of 10 V, are shown in Figure 3c,f, while the raw data, both for linear and saturation regime, can be found in Figure S3. The average maximum current, which has been extracted from the 10 V transfer curves, amounts to 1.67 ± 0.29 A for the n-type devices, while for the p-type devices it amounts to 4.02 ± 0.28 A. This difference in the performance can be explained by considering the combined effect of their average field effect mobilities and of their threshold voltages. The average field effect mobility of electrons (µe) and holes (µh), over 100 devices in both cases, was extracted from the slope of the transfer characteristic curves in the VG range from 8 to 10 V, according to the gradual channel approximation.[43] The areal capacitance for the dielectric stack employed in these devices has been measured for frequencies down to 5 Hz (Figure S2c), where we recorded a capacitance equal to 22.6 nF cm-2; this value has been used for the mobility extraction. For n-type devices, µe = 0.08 cm2 V-1 s-1 in the linear regime (VDS = 2 V), and µe = 0.12 cm2 V-1 s-1 in the saturation regime (VDS = 10 V), while for p-type devices µh = 0.20 cm2 V-1 s-1 in the linear regime, and µh = 0.21 cm2 V-1 s-1 in the saturation one. In the latter case, the mobility values do not show substantial gate voltage dependence in full accumulation (Figure S4b). n-type OFETs, on the other hand, exhibit a slightly non-ideal behavior, given by a small gate dependence of the mobility curves (Figure S4a). We extracted the measurement reliability factors (r), as recommended by Choi et al.,[44] and obtained in both cases high values: close to 100 % for holes mobility in both linear (96 %) and saturation (95 %), and higher than 70 % for electrons mobility (81 % in linear and 73 % in saturation regime). As a result, average effective mobilities (µeff), which are more robust figure of merits to describe carriers field-effect mobility in thin-film transistors and are calculated as µh×r and µe×r, are very close to µh and µe: for p-type devices, µeff,lin = 0.19 cm2 V-1 s-1 and µeff,sat = 0.20 cm2 V-1 s-1, and for n-type devices µeff,lin = 0.06 cm2 V-1 s-1 and µeff,sat = 0.08 cm2 V-1 s-1. The lower r for n-type OFETs is due to the combined effect of a weak gate voltage dependence and a higher threshold voltage: P(NDI2OD-T2) devices present a threshold voltage of 1.35 ± 0.41 V and a turn-on voltage of 1 ± 0.22 V, while 29-DPP-TVT transistors show lower current onset values, with a threshold voltage of 8 0.49 ± 0.21 V and 0.68 ± 0.3 V as turn-on voltage. The subthreshold slope (SS) is 148 ± 23 mV dec- 1 and 126 ± 25 mV dec-1 for p- and n-type OFETs, respectively. On-Off ratios, defined as the maximum ratio of the drain current values in the "on" and "off" states, are relatively high, with average values of 5x105 for the p-type transistors and of 7x106 for the n-type ones. The transfer curves for devices of both polarities present a small hysteresis. For what concerns p-type devices, during the backward sweep, the threshold voltage slightly shifts toward positive voltage and the subthreshold slope (SS) is increased. The n-type OFETs, on the other hand, present a shift toward negative voltages and the same increase in SS. This feature can be explained with the presence of shallow traps at the dielectric interface, which are filled during the forward sweep and remain so during the backward one, which thus appears to be slightly more ideal. Being this hysteresis limited in size and reproducible, it does not affect the performances of our devices in a significant way and can be considered to be negligible. Overall, of the 200 transistors fabricated for this electrical analysis, only two were not working, leading to a 99% yield. From a simple inspection with the optical microscope of the defective devices, we found that failure in the two devices was caused by particulate that landed onto the active channel (Figure S5a,b). It is therefore reasonable to expect that in a more controlled environment, such as a classified cleanroom, the yield may approach 100%. Gate leakage currents of our transistors are very low, with a flat curve and leakage current values below 100 pA (Figure 3a,d). More than 90% of the transistors are characterized by such low leakage, as shown in the raw data in Figure S3, while only about 5 devices for each polarity present an increased leakage current, while still working properly, with current and mobility values comparable with the ones of the very low leakage devices. For what concerns the transistor arrays presented here, the presence of the leakage current is to be related in most of the cases to the accidental presence of small droplets of the silver nanoparticles ink into the transistors channel, as shown in Figure S5c. Silver pads are needed uniquely for the characterization of transistors and circuits with external probes, but they are not needed for the actual operation of these devices; the main outcome of this 9 work is in fact the development of fully transparent integrated circuits, in which the main cause of defective behavior here highlighted is eliminated. Figure 3. Electrical characterization of the printed all-polymeric OFETs (L = 65 µm, W = 1000 µm). Transfer and output curves for p-type (a, b) and n-type (d, e) devices, based on 29-DPP-TVT and P(NDI2OD-T2) respectively. Average transfer curves and their standard deviation, for the 10x10 OFETs p-type (c) and n-type (f) arrays. The bendability of the printed transistors has been assessed, by means of bending tests with different applied bending radii, equal to 5.5 mm, 8 mm and 13 mm, which lead to strain values equal to about 0.5 %, 0.8 % and 1.1 %. For each bending radius, the transfer curve of one transistor has been measured before the test and after 1, 10, 100 and 1000 bending cycles; the results of this analysis are presented in Figure 4. For what concerns the p-type devices, there is only a slight increase in the 10 maximum current for increasing number of bending cycles, which is mirrored in the mobility values shown in Figure S6a; overall, there is no significant change in performances after 1000 bending cycles. Considering n-type transistors, maximum current values stay almost constant up to 1000 bending cycles for applied strains below 1 %, while for a 1.1 % strain value the current and mobility values are unchanged up to 100 bending cycles, while there is a small loss in performances after 1000 cycles, as it can be seen in Figure 4 and Figure S6b. Figure 4. Bendability tests results. Transfer curves and maximum current values, recorded before bending and in flat configuration after 1, 10, 100 and 1000 bending cycles, for p-type (a, b) and n- type (c, d) devices. Transfer curves are presented for the case with an applied strain higher than 1%. 11 The presented p- and n-type OFETs have been fabricated with the same techniques on the same substrates, employing identical electrodes and dielectrics. Therefore, their integration into complementary logic circuits is highly simplified. We started by fabricating a logic inverter, the simplest complementary circuit that can be realized, by integrating one p- and one n-type device. The inverter is fabricated by printing the two transistors with a shared drain contact and the same gate electrode, which act as output and input nodes, following the same fabrication steps presented for the transistor preparation (Figure 5a, b). Since the performances of the two different types of devices are not perfectly balanced, with the p-type transistors presenting slightly higher currents compared to the n-type ones, the channel width of the p-type devices (Wp) has been designed to be about one third of the n-type transistors' one (Wn), with Wp = 300 m and Wn = 1000 m, while keeping the same channel length, Lp = Ln = 65 m. The latter dimensions have been adopted for all complementary circuits in this work. First, a static characterization of the printed inverters has been performed. Voltage transfer curves (VTC) have been measured, sweeping the voltage input signal from a "0" logic state to a "1"; supply voltages (VD) varying from 2 V to 10 V have been used, and the so obtained VTC are plotted in Figure 5c as an average over the performances of 5 printed inverters.[45] By finding the intersection point between the VTC and the bisector of the axes, an average inverting threshold voltage of 4.90 V has been obtained for devices operated at 10 V, with a standard deviation of 0.05 V. The inverters are functional down to a supply voltage of only 2 V, for which the average inverting threshold voltage amounts to 1.19 V, with a standard deviation of 0.07 V. The average gain values, obtained as the derivative of the VTC (Figure 5d), amount to about 14 for VD = 10 V, with slightly higher values recorded for smaller input voltages. The maximum recorded gain is 17, and has been achieved with an input voltage of 2 V. The noise margins (NM), which give an estimation of the immunity of the printed inverters to noise on the input signal, have been calculated using the maximum equal criteria.[46] NM lay in the range between 50 % and 60 % of VD/2 for all supply voltages from 2 to 10 V, with the highest value obtained for VD = 2 V, for which the average NM is 12 equal to 0.62 V. The obtained figures of merit are ideal, indicating that low voltage complementary logic gates based on our printed transistors can be used in order to realize more complex integrated circuits.[45] The voltage transfer characteristics of these inverters show a small hysteresis, which can be related to the hysteresis shown by the single transistors and has already been addressed. This phenomenon is limited and reproducible, and thus doesn't affect their correct operation. Figure 5. (a) Schematic representation and (b) optical micrograph of the printed inverter. (c) Voltage transfer curves and (d) gain curves of the inverters with input voltage varying from 2 V to 10 V. In order to perform the dynamic characterization of these inverters, 7-stage ring oscillators (RO) have been fabricated (Figure 6a). An RO is a loop containing an odd number of inverters, where the output of an inverter is fed into the subsequent one, leading to its characteristic self-oscillation, whose frequency fRO depends on the number of stages (N) and on the stage delay (SD), according to 13 𝑓𝑅𝑂 = 1 2∗𝑁∗𝑆𝐷 ( 1 ) The oscillating behavior of the circuit has been characterized, with supply voltages varying from 2 V to 10 V, and results for 3 different supply voltages are presented in Figure 6c. The average oscillating frequency, evaluated on 3 different samples, goes from almost 6 Hz at 2 V to 57 Hz for a supply voltage of 10 V, leading to average stage delays of 13.1 ms and about 1.3 ms for supply voltages of 2 and 10 V, respectively (Figure 6b). The best performing circuit oscillates with a frequency of 62.5 Hz when a voltage supply of 10 V is applied, which leads a minimum stage delay achieved of 1.14 ms. At the best of our knowledge, this is the first demonstration of a printed, all-polymeric, transparent ring oscillator operating at low voltage.[8, 47-50] Figure 6. (a) Optical micrograph of the printed 7-stage ring oscillator. Shown are the points were electrical connections are created during characterization, namely VD for the supply voltage, GND for the ground and Vout for the output signal. (b) Average oscillating frequency and stage delay, evaluated over 3 different samples, for voltage supply varying from 2 to 10 V. (c) Time response of the 7-stage ring oscillator. 14 We then proceeded to the fabrication of more complex logic circuitry, in particular we integrated the printed transistor in a D-Flip-Flop (DFF), one of the basic memory elements needed to build sequential logic circuits, counters and timers.[45, 51-54] This DFF has been implemented using a master- slave configuration, in which two D-latches (details in SI) are connected in series with opposite transparency phases. During its operation, this circuit retains the input data, and it adjusts the output according to the stored state only in correspondence of the falling edge of the clock signal; further details about the working mechanism can be found in the supplementary information. For the circuit design we selected a pass-transistor based logic. This logic requires two main building blocks, the inverter, whose development and characterization has already been addressed, and the transmission gate. The latter is a logic port acting as a switch, and in its typical design it is made of two complementary transistors with shared source and drain contacts, and with opposite signals driving the gate electrodes. In our design, aiming at a reduction of the number of transistors used in each circuit, we decided to use a single p-type transistor as transmission gate, instead of the complementary port, also because in this case there is no net gain in adopting the complementary pass gate. As a consequence, our modified DFF design requires only 12 transistors. The schematic and the optical micrograph of our low-voltage transparent DFF are shown in Figure 7a,b. In Figure 7d,e the proper operational behavior of our devices is presented, for VD values equal to 10 V (d) and 2 V (e) (the operation of the D-latches sub-units is shown in Figure S8b,c). One of the characteristic feature of these devices is their transparency. We have thus evaluated the transmittance of the printed circuits by means of UV-vis spectroscopy in the visible range, and we have achieved a transmittance equal or higher than 90 % for the overall system; in Figure S9 we show the spot on the circuit where this measurement has been performed. Taking into account the contribution of the PEN substrate, and normalizing the transmittance for the circuit alone, we can show that our printed circuits have a transparency equal or higher than 95 % in the visible range, as it can be seen in Figure 7c. While previous low-voltage printed DFF has been presented in literature,[55-57] this is the first example, to 15 the best of our knowledge, of a low-voltage, all-polymer, transparent DFF, additionally obtained combining scalable processes. We assessed the shelf-life stability of non-encapsulated printed circuits stored in nitrogen. After 4 months of storage the DFF are still perfectly operational (Figure 7f), thanks to the fact that characteristics of single transistors and inverters (Figure S7) are negligibly altered. Figure 7. Schematic representation (a) and optical micrograph (b) of the transparent, all-polymer D- Flip-Flop. (c) Transparency of the printed DFF with respect to the PEN substrate. Operation of the DFF, for supply voltages of 10 V (d) and 2 V (e). (f) Operation of the same DFF, after 4 months from fabrication. 3. Conclusions We have demonstrated the integration of a parylene-based bilayer as top-gate dielectric for all- polymer, printed, transparent and low-voltage OFETs. Except for the dielectric, in which parylene was deposited through an already industrially scaled chemical vapor deposition method, all the 16 functional layers were inkjet printed on a PEN substrate. We used P(NDI2OD-T2) and 29-DPP-TVT semiconducting co-polymers for n- and p-type transistors, respectively, and we have reported correct field-effect behavior down to 1 V. The printed OFETs achieve currents higher than 1 A in both cases for operating voltages in the order of 10 V, with a high yield (99%) and uniform performances. The flexibility of these devices was assessed through bending tests, and it has been shown that they can sustain up to 1000 bending cycles with strains up to 1%. We successfully demonstrated the integration of these transistors into complementary logic circuits. First, complementary inverters were fabricated, and well balanced output characteristics were obtained, with NM equal or higher than 50 % and an average gain of 14. Next, we demonstrated the first all-polymer, transparent, printed ring oscillators operating at voltages as low as 3 V, and achieving a maximum oscillation frequency of 62.5 Hz for a supply voltage of 10 V, thus showing a stage delay of 1.14 ms. Finally, we demonstrated all-polymer, transparent, printed D-Latches and D- Flip-Flops, correctly operating at voltages as low as 2 V. The transparency of these circuits is higher than 90% for the whole device has been obtained. Life-shelf stability in nitrogen of non-encapsulated circuits exceeds 4 months after fabrication. Our results show that it is possible to fabricate fundamental logic electronics blocks, as required in serialization circuits, timers and counters, through all-polymer, transparent OFETs on plastic capable of operating at voltages compatible with thin film batteries or energy harvesters, paving the way to their integration into consumer products with low additional costs. 4. Experimental Section Bottom-contact/top-gate organic field effect transistors, whose structure is presented in Figure 2a, have been fabricated on a 125 m-thick poly(ethylene 2,6-naphthalate) (PEN) substrate, purchased from Du Pont. Poly(3,4-ethylenedioxithiophene):polystyrene sulphonate (Clevios PJ700 formulation, 17 purchased from Heraeus) source and drain contacts (Figure 2b) have been inkjet-printed by means of a Fujifilm Dimatix DMP2831. This process allows to fabricate transistors with a channel width of about 1000 m and channel length in the order of 65 m. In order to facilitate the characterization of these devices, small pads of a silver nanoparticles-based ink have been printed (Silverjet DGP-40LT-15C, purchased from Advanced Nano Products (ANP)) in correspondence of the points where electrical contacts with external probes are made during measurements. This is mainly related to the fact that these devices are transparent and it is thus quite difficult to characterize them without the help of clearly visible contact pads. All the devices presented in this work have been fabricated with silver pads, but it is necessary to point out that they are not required for the actual operation of any of the circuits presented here. For what concerns the n-type transistors, a PEI-based injection layer has been printed on top of the contacts (Figure 2c). The printed solution contains ethylene glycol (20% vol, purchased Sigma Aldrich), a solution of zinc oxide (ZnO) nanoparticles in isopropanol (nanoparticles concentration 2.5% wt, this solution amounts to 30% vol of the total volume of the final solution) and PEI (branched, purchased from Sigma Aldrich, average Mw ≈ 10 000), dissolved in water with a weight concentration of 0.2% (50% vol of the final solution). As semiconductors, P(NDI2OD-T2) (ActivInk N2200, Flexterra Corporation) has been used for the n-type OFETs, and 29-DPP-TVT (synthesized according to Yu et al.[58]) for p-type devices. Both semiconductors were patterned by inkjet-printing (Figure 2d,e). P(NDI2OD-T2) has been printed from a mesitylene-based solution, with a concentration of 7 mg ml-1, while 1,2-dichlorobenzene was used as solvent for 29-DPP-TVT, to yield a concentration of 2.5 mg ml-1. After printing, annealing in nitrogen atmosphere have been performed; P(NDI2OD-T2) has been annealed for 3 hours at 120 °C, while 29-DPP-TVT for 5 minutes at 110 °C. The bilayer dielectric stack is composed of a thin poly(methyl methacrylate) (PMMA) layer (purchased from Sigma Aldrich, average Mw ≈ 120 000), and a thicker parylene-C layer (dimer purchased from Specialty Coating Systems). The thickness of the PMMA layer is in the order of 20 18 nm, while the parylene-C layer is about 100 nm thick. The PMMA layer has been spin-coated and annealed at 100 °C for one hour in a nitrogen atmosphere, while parylene-C is deposited by means of chemical vapor deposition, using a SCS Labcoater 2 -- PDS2010 system. Parylene dimer underwent pyrolysis inside the furnace chamber at about 690 °C, while deposition has been performed at room temperature, with a pressure in the chamber in the order of 15 mTorr. PEDOT:PSS gate contacts have been inkjet printed on top of the dielectric layer (Figure 2f). For what concerns circuits, via-holes contacting different conductive layers have been fabricated by either by laser or chemical drilling. The characterization of the transistors and circuits has been performed in an inert nitrogen atmosphere. The transfer and output curves for the transistors, together with the voltage transfer curves for the inverters, have been measured using an Agilent B1500A Semiconductor Parameter Analyzer. The dynamic characterization of the integrated circuits has been performed using the above mentioned Semiconductor Parameter Analyzer for recording the output signals, while the input signals and the voltage supply were provided by an Agilent 81150A waveform generator and an Agilent B2912A source meter. For what concerns the dielectric characterization, an Agilent 4294A impedance analyzer has been employed. 19 Supporting Information Supporting Information is available from the Wiley Online Library or from the author. Acknowledgements We would like to acknowledge Filippo Melloni for his help with the electrical characterization. This work would not have been possible without the support by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program "HEROIC", Grant Agreement No. 638059. This work has also been partially supported by the Italian Ministry of Economic Development through the project F/050241/03/X32 under the "Fondo per la Crescita Sostenibile" -- Call HORIZON2020 PON I&C 2014-2020. The authors are thankful to the company FLEX for very useful discussions. Received: ((will be filled in by the editorial staff)) Revised: ((will be filled in by the editorial staff)) Published online: ((will be filled in by the editorial staff)) References [1] H. Sirringhaus, Adv. Mater. 2014, 26, 1319. [2] H. Klauk, Chem. Soc. Rev. 2010, 39, 2643. [3] G. Gelinck, P. Heremans, K. Nomoto, T. D. Anthopoulos, Adv. Mater. 2010, 22, 3778. [4] K. J. Baeg, M. Caironi, Y. Y. Noh, Adv. Mater. 2013, 25, 4210. [5] X. Guo, Y. Xu, S. Ogier, T. N. Ng, M. Caironi, A. Perinot, L. Li, J. Zhao, W. Tang, R. A. Sporea, A. Nejim, J. Carrabina, P. Cain, F. Yan, IEEE Trans. Electron Devices 2017, 64, 1906. [6] J. Kwon, Y. Takeda, K. Fukuda, K. Cho, S. Tokito, S. Jung, ACS nano 2016, 10, 10324. 20 A. Sou, S. Jung, E. Gili, V. Pecunia, J. Joimel, G. Fichet, H. Sirringhaus, Org. Electron. [7] 2014, 15, 3111. [8] S. Mandal, G. Dell'Erba, A. Luzio, S. G. Bucella, A. Perinot, A. Calloni, G. Berti, G. Bussetti, L. Duò, A. Facchetti, Y.-Y. Noh, M. Caironi, Org. Electron. 2015, 20, 132. [9] M. Koo, K.-I. Park, S. H. Lee, M. Suh, D. Y. Jeon, J. W. Choi, K. Kang, K. J. Lee, Nano Lett. 2012, 12, 4810. [10] L. Hu, H. Wu, F. La Mantia, Y. Yang, Y. Cui, ACS nano 2010, 4, 5843. [11] N. Piva, F. Greco, M. Garbugli, A. Iacchetti, V. Mattoli, M. Caironi, Adv. Electron. Mater. 2017. [12] C. J. Brabec, N. S. Sariciftci, J. C. Hummelen, Adv. Funct. Mater. 2001, 11, 15. [13] A. Facchetti, M. H. Yoon, T. J. Marks, Adv. Mater. 2005, 17, 1705. [14] R. P. Ortiz, A. Facchetti, T. J. Marks, Chem. Rev. 2009, 110, 205. [15] J. Veres, S. D. Ogier, S. W. Leeming, D. C. Cupertino, S. Mohialdin Khaffaf, Adv. Funct. Mater. 2003, 13, 199. [16] A. Stassen, R. De Boer, N. Iosad, A. Morpurgo, Appl. Phys. Lett. 2004, 85, 3899. [17] A. Luzio, F. G. Ferré, F. D. Fonzo, M. Caironi, Adv. Funct. Mater. 2014, 24, 1790. [18] M. Zirkl, A. Haase, A. Fian, H. Schön, C. Sommer, G. Jakopic, G. Leising, B. Stadlober, I. Graz, N. Gaar, R. Schwödiauer, S. Bauer-Gogonea, S. Bauer, Adv. Mater. 2007, 19, 2241. [19] V. Pecunia, K. Banger, H. Sirringhaus, Adv. Electron. Mater. 2015, 1. [20] Y. Wang, S.-G. Lu, M. Lanagan, Q. Zhang, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 2009, 56. [21] S. H. Kim, K. Hong, W. Xie, K. H. Lee, S. Zhang, T. P. Lodge, C. D. Frisbie, Adv. Mater. 2013, 25, 1822. [22] B. Nketia‐Yawson, S. J. Kang, G. D. Tabi, A. Perinot, M. Caironi, A. Facchetti, Y. Y. Noh, Adv. Mater. 2017, 29. [23] W. F. Gorham, J. Polym. Sci., Part A: Polym. Chem. 1966, 4, 3027. 21 [24] T. Marszalek, M. Gazicki-Lipman, J. Ulanski, Beilstein J Nanotechnol 2017, 8, 1532. [25] K. N. N. Unni, S. Dabos-Seignon, J.-M. Nunzi, J. Mater. Sci. 2006, 41, 317. [26] M. Kondo, T. Uemura, T. Matsumoto, T. Araki, S. Yoshimoto, T. Sekitani, Appl. Phys. Express 2016, 9. [27] R. A. Nawrocki, N. Matsuhisa, T. Yokota, T. Someya, Adv. Electron. Mater. 2016, 2. [28] K. Fukuda, T. Sekine, R. Shiwaku, T. Morimoto, D. Kumaki, S. Tokito, Sci Rep 2016, 6, 27450. [29] T. Yasuda, K. Fujita, H. Nakashima, T. Tsutsui, Jpn. J. Appl. Phys. 2003, 42, 6614. [30] R. Shiwaku, Y. Yoshimura, Y. Takeda, K. Fukuda, D. Kumaki, S. Tokito, Appl. Phys. Lett. 2015, 106. [31] R. Shiwaku, Y. Takeda, T. Fukuda, K. Fukuda, H. Matsui, D. Kumaki, S. Tokito, Sci Rep 2016, 6, 34723. [32] B. Peng, P. K. L. Chan, Org. Electron. 2014, 15, 203. [33] P. Cosseddu, S. Lai, M. Barbaro, A. Bonfiglio, Appl. Phys. Lett. 2012, 100. [34] P. Darmawan, T. Minari, Y. Xu, S.-L. Li, H. Song, M. Chan, K. Tsukagoshi, Adv. Funct. Mater. 2012, 22, 4577. [35] D. Natali, J. Chen, F. Maddalena, F. García Ferré, F. Di Fonzo, M. Caironi, Adv. Electron. Mater. 2016, 2. [36] D. Natali, M. Caironi, Adv. Mater. 2012, 24, 1357. [37] H. Sirringhaus, Adv. Mater. 2005, 17, 2411. [38] Y. Xu, C. Liu, D. Khim, Y. Y. Noh, Phys. Chem. Chem. Phys. 2015, 17, 26553. [39] E.-Y. Shin, E.-Y. Choi, Y.-Y. Noh, Org. Electron. 2017, 46, 14. [40] S. Y. Park, M. Park, H. H. Lee, Appl. Phys. Lett. 2004, 85, 2283. [41] Y. Zhou, C. Fuentes-Hernandez, 2012. [42] S. Fabiano, S. Braun, X. Liu, E. Weverberghs, P. Gerbaux, M. Fahlman, M. Berggren, X. Crispin, Adv. Mater. 2014, 26, 6000. 22 [43] S. M. Sze, K. K. Ng, Physics of semiconductor devices, John wiley & sons, 2006. [44] H. H. Choi, K. Cho, C. D. Frisbie, H. Sirringhaus, V. Podzorov, Nat. Mater. 2017, 17, 2. [45] J. M. Rabaey, A. P. Chandrakasan, B. Nikolic, Digital Integrated Circuits: A Design Perspective, 2003. [46] J. R. Hauser, IEEE Transactions on Education 1993, 36, 363. [47] A. Knobloch, A. Manuelli, A. Bernds, W. Clemens, J. Appl. Phys. 2004, 96, 2286. [48] H. Kempa, M. Hambsch, K. Reuter, M. Stanel, G. Schmidt, B. Meier, A. Hubler, IEEE Trans. Electron Devices 2011, 58, 2765. [49] A. Hübler, G. Schmidt, H. Kempa, K. Reuter, M. Hambsch, M. Bellmann, Org. Electron. 2011, 12, 419. [50] M. Hambsch, K. Reuter, M. Stanel, G. Schmidt, H. Kempa, U. Fügmann, U. Hahn, A. Hübler, Mater. Sci. Eng., B 2010, 170, 93. [51] G. Dell'Erba, A. Luzio, D. Natali, J. Kim, D. Khim, D.-Y. Kim, Y.-Y. Noh, M. Caironi, Appl. Phys. Lett. 2014, 104. [52] Y. Takeda, K. Hayasaka, R. Shiwaku, K. Yokosawa, T. Shiba, M. Mamada, D. Kumaki, K. Fukuda, S. Tokito, Sci Rep 2016, 6, 25714. [53] S. Jacob, S. Abdinia, M. Benwadih, J. Bablet, I. Chartier, R. Gwoziecki, E. Cantatore, A. Van Roermund, L. Maddiona, F. Tramontana, Solid·State Electron. 2013, 84, 167. [54] M. Uno, N. Isahaya, B. S. Cha, M. Omori, A. Yamamura, H. Matsui, M. Kudo, Y. Tanaka, Y. Kanaoka, M. Ito, Adv. Electron. Mater. 2017, 3. [55] Y. Xia, W. Zhang, M. Ha, J. H. Cho, M. J. Renn, C. H. Kim, C. D. Frisbie, Adv. Funct. Mater. 2010, 20, 587. [56] K. Hayasaka, H. Matsui, Y. Takeda, R. Shiwaku, Y. Tanaka, T. Shiba, D. Kumaki, S. Tokito, Adv. Electron. Mater. 2017, 3. 23 [57] M. Elsobky, M. Elattar, G. Alavi, F. Letzkus, H. Richter, U. Zschieschang, M. Strecker, H. Klauk, J. N. Burghartz, Org. Electron. 2017, 50, 491. [58] H. Yu, K. H. Park, I. Song, M.-J. Kim, Y.-H. Kim, J. H. Oh, Journal of Materials Chemistry C 2015, 3, 11697. 24 A parylene-based high-capacitance dielectric bilayer enables high-yield fabrication of low-voltage printed polymer integrated circuits on plastic, based on top-gate field effect transistors with low-leakage and high-uniformity. All-polymer complementary circuits, such as 7-stages ring oscillators and D-Flip-Flops, properly operating at voltages as low as 2 V, are demonstrated. Keyword organic field-effect transistors, printed integrated circuits, parylene, inkjet printing, flexible electronics E. Stucchi, G. Dell'Erba, P. Colpani, Y.-H. Kim, M. Caironi* Low-Voltage Printed, All-Polymer Integrated Circuits Employing a Low-leakage and High-Yield Polymer Dielectric 25 Copyright WILEY-VCH Verlag GmbH & Co. KGaA, 69469 Weinheim, Germany, 2016. Supporting Information Low-Voltage Printed, All-Polymer Integrated Circuits Employing a Low-leakage and High-Yield Polymer Dielectric Elena Stucchi, Giorgio Dell'Erba, Paolo Colpani, Yun-Hi Kim and Mario Caironi Role of PMMA interlayer Figure S1. Transfer curves of two transistors employing identical structures except for the dielectric layers, which is parylene only (orange curve) or a bilayer composed of PMMA and parylene (black curve). 26 Bode plot for the capacitors Figure S2. Bode plot of the capacitors. Shown are the impedance (a) and the phase (b) as a function of frequency. (c) Capacitance as a function of frequency for capacitor employing a 120 nm thick parylene layer. The capacitance here is measured down to 5 Hz, and the value recorded at the minimum frequency is the one used in order to extract the field effect mobilities. 27 Transistors yield -- raw data of all the measured transfer curves Figure S3. Raw data of the transfer curves for the 10x10 transistors array. Data for the n-type (a) and p-type (b) transistors, with linear regime shown on the left (measured at 2 V), and saturation regime on the right (measured at 10 V). Red curves refer to the leakage current. 28 Mobility curves Figure S4. Average mobility curves as a function of the gate voltage, together with their standard deviation, for n-type (a) and p-type (b) devices, in linear (±2 V, black curve) and saturation (±10 V, blue curve) regime. Optical micrograph of defective transistors Figure S5. Optical micrograph of the defective transistors. (a) and (b) show the two transistors that are not working. In both cases the cause of failure is to be related with the presence of dust particles inside the channel. (c) Example of a transistor with non-ideal leakage, caused by the presence of silver droplets inside the channel. 29 Bending test results Figure S6. Effect of the bending tests on different transistors' figures of merit, for p-type (a) and n-type (b) devices. From left to right, mobility evaluated in linear regime (2 V), mobility in saturation regime (10 V) and onset voltage. 30 Performances of the inverter after 4 months storage in nitrogen atmosphere Figure S7. Performances of the same n-type transistor (a), p-type transistor (b) and of the same inverter (c-d) right after the device fabrication, after 2 months and after 4 months of storage in nitrogen atmosphere. D-Latch The D-Latch is a device used to store 1 bit of information. This circuit is composed of two inverters and two transmission gates, working with opposite phases, and it requires only 6 transistors to be implemented with the technology we are using. The operation of the D-latch can be divided into two main phases, the transparency and the hold phase. During the first, the clock is high and the input signal passes through the circuit and reaches the output, so the latch is said to be in transparent mode. On the contrary, when the clock is low, the device holds stably as output the signal received as input on the falling edge of the clock. Analyzing in details what happens inside the circuit, during the transparency phase, when the clock (CK) is high, and so 31 CK' is low, the first transmission gate TG1 is acting as short circuit, while TG2 is acting as an open circuit. The input DATA is thus inverted twice, by both the first inverter INV1 and the INV2, and then transmitted to the output node, so the output of the transparency phase is the delayed copy of the input signal. On the contrary, during the hold phase, the clock CK is low and CK' is high, so TG1 is acting as open circuit and TG2 as short, which means that they form a closed loop and the output node corresponds to the input of the inverter INV1. The output is not sensitive to any change in DATA during the opaque phase. The operation of the two latches DL1 and DL2 of the working D-Flip-Flop presented in this paper is shown in Figure S7. Figure S8. (a) Schematic representation of the printed, all-organic D-Latch. Operation of the first (b) and the second (c) D-Latch composing the D-Flip-Flop presented in this paper. D-Flip-Flop The Master-Slave D-Flip-Flop is one of the basic circuits used to store information. The Master- Slave DFF is implemented by making use of two D-Latch circuits, connected in series and with opposite clock signals, so that when one is in its transparency phase, the other is in opaque phase, and vice versa. The output signal of this device is kept constant all the time and, being a falling-edge triggered device, might change its output value only when the clock signal is 32 switching from high to low. In fact, in that instant, the first D-Latch DL1 goes from its transparency phase to the hold one, sampling the input data and then storing it into its feedback loop, while the second D-Latch DL2 switches from opaque to transparent, making thus possible for the input signal sampled during the falling edge of the clock to reach the output node. UV-vis spectra measurement details Figure S9. Schematic of the spot where the UV-vis spectra has been measured. 33
1912.01985	1	1912	2019-12-04T13:55:32	Unraveling the degradation mechanism in FIrpic based Blue OLEDs: II. Trap and detect molecules at the interfaces	[ "physics.app-ph", "cond-mat.mtrl-sci", "physics.chem-ph" ]	The impact of organic light emitting diodes (OLEDs) in modern life is witnessed by their wide employment in full-color, energy-saving, flat panel displays and smart-screens; a bright future is likewise expected in the field of solid state lighting. Cyclometalated iridium complexes are the most used phosphorescent emitters in OLEDs due to their widely tunable photophysical properties and their versatile synthesis. Blue-emitting OLEDs, suffer from intrinsic instability issues hampering their long term stability. Backed by computational studies, in this work we studied the sky-blue emitter FIrpic in both ex-situ and in-situ degradation experiments combining complementary, mutually independent, experiments including chemical metathesis reactions, in liquid phase and solid state, thermal and spectroscopic studies and LC-MS investigations. We developed a straightforward protocol to evaluate the degradation pathways in iridium complexes, finding that FIrpic degrades through the loss of the picolinate ancillary ligand. The resulting iridium fragment was than efficiently trapped "in-situ" as BPhen derivative 1. This process is found to be well mirrored when a suitably engineered, FIrpic-based, OLED is operated and aged. In this paper we (i) describe how it is possible to effectively study OLED materials with a small set of readily accessible experiments and (ii) evidence the central role of host matrix in trapping experiments.	physics.app-ph	physics	Unraveling the degradation mechanism in FIrpic based Blue OLEDs: II. Trap and detect molecules at the interfaces Marta Penconia, Marco Cazzanigaa†, Walter Panzerib, Andrea Meleb,c, Fausto Cargnonia, Davide Ceresolia, Alberto Bossia* a Istituto di Scienze e Tecnologie Molecolari - CNR, PST di via Fantoli 16/15, 20138 Milan and via C. Golgi 19, 20133 Milano, Italy; SmartMatLab Center, Via C. Golgi 19, 20133 Milano, Italy, b Istituto di Chimica del Riconoscimento Molecolare - CNR, "U.O.S. Milano Politecnico" Via Mancinelli 7, 20131 Milano Italy, c Dipartimento di Chimica, Materiali e Ingegneria Chimica "Giulio Natta", Politecnico di Milano, Via Mancinelli, 7, 20131 Milano, Italy ABSTRACT: The impact of organic light emitting diodes (OLEDs) in modern life is witnessed by their wide employment in full-color, energy-saving, flat panel displays and smart-screens; a bright future is likewise expected in the field of solid state lighting. Cyclometalated iridium complexes are the most used phosphorescent emitters in OLEDs due to their widely tunable photophysical properties and their versatile synthesis. Blue-emitting OLEDs, suffer from intrinsic instability issues hampering their long term stability. Backed by computational studies, in this work we studied the sky-blue emitter FIrpic in both ex-situ and in-situ degradation experiments combining complementary, mutually independent, experiments including chemical metathesis reactions, in liquid phase and solid state, thermal and spectroscopic studies and LC-MS investigations. We developed a straightforward protocol to evaluate the degradation pathways in iridium complexes, finding that FIrpic degrades through the loss of the picolinate ancillary ligand. The resulting iridium fragment was than efficiently trapped "in-situ" as BPhen derivative 1. This process is found to be well mirrored when a suitably engineered, FIrpic-based, OLED is operated and aged. In this paper we (i) describe how it is possible to effectively study OLED materials with a small set of readily accessible experiments and (ii) evidence the central role of host matrix in trapping experiments. INTRODUCTION Organic electronic technologies, which include light emitting diodes, field effect transistors and organic photovoltaics, are vibrant research sectors from both the academic and industrial standpoint.1 In particular organic light emitting diodes (OLEDs) stand out like the reference technology employed in energy saving flat panel displays and looked as promising alternative to develop solid state lighting.2 These devices consist of organic and organometallic materials arranged in a thin-film multilayer architecture, either solution-processed or vacuum-deposited, sandwiched between two electrodes. In OLEDs, light is generated and emitted as the result of the electroluminescence (EL) process which involves i) injection of electrons and holes under applied voltage from the cathode and anode, respectively, ii) migration of the charges through the device layers and iii) charge trapping and recombination, with formation of excitons, at the emitting material. Excitons will ultimately and radiatively decay to the ground state. Phosphorescent transition metal complexes (mostly based on Ir(III) and Pt(II) metal ions), doped in a host material, have been shown to yield 100% internal quantum efficiency by harvesting both singlet triplet electrogenerated excitons.3 High external quantum efficiencies have been achieved combining light extraction techniques with device engineering. Improved exciton recombination in the emissive layer has been obtained using blocking layers and carefully selecting the layered materials to improve the charge injection and charge transport balance. Cyclometalated iridium complexes are the most used phosphorescent emitters because of their fine tunable photophysical properties and the versatile synthesis.3,4 Iridium complexes have been developed to span emissions from the deep UV down to the near IR spectral region as a result of the great demand for commercial applications. Yet, two big issues are evident: the lack of efficient emitters for the NIR region3 and the if severe instability issues of blue emitters. Especially in the latter contest, the study of the degradation processes, ultimately leading to OLED failure, is essential to meet the device lifetime and the color stability needed for commercialization. Degradations in OLEDs can occur through extrinsic mechanisms as well as through a long-term intrinsic degradation.5,6 Extrinsic mechanisms, such as the formation of non-emissive dark spots and electrical shorts, are tightly linked to the fabrication conditions. A careful control and optimization of the processing methods can ultimately limit and avoid their occurrence; for instance, high-vacuum thermal evaporation methods offer the best conditions to meet the standards of efficiency and lifetime of commercial devices.7-9 On the contrary, the intrinsic mechanisms, still holding many open issues, mainly consist in the chemical alteration of the organic or metal-organic materials affecting the device emission efficiency (i.e. brightness decrease and operating voltage rise),10 and they are particularly relevant in the emitting layer. While a recent work highlights the role of the exciton-induced generation of radical ion pairs between the host and the dopant emitter,11 we are mainly interested in the fate of the exciton once it is trapped on the dopant emitter and it can initiate intramolecular degradations from the molecular excited state (therefore forming nonradiative recombination centers and luminescence quenchers). The understanding of degradations occurring in phosphorescent blue-emitters is far more important compared to degradation of the green or red ones. In fact, the higher energy of the emitted blue photons,4 i.e. the higher energy of the emitter excited state, combined to the long (~0.5-5 μs) triplet lifetime, can lead to the thermal population of non-radiative excited states (involving antibonding orbitals). As a result, blue OLEDs display shorter lifetime. chemical The study of emitter degradation is challenging and requires the a priori identification of the decomposition products to frame a mechanistic understanding. The difficulties, from the experimental standpoint, lie in the fact that emitters are doped in low concentration in the EML and only a small fraction of them degrade; on one hand, even tiny amount of degraded compounds produce efficient traps, on the other hand, very small quantity of material is available for analysis (considering the average of 20-40 nm thickness of the emissive layer), limiting the number of suitable analytical techniques. The high-performance liquid chromatography coupled with UV-Vis12 and mass spectrometer (HPLC-MS)13 detectors or laser desorption/ionization time-of- flight mass spectrometry (LDI-TOF-MS)14,15 also Ir(ppy)318,19 combined with X-ray photoelectron spectroscopy (XPS)16 offer the highest sensitivity for such studies and have been employed in early investigations. Several studies reported ligand dissociation reaction under device operation and demonstrated that it can be an irreversible process the resulting coordinatively unsaturated fragment undergo further reactions with surrounding molecules from the adjacent transport layers, for example 4,7-diphenyl-1,10- phenanthroline (BPhen). This has been investigated by Leo et al. for a variety of organometallic iridium complexes such as Ir(MDQ)2(acac),17 and FIrpic20,21 employing LDI-TOF-MS technique, although a possible damage effect of the incident laser during the analysis cannot be completely ruled out as source of fragmentation.5,16 Moreover, evidences of red shift in the electroluminescence spectra under device operation have been reported, ligand substitution22,23 or related to emitter aggregation (FIrpic).9,24 Thompson and co-worker reported on the device aging and photochemical stability of the well-known Ir(ppz)3.23 This is also the case of FIrpic (bis[2-(4,6-difluorophenyl)pyridyl- C2,N] (picolinato) iridium(III)) (Scheme 1), the most representative blue phosphorescent emitter. It provides highly efficient blue OLED as dopant in host matrix but it is highly instable leading to fast failure in operating device.25 Several studies reported that ligand dissociation in FIrpic can be irreversible since it can be followed by CO2 loss.20 Moreover, the cleavage of fluorine atoms has been observed after thermal evaporation of the material and device aging.26 ascribed whether to In summary, Cargnoni and Ceresoli thoroughly detailed the degradation mechanism of FIrpic in both ground and excited states and by exciton formation.27 Among their results, it stands out the weakness of the picolinate ligand (pic) towards excited state dissociation. FIrpic sequentially undergoes a bond breaking at the Ir- N(pic) site followed by loss of the picolinate ligand.27 In this hypothesis, the coordinative open, left-over, Ir fragment could be trapped with proper molecules. On this ground, our experiments are designed to track the degradation follow-up. In this study we speculated, thought later proved, that once picolinate falls apart (Scheme 1), the degraded molecule could be trapped by a strongly coordinating bidentate phenantroline ligand, BPhen, thus obtaining the charged heteroleptic [Ir(F2ppy)2(BPhen)]+, complex 1. The simplified scheme of the overall degradation/trapping process is depicted in Scheme 1. The use of BPhen provides complex 1 with properties significantly different compared to FIrpic allowing for multi-technique detection/analysis; in particular, complex 1 is a charged compound, its molecular weight is different from that of the FIrpic and in addition it is reported to have orange luminescence.28 through This work aims at presenting a critical experimental investigation of the FIrpic degradation process studies encompassing the intrinsic chemical lability of the FIrpic ligands, the optical properties of FIrpic and derivative 1, and finally the degradation environment inside an OLED under electrical stress (i.e. promoting degradation under electrically pumped exciton formation). Therefore, the next sections are organized as follow: 1) synthesis of the complex 1, its optical properties and those of FIrpic; 2) FIrpic reactivity towards thermal ligand metathesis reaction with BPhen, both in solution and in solid phase, and development of suitable analytical and optical frameworks; 3) design of engineered OLED with a bulk interface between FIrpic and BPhen (the latter used also as electron transport layer - ETL), aging and real time monitoring of device degradations under electrical stress; 4) morphological, spectroscopic and chemical analysis on OLED devices. For its conception, the experimental set up is worth for a broader exploitation in OLED material analysis and would provide guidelines in the molecular engineering of stable phosphorescent emitter. RESULT AND DISCUSSION 1. Synthesis of the complex 1, optical properties of complex 1 and FIrpic According to the theoretical framework, that suggests the detachment of the picolinate ligand from FIrpic,27 the fate of the emitter can be followed by inducing the coordination of the BPhen molecule to the unsaturated residue [Ir(F2ppy)2]+ giving the cationic complex 1, as depicted in Scheme 1. For comparative purpose, complex 1 has been synthesized according to literature procedure (see supporting information for details).28 In Figure 1a, the absorption and emission spectra of the two iridium complexes, FIrpic and 1, are reported and their photophysical properties are summarized in Table S1. In deaerated CH2Cl2 solution at room temperature, FIrpic displays a structured emission with maximum at 468 nm, arising from a mixed LC- MLCT character of the transition.24 It has phosphorescence quantum yield of 97% and a lifetime of 1.7 μs. In comparison, the emission spectrum of complex 1 is significantly red shifted with peak maxima at 514 nm. The phosphorescence quantum yield and lifetime are 51% and 5.4 μs, and broader respectively. In this case, a dominant MLCT character of the emission is suggested by the slightly featureless phosphorescence. When deposited as neat thin film by spin coating, the emission of both complexes are red-shifted, with maxima at 481 nm and 548 nm for FIrpic and complex 1, respectively (Figure 1b); whereas FIrpic still shows a partial structured emission, complex 1 has a broad gaussian profile. Scheme 1. Ligand dissociation process of FIrpic and coordination to BPhen molecule to form complex 1. The quantum yields and lifetimes in neat thin film are respec tively 15% and 0.16 μs for FIrpic and 37% and 2.8 μs for complex 1. 2. Thermal ligand metathesis reactions In order to prove the dominant mechanism of picolinate loss in FIrpic, its reactivity towards thermal ligand metathesis with BPhen was explored in solution and also in the solid state; the latter mirroring the conditions of the stacks in a OLED device. An equimolar solution of FIrpic and BPhen in argon degassed 2-ethoxyethanol was heated at 100 °C for few hours. The reaction, monitored via thin layer chromatography, shows the slow disappearance of FIrpic and the formation of a new brightly luminescent compound with lower Rf, signature of the more polar/ionic nature of the newly formed compound (as expected by the replacement of a monoanionic ligand -- i.e. F2ppy or pic - with the neutral BPhen). Electrospray mass spectroscopy (ESI-MS) analysis were carried out to monitor the crude reaction mixture after 3, 6 and 24 hours. Figure 2 displays the ESI-MS spectra after 1 day; the base peak at m/z = 905 and the isotopic pattern are consistent with the ion of elemental composition [C46H28N4F4Ir]+, assignable to complex 1. Noticeable, no co-products deriving from metathesis of BPhen with ligand other than the pic were found (i.e. excluding the abstraction of F2ppy), as well as no significant amount of residual FIrpic. A freshly prepared FIrpic-BPhen mixture was injected as a control to rule out any artefact deriving from gas phase reaction inside the ESI source; the ESI analysis (Figure S4) shows the protonated or sodium cationized species of the BPhen and FIrpic as well as a possible clustered cationic system of [FIrpic-BPhen-Na]+ at m/z 1050, while no evidence of the signal at m/z 905 has been observed. Figure 1. a) UV/Vis absorption (straight line) and emission (line with drawings) normalized spectra in deaerated CH2Cl2 solution and b) normalized emission spectra of neat thin films of FIrpic (black) and complex 1 (blue). Hence, the reactivity of an equimolar blend of FIrpic and BPhen was investigated in solid state. Thermal analysis was followed via differential scanning calorimetry (DSC) in the temperature range up to 300 °C (N2 flushed, open vessel). Upper limit temperature was chosen according to the melting point of BPhen (218-220°C) and that of FIrpic (330-335°C). DSC was run either placing the finely grinded powder mixture of the two compounds in the DSC crucible (experiment DSC- A) or dissolving the mixture in dichloromethane and let the solution evaporate in the crucible (experiment DSC-B). The DSC curves are reported in Figure S5. In both cases, a broad exothermic peak appears at 250°C with onset around 180°C. DSC-A showed the endothermic melt of BPhen at ca. 220°C immediately prior the exothermic reaction. In the second case (DSC-B), a slight melting is only perceivable around 180°C followed by exothermic degradation. The residue in the crucibles were than subjected to HPLC-MS analysis and compared to the as-prepared FIrpic- BPhen mixture. After both DSC-A and DSC-B experiments, the residue analysis pointed out the disappearance of FIrpic signal and the presence of 1 (see Figures S6, S7 and S8). The solid state reactivity was also followed by photoluminescence (PL) experiments. Thin films with similar architecture as later used in the OLEDs were prepared by spin coating. A FIrpic- BPhen neat film (1 to 1 ratio) was spin-coated onto a quartz slide and then thermally annealed for 30 min at 100°C. The photoluminescence spectrum of the pristine film, peaked at 470 nm, evolves upon annealing to a broad peak with maximum at 560 nm grown as a shoulder to the FIrpic emission (Figure 2); this band matches the solid state luminescence of 1. Proving that the spectral change is the consequence of the FIrpic/BPhen metathesis reaction, PL upon thermal annealing was investigated both in the neat film of FIrpic as well as for a FIrpic-BPhen pair dispersed into a PMMA matrix (4%w/w), showing, in both cases, no significant spectral alteration (Figures 2 and S9). Altogether, formation of complex 1 is confirmed both by ESI-MS analysis and PL experiments; the two techniques provide accessible and independent methods to follow the metathesis process. Mass spectrometry can discriminate FIrpic from 1 on the basis of an over 200 Da difference in molecular weight, likewise their emission wavelengths differ of 50 - 70 nm (respectively for the solution and the neat thin film). The trapping reaction is efficient both in solution and, worth standing, in solid state when the two compounds are intimately mixed; DSC and PL prove (i) the formation of 1 as the single degradation/trapped product, (ii) the possibility to follow the process by mass-spectrometry and photoluminescence, and (iii) the occurrence of an intermolecular reaction to form 1 when the two molecule are in close proximity. 3. Engineered OLED with an interface between FIrpic and the chemical trap BPhen Under operation, OLEDs would provide suitable conditions for the emitter degradation through the electrical generation of excitons (rather than by thermal mechanism). In particular, according to the theoretical framework, the emitter is even more prone towards ligand dissociation when triplet exciton are formed, given the very low activation barrier for Ir-pic detachment.27 Four architectures of FIrpic based devices (Figures 3 and S10) were prepared by physical vapor deposition (PVD). With the aim to promote efficient trapping of the FIrpic degra and B80, the two differ in the thickness of the heterojunction layer, been respectively 20 and 80 nm). A third architecture was built as a reference, where FIrpic was doped into N,N'-dicarbazolyl- 3,5-benzene (mCP) matrix (device C in Figure 3).30 Figures 4, 5 and 6 report the energy levels and layer thicknesses of the materials in the respective stacks, the electroluminescent spectrum before and after aging, and the I/V/ELintensity profiles. Voltage independent EL is obtained, indicating the correct charge trapping and exciton confinement in the EML. Device A switches on at a Von = 4 - 5 V. The initial EL spectrum perfectly matches the PL one of FIrpic neat film (Figure 4b). The device was then aged at constant current density (160 mA/cm2) for 50 min. The EL intensity drops to 17% of the initial value after only 10 min of electrical aging and continues to decrease as time elapses. The EL spectrum after aging shows a remarkable change: the FIrpic emission decreases, becoming almost negligible, while a new band at 580 nm appears. Visually, the color of the emitted light changes from blue to green- yellow and the brightness noticeably drops. The current density behavior versus the applied potential decreases by four times at the end of the electrical stress (Figure 4c). Analogously, the aging at 160 mA/cm2 for 50 min applied to the device B20 (Figure 5) results in a drastic decrease of the EL intensity (being 20% of the initial value after 10 min); the FIrpic emission is quickly replaced by a broad band centered 550 nm, again resembling the luminescence of the complex 1. The I/V curve of the device before and after aging shows the same behavior observed for the device A. In both devices, the prominent EL redshift reproduces very nicely the spectral shift observed in the thermal experiments described above (section 2). As a control, device C was made in a typical doped architecture with FIrpic dispersed in 6% weight into the mCP host. Figure 6 reports the energetics, thickness and properties of the device. Prior aging the EL spectrum matches the PL emission of FIrpic in solution (Figure 6b) while driving the device at 160 mA/cm2 or even at 16 mA/cm2, results in a remarkable decrease of both EL intensity (over 90% of intensity is lost after 10 min at 160mA/cm2) and current density in similar manner as in devices A and B. On the other hand, its EL spectrum shows no significant modification given the lack of any macroscopic contact between the FIrpic and the BPhen ETL; the loss of efficiency would arise from exciton induced formation of dark traps originated by FIrpic degradation. This observation fits with the Figure 2. (top) Positive polarity ESI-MS spectrum of the crude thermal metathesis reaction mixture FIrpic-BPhen after 1 day. MWFIrpic=695.1 g/mol, MWBPhen=332.1 g/mol, MW1=905.2 g/mol.(bottom) Normalized emission spectra of FIrpic-BPhen mixture in neat film before (black line) and after (blue line) thermal treatment at 100°C for 30 min and in PMMA matrix before (red line) and after (grey line with drawings) thermal treatment, λex=430 nm. dation product by coordination to BPhen, the emitter is employed as neat layer sandwiched between the common hole transport layer (HTL) N,N'-di[(1-naphthyl)-N,N′-diphenyl]-1,1'- biphenyl)-4,4'-diamine (NPD) and BPhen (used as ETL). This type of device, with the simplest architecture ITO/NPD/FIrpic/BPhen/LiF/Al, was of limited utility in this study due to significant voltage dependent electroluminescence (Figure S10), caused by electrons leaking into the NPD layer (as evidenced by the growth of the 450 nm band at high driving potentials). In order to confine electrons into the emitting/reacting FIrpic/BPhen interphase and generate excitons on the FIrpic molecules, Ir(ppz)3 (fac-tris-iridium(III)- (1-phenylpyrazole)) was introduced as electron blocking layer (EBL) between the NPD and FIrpic ones.29 Two device architectures were prepared, the first one having a planar junction configuration with the FIrpic layer adjacent to the BPhen one (device A in Figure 3); the second device having a bulk heterojunction layer of the two materials co-evaporated in 1 to 1 ratio and sandwiched between the neat layers of the two compounds (configuration B in Figure 3, namely devices B20 experiment carried out dispersing FIrpic and BPhen in PMMA (section 2). Taken together these results confirm that only in the presence of a macroscopic interface between FIrpic and BPhen (the bilayer heterojunction or the bulk heterojunction), the exciton induced degradation of FIrpic is quickly sidestepped by the efficient entrapment of coordinatively unsaturated Ir system by the BPhen molecule in close contact. Results in line with the solid state behavior (section 2) upon thermal annealing. Figure 3. Configuration of the devices with planar FIrpic-BPhen bulk hererojunction (device B) and FIrpic doped into mCP matrix (device C). (device A), interface 4. a) Figure A: Normalized ITO/NPD(50nm)/Ir(ppz)3(10nm)/FIrpic(20nm)/BPhen(50nm)/LiF(1nm) electroluminescence of the unaged device (red line with squares) and of the device driven at 160 mA/cm 2 for 50 minutes (blue line with circles) compared with the photoluminescence of FIrpic (black) and complex 1 (grey) in neat film; reduced intensity of electroluminescence versus time during aging. c) Current density- voltage curve of the unaged (black) and aged (green) device. the /Al(80nm). diagram Energy device for b) Figure 5. a) Energy diagram for the device B20: ITO/NPD(50nm)/Ir(ppz)3(10nm)/FIrpic(10nm)/ FIrpic:BPhen 50% mol (20nm)/BPhen(40nm)/LiF(1nm)/Al(70nm). b) Normalized electroluminescence of the unaged device (red line with squares) and of the device driven at 160 mA/cm2 for 50 minutes (blue line with circles) compared with the photoluminescence of FIrpic (black) and complex 1 (grey) in neat film, reduced intensity of electroluminescence versus time during aging. c) Current density-voltage curve of the unaged (black) and aged (green) device. 4. Morphological, spectroscopic and chemical analysis on OLED devices Although the EL signature in the aged devices A and B are in optimal agreement with the degradation-entrapment mechanism described, a direct evidence of the formation of 1 in aged OLEDs would be more assertive and conclusive. We therefore employed PL and ESI-MS analysis to study the aging processes. We fabricate at least two identical devices (for each configuration, A and B) in a single chamber load; one was kept as benchmark without being electrically aged while the second was aged according the described procedure. To mimic the FIrpic/BPhen interface an incomplete device A was built up to the FIrpic deposition, the morphology and topography of outermost layer (Figure S12) was analyzed by non-contact Atomic Force Microscopy (AFM). The surface appears, overall, quite flat with an average hill-to-valley roughness of 4 nm. Nonetheless the FIrpic layer is characterized by depressions where more than 60% of the mapped-area have depth greater that the average roughness. An hill-to-valley distances even exceeding 10 nm are found on the surface implying that the real FIrpic/BPhen interfacial contact area is much larger than the nominal pixel area. Full device A was thus studied by PL experiments applying on the device a positive mask with a 6.25mm2 aperture on the probed pixel. Without any change in device position and instrumental optical setup, we recorded the unaged pixel emission (excitation at 425 nm). The pixel was than aged according to the described protocol and PL spectrum was measured again. Although the EL changes as in Figure 4b, the PL spectrum is only slightly changed (Figure S13); indeed when we subtracted the normalized unaged pixel PL spectrum from that of its aged one, surprisingly we observed a bleach of the FIrpic emission at 460 nm and the raise of the emission of 1 at 550 nm. PL emission was also measured by applying to the aged device a bias voltage at which no EL can be detected, namely 5 V forward bias and 5 V reverse bias. The corresponding PL spectra are identical to the one in absence of bias, indicating that 1 is formed irreversibly at the FIrpic/BPhen interface. For the ESI-MS chemical analysis, pixels of device B20 were aged as described in section 3. The cathode was tape-peeled31 and the underneath organics of the pixels were dissolved in 1 drop of acetonitrile then diluted into 100 µL of the same solvent. At first, the samples were subject to a direct-infusion ESI-MS scan. The mass spectra of the unaged and aged devices (Figure S14) show overall very similar pattern in which all the chemicals used in the fabrication can be readily identified; the sample obtained from the aged device displays also few other signals, possibly due to the contaminations from the tape-pealing process instead of being correlated to the possible matrix alteration described elsewhere.32 Yet this analysis does not provide the striking evidence for the presence of complex 1 as the intensity of m/z = 905 is at the noise level). Figure 6. a) Energy diagram for the device C: ITO/NPD(50nm)/Ir(ppz)3(10nm)/FIrpic:mCP(6%wt, 30nm)/BPhen(40nm) /LiF(1nm)/ Al(75nm). b) Normalized electroluminescence of the unaged device (black) and of the device driven at 16 mA/cm2 for 50 minutes (blue) compared with the photoluminescence of FIrpic in deaerated dichloromethane solution (red); reduced intensity of electroluminescence versus time during aging at 16 mA/cm2 (filled squares) and 160 mA/cm2 (open circles). c) Current density-voltage curve of the unaged (black) and aged (green) device. In order to avoid possible artefacts due to the direct-infusion of a "multi-component" mixture and reduce the total ionic load of the MS-detector with compounds at concentration order of magnitudes higher than the one expected for 1 (see below),33 we developed a 20 minutes reverse-phase gradient HPLC method. This method allows to perfectly separate all the organic (BPhen and NPD) and organometallic (FIrpic, Ir(ppz)3) components in the OLED stacks as well as to elute complex 1 in a clean region of the chromatogram (at ca 9-10 min). In Figure S15 are reported the HPLC chromatograms of the compounds mixture as well as the ones of all the chemicals separately. We used the retention times to locate complex 1 in the HPLC profile and confirm its presence / absence in aged and unaged samples via MS analysis. The first HPLC- MS study performed on the device B20 was, in this case, of limited utility given the low S/N ratio and the presence of background m/z 905 signal; the whole (ESI)MS spectra extracted from the HPLC run at the retention time of complex 1 reveal in both the unaged and aged pixels (Figure S16) presence of variable amount of 1. A new attempt to prove the in-situ formation of 1 upon aging was done replicating the previous analysis on the prototype device B80. Given the four times thicker bulk heterojunction emissive layer, the formation of a significantly larger amount of 1 was expected. In addition the MS spectrometer was set mass isolation mode on m/z 905 in order to achieve the highest instrumental sensitivity. The mass chromatogram of the unaged pixels of the isolated m/z 905 signal gave an experimental intensity larger than expected for a "fresh" system. The integrated area at the retention time of ca 10 min, was 8.9·106 units; the corresponding mass peak showed the expected isotopic pattern for 1. Although surprising this finding is not unexpected, given the recent observation of Forrest et al..16 In the case of the aged B80 pixels, the integrated intensity of the same peak was as large as 15.1·106 units; in addition a new chromatographic peak at 10.8 min (integration 1.82·106 units) appeared with the same isotopic pattern of the peak at 10 min. It is worth pointing out that such signal, appearing only in the aged device and at slightly longer retention time, can be assigned to a "facial-like" isomer of 1, resulting from the loss of the trans-arrangements of the F2ppy ligands in the Ir fragment before the trapping event (figure S17 and 18).34 In order to finally establish whether the area of the mass chromatogram of 1 in the aged pixels is significantly larger than that of the same signal in the unaged device, we compared the NPD signal intensity on the integrated UV chromatograms for the two experiments. NPD is selected as internal standard because is the bottom layer in the device. In the sample preparation, it is dissolved after all the other over-layers are necessarily dissolved. The NPD area in unaged B80 is 238 arbitrary units while in the aged one is 290. In the assumption that the aging process is not affecting the elute composition, and set to 1 the intensity of the chromatographic peak of isolated 905 in unaged device, the m/z 905 peak in the aged one is 1.6 (table S2), that is considered a significantly higher value suitable to demonstrate the degradation -- trapping process proposed in this work. CONCLUSION In this study we have set up simple and selective trapping experiments proved to be effective in simulating the ex-situ and in-situ degradation of FIrpic. We developed a combination of mutually complementary, independent analytical techniques, including DSC, HPLC/MS and PL experiments easily accessible to any chemistry lab, to follow this process. We safely demonstrated that the results of these "chemical studies" are indeed well mirrored when suitable OLED prototypes are fabricated and aged. We investigated the reactivity of FIrpic in the ground state towards ligand substitution with BPhen by thermal reaction, as alternative to photo-induced ligand metathesis reaction. The formation of complex 1 was ascertained in solution and, most importantly, in solid state. This allowed to elucidate in a simple and easily reproducible way the occurrence of degradation- trapping in working OLED devices. aging, Upon electrical Two engineered OLED architectures, characterized by the presence of a planar or bulk interface between FIrpic and BPhen, were designed. the electroluminescence spectra evolve from the typical bluish emission of FIrpic to a broad orange color that fits with emission of complex 1. This color change perfectly matches the PL color evolution of a thin film FIrpic-BPhen blend upon thermal treatment. In these experiments, the raise of the orange band is indisputably associated with formation of complex 1 as proved by ESI-MS analysis. By means of PL difference spectra we were able to probe that on the aged pixel, it is possible to observe the raise of the emission of 1 in lieu of the bleach of the FIrpic one. Most importantly, applying HPLC-MS techniques with the isolation of the target molecular ion (m/z 905) it has been undisputable proved the formation of the complex 1 as the result of a trapping event between the degraded FIrpic and the BPhen trap. Interestingly in this case it was possible to discriminate in the MS chromatogram two peaks with 905 m/z belonging to two geometrical isomers We believe that these experiments have provided an easy reproducible framework to circumvent the problems linked to trace analysis of the formation of degradation product in OLEDs and can be applied even in a feedback scheme for a simple evaluation of the stability of new emissive dopants. It is worth mentioning that these results also point to the fact that if a host matrix is designed to efficiently trap the degradation intermediates into high energy derivative which will not funnel the excitonic energy of FIrpic, this would lead to a new paradigm to extend the device lifetime. EXPERIMENTAL SECTION Sublimated 4,7-diphenyl-1,10- phenanthroline (BPhen), N,N′-di[(1-naphthyl)-N,N grade from (mCP) ′-diphenyl]-1,1′-biphenyl)-4,4′-diamine (NPD) and N,N'-dicarbazolyl-3,5-benzene were purchased bis[2-(4,6- TCI, difluorophenyl)pyridyl- C2,N] (picolinato)iridium(III) (FIrpic) and LiF were purchased from Aldrich, Al pellets was purchased from Kurt J. Lesker Company. Fac-Ir(ppz)3 and complex were synthesized according to literature procedures.28 from commercially available 2-(2',4'- difluorophenylpyridine) and 1-phenylpyrazole purchased from TCI fac-Ir(ppz)3 was sublimated before use. All solvents are degassed under argon prior the use. [Ir(F2ppy)2(BPhen)]PF6, 1, NMR spectra were recorded at 300 K on Bruker AV400 and Bruker AC300 spectrometers. Chemical shifts are expressed in parts per million (ppm) and referred to the residual signal of the solvent (2.50 ppm in DMSO-d6).); coupling constants are given in Hz. Reverse phase gradient HPLC analysis were performed on Agilent 1100 equipped with VWD UV-Vis detector (set 300 nm), a Waters Atlantis column (4.6 mm x 10.0 cm, 3 μm particle size) and a mass spectrometer BRUKER Esquire 3000 PLUS with quadrupole ion-trap detector. HPLC analysis ran in a 20 min gradient elution, flow 0.5 ml/min starting from 70-30 acetonitrile -- water mixture up to 100% acetonitrile in the presence of 0.05% trifluoroacetic acid (TFA). Direct infusion electrospray ionization mass spectrometry (ESI- MS) experiments were carried out on the same mass spectrometer. The samples were prepared by diluting suitable stock solutions of the substrates in CH3CN up to an overall concentration of 10-6 M. The solutions were infused directly in the ESI source at 4 µL min-1. UV/Vis absorption spectra were obtained on Shimadzu UV-Vis-NIR Spectrophotometer in 1 cm path length quartz cell with dichloromethane. Photoluminescence quantum yields were measured with a C11347 Quantaurus - QY Absolute Photoluminescence Quantum Yield Spectrometer (Hamamatsu Photonics U.K), equipped with a 150 W Xenon lamp, an integrating sphere and a multi-channel detector. Photoluminescence emission spectra, lifetimes and electroluminescence spectra were obtained with a FLS 980 spectrofluorimeter (Edinburg Instrument Ltd.). Continuous excitation for the steady state measurements was provided by a 450 W Xenon arc lamp. Emission spectra were corrected the detector sensitivity. Photoluminescence lifetime measurements, determined by TCSPC (time-correlated single- photon counting) method, were performed using an Edinburgh Picosecond Pulsed Diode Laser EPL- 375 (Edinburg Instrument Ltd.), with central wavelength 374.0 nm, pulse width 66.2 ps and for L-1. repetition rates 50 μs. Photoluminescence experiments at room temperature were carried out in nitrogen degassed dichloromethane solution 2x10-5 mol Current-voltage measurements were obtained from AUTOLAB PGSTAT302N equipped with software NOVA 2.0. Thin films were prepared by spin coating on quartz substrate from dichloromethane solutions. Devices were made by physical vapor deposition (PVD) with a Kenosistec KE500KF vacuum chamber connecting to a glovebox. The base chamber pressure is 5 × 10-7 mbar. The chamber includes four effusive (Knudsen) sources and two thermal (Joule) sources. Two microbalances equipped with gold-coated crystal sensor Inficon quartz monitor allow to monitor the thickness of deposited material by the Inficon SQS-242 software. The crystal sensor was calibrated for each material by measuring the thickness of at least three evaporated films by means of a Bruker DektakXT contact profilometer. All the materials were deposited at pressures ≤ 2 × 10-6 mbar and deposition rates lower than 1 Å/s. Device were made on ITO-coated glass substrates purchased from KINTEC Company (width of patterned stripe 2.5 mm; thickness, 150 ± 10 nm; sheet resistance 15-20 Ω cm−2). ITO substrates were cleaned with detergent and water, sonicated in deionized water, acetone, isopropanol for 15 min each and exposed to an ozone atmosphere for 15 min before loading into the high-vacuum chamber. After organic depositions, LiF and Al layers were deposited by using a mask with 2.5 mm stripe width (placed onto substrates under N2). Each device consists of 4 independent pixels with area of 6.25 mm2. Non-contact AFM images were obtained with a 5500 Keysight system ASSOCIATED CONTENT Supporting Information. Synthesis of complex 1 and its characterization, photophysical properties of FIrpic and complex 1, ESI-MS spectra,, DSCs thermal studies and corresponding HPLC-MS analyses, optical spectrum of FIrpic neat films, device characterization for the unoptimized architecture without Ir(ppz)3 blocking layer, electrical features of the device B80, AFM morphological analysis of the FIrpic interface, photoluminescence characterizations of unaged and aged pixels of device A, direct infusion MS and HPLC-MS analysis of unaged and aged devices, calculated absorption spectra of FIrpic and 1. These materials are supplied as Supporting Information. "This material is available free of charge via the Internet at http://pubs.acs.org." AUTHOR INFORMATION Corresponding Author * Alberto Bossi, email: [email protected], ISTM-CNR via Fantoli 16/15, 20138 Milan, tel: +39 0250995627. Davide Ceresoli, email: [email protected], ISTM-CNR via Golgi 19, 20133 Milan, tel: +39 0250314276. Fausto Cargnoni, email: [email protected], ISTM-CNR via Golgi 19, 20133 Milan, tel: +39 0250314272. Present Addresses †Currently at Department of Chemistry, University of Milan, via Golgi 19, 20133 Milano. Author Contributions The manuscript was written through contributions of all authors. / All authors have given approval to the final version of the manuscript. Funding Sources The project has been founded by the Samsung GRO 2014 program "Exciton and polaron induced OLED degradation by combined ab-initio molecular dinamics and experiments" and the progetto Integrato Regione Lombardia and Fondazione CARIPLO (grant numbers 12689/13, 7959/13; Azione 1 e 2, "SmartMatLab centre" and Cariplo Foundation grant 2013-1766). ISTM-CNR ACKNOWLEDGMENT AB and MP thanks Dr. Alessio Orbelli Biroli and Dr. Daniele Marinotto of for the measurements of the film thickness of the organic materials used in OLEDs and Dr. Ivan Andreosso for some preliminary calibrations. Authors greatfully thanks Prof. Gianlorenzo Bussetti, Department of Physics of the Politecnico di Milano, for the AFM investigation and useful discussion. AFM were realized at the Solid-Liquid Interface and Nanomicroscopy (SoLINano) laboratory which is an inter-Departmental lab of the Politecnico di Milano REFERENCES 1. Muccini, M.; Toffanin, S. Organic Light-Emitting Transistors: Towards the Next Generation Display Wiley-Science Wise Co-Publication: Technology; Hoboken, NJ, 2016. 2. Yersin, H. Highly Efficient OLEDs with Phosphorescent Materials; Wiley-VCH: Weinheim, 2007. 3. Thompson, M. E.; Djurovich, P. E.; Barlow, S.; S. Marder, for Optoelectronic Comprehensive Organometallic Chemistry, ed. D. O'Hare, Elsevier: Oxford, 2007; 12, 101-194. 4. Huckaba, A. J.; Nazeeruddin, M. K. Strategies for Tuning Emission Energy in Phosphorescent Ir(III) Complexes. Comments Inorg. Chem. 2016, 37, 117- 145. Organometallic Complexes Applications, 5. Scholz, S.; Kondakov, D.; Lussem, B.; Leo, K. Degradation Mechanisms and Reactions in Organic Light-Emitting Devices. Chem. Rev. 2015, 115, 8449- 8503. 6. Schmidbauer, S.; Hohenleutner, A.; Konig, B. Chemical Degradation in Organic Light-Emitting Devices: Mechanisms and Implications for the Design of New Materials. Adv. Mater. 2013, 25, 2114-2129. 7. Yamamoto, H.; Weaver, M. S.; Murata, H.; Adachi, C.; Brown, J. J. Understanding extrinsic degradation in phosphorescent OLEDs. SID Int. Symp. Dig. Tec. 2014, 45, 758 -- 761. 8. Forrest, S. R. Ultrathin Organic Films Grown by Organic Molecular Beam Deposition and Related Techniques. Chem. Rev. 1997, 97, 1793-1896. 9. Spindler, P.; Hamer, J. W.; Kondakova, M. E. OLED Manufacturing Equipment and Methods, Handbook of Advanced Lighting Technology, Springer: Cham, 2017, 417-441. 10. Zhang, Y.; Aziz, H. Degradation Mechanisms in Blue Phosphorescent Organic Light- Emitting Devices by Exciton−Polaron Interactions: Loss in Quantum Yield versus Loss in Charge Balance. ACS Appl. Mater. Inter. 2017, 9, 636-643. 11. Kim, S.; Bae, H. J.; Park, S.; Kim, W.; Kim, J.; Kim, J. S.; Jung, Y.; Sul, S.; Ihn, S.-G.; Noh, C.; Kim, S.; You Y. Degradation of blue-phosphorescent organic light- emitting devices involves exciton-induced generation of polaron pair within emitting layers, Nat. Comm. 2018, 9, 1211. 12. Kondakov, D. Y.; Lenhart, W. C.; Nichols, W. F. Operational degradation of organic light-emitting diodes: Mechanism and identification of chemical products. J. Appl. Phys. 2007, 101, 024512. 13. Sivasubramaniam, V.; Brodkorb, F.; Hanning, S.; Loebl, H. P.; van Elsbergen, V.; Boerner, H.; Scherf, U.; Kreyenschmidt, M. Investigation of FIrpic in PhOLEDs via LC/MS technique. Cent. Eur. J. Chem. 2009, 7, 836- 845. 14. Scholz, S.; Meerheim, R.; Lussem, B.; Leo, K. Laser desorption/ionization time-of-flight mass spectrometry: A predictive tool for the lifetime of organic light emitting devices. Appl. Phys. Lett. 2009, 94, 043314. 15. Scholz, S.; Walzer, K.; Leo, K. Analysis of Complete Organic Semiconductor Devices by Laser Mass Desorption/Ionization Spectrometry. Adv. Funct. Mater. 2008, 18, 2541-2547. 16. Jeong, C; Coburn, C.; Idris, M.; Li, Y.; Djurovich, P. I.; Thompson, M. E.; Forrest, S. R. Understanding molecular fragmentation in blue phosphorescent organic light-emitting devices. Org. Electron. 2019, 64,15 -- 21. 17. Meerheim, R.; Scholz, S.; Olthof, S.; Schwartz, G.; Reineke, S.; Walzer, K.; Leo, K. Influence of charge balance and exciton distribution on efficiency and lifetime of phosphorescent organic light-emitting devices. J. Appl. Phys. 2008, 104, 014510-014518. 18. de Moraes, I. R.; Scholz, S.; Lussem, B.; Leo, K. Role of oxygen-bonds in the degradation process of phosphorescent organic light emitting diodes. Appl. Phys. Lett. 2011, 99, 053302-053303. 19. de Moraes, I. R.; Scholz, S.; Leo, K. Influence of the applied charge on the electro-chemical degradation in green phosphorescent organic light emitting diodes. Org. Electron. 2016, 38 164-171. Time-of-Flight Proc. SPIE 2008, iridium complexes: 20. de Moraes, I. R.; Scholz, S.; Lussem, B.; Leo, K. Analysis of chemical degradation mechanism within sky blue phosphorescent organic light emitting diodes by laser-desorption/ionization time-of-flight mass spectrometry. Org. Electron. 2011, 12, 341-347. 21. Scholz, S.; Meerheim, R.; Walzer, K.; Leo, K. Chemical degradation mechanisms of organic 6999, semiconductor devices. 69991B. 22. Ragni, R.; Plummer, E. A.; Brunner, K.; Hofstraat, J. W.; Babudri, F.; Farinola, G. M.; Naso, F.; De Cola, L. synthesis, Blue emitting photophysics and phosphorescent devices. J. Mater. Chem. 2006, 16, 1161-1170. 23. Jurow, M. J.; Bossi, A.; Djurovich, P. I.; Thompson, M. E. In Situ Observation of Degradation by Ligand Substitution in Small-Molecule Phosphorescent Organic Light-Emitting Diodes. Chem. Mater. 2014, 26, 6578- 6584. 24. Wang, Q.; Aziz, H. Exciton -- Polaron-Induced Aggregation of Organic Electroluminescent Materials: A Major Degradation Mechanism in Wide-Bandgap Phosphorescent and Fluorescent Organic Light-Emitting Devices. Adv. Optical Mater. 2015, 3, 967-975. 25. Baranoff, E.; Curchod, B. F. E. FIrpic: archetypal blue phosphorescent emitter for electroluminescence. Dalton Trans. 2015, 44, 8318-8329. 26. Sivasubramaniam, V.; Brodkorb, F.; Hanning, S.; Loebl, H. P.; van Elsbergen, V.; Boerner, H.; Scherf, U.; Kreyenschmidt, M. Fluorine cleavage of the light blue heteroleptic triplet emitter FIrpic. J. Fluorine Chem. 2009, 130, 640-649. 27. Cazzaniga, M.; Cargnoni, F.; Penconi, M.; Bossi, A.; Ceresoli, D. Unraveling the degradation mechanism cells using in FIrpic based blue OLEDs: I. A theoretical investigation. submitted to Chem. Mat. this issues. 28. Sunesh, C. D.; Sunseong, O.; Chandran, M.; Moon, D.; Choe, Y. Effect of ionic liquids on the electroluminescence of yellowish-green light emitting electrochemical bis(2-(2,4- difluorophenyl)pyridine)4,7-diphenyl-1,10- phenanthroline-iridium(III) hexafluorophosphate. Mat. Chem. Phys. 2012, 136, 173-178. 29. Adamovich, V. I.; Cordero, S. R.; Djurovich, P. I.; Tamayo, A.; Thompson, M. E.; D'Andrade, B. W.; Forrest, S. R. Org. Electron. 2003, 4, 77-87. 30. Holmes, R. J.; Forrest, S. R.; Tung, Y. J.; Kwong, R. C.; Brown, J. J.; Garon, S.; Thompson, M. E. Blue organic electrophosphorescence using exothermic host -- guest energy transfer. Appl. Phys. Lett. 2003, 82, 2422-2424. 31. Aziz, H.; Popovic, Z.; Tripp, C. P.; Hu, N.-X.; Hor, A.-M.; Xu, G. Degradation processes at the cathode/organic interface in organic light emitting devices with Mg:Ag cathodes. Appl. Phys. Lett. 1998, 72, 2642-2644. 32. de Moraes, I. R.; Scholz, S.; Lussem, B.; Leo, K. Chemical degradation processes of highly stable red phosphorescent organic light emitting diodes. Org. Electron. 2012, 13, 1900-1907. 33. In an oversimplified scheme, approximately the amount of 1 in device B, based on the volume of the pixel emitting layer and considering a reasonable estimate of the lower limit for FIrpic degradation of about 5% of molecules, yields less than 5x10-8 M (0.03 µg/ml) of complex 1. 34. Note: in ACN-H2O reverse phase HPLC, Fac-Mer mixture of tris-cyclometalated Ir complexes (homo or heteroleptic), the Mer isomer is eluted a shorted time than the corresponding Fac one. SYNOPSIS TOC Exciton induced degradation of FIrpic emitter in Blue-OLEDs. Paramount importance of addressing the chemical stability of the molecules both ex-situ and in-situ. 13
1912.03601	1	1912	2019-12-08T03:02:46	Feasibility Study of a Laser-Based Approach for Diagnosing Deuterium Neutrals in the Edge of Fusion Devices	[ "physics.app-ph", "nucl-ex", "physics.plasm-ph" ]	In magnetically-confined plasmas of tokamaks, neutral deuterium/hydrogen (D/H) atoms play a role in energy, momentum, and particle balance, as well as the stabilization of plasma turbulence. One key important fusion performance parameter is the pedestal density. Understanding the pedestal density formation is critical for the development of predictive model of future fusion devices. Typically, measurements of the neutrals are obtained using optical emission spectroscopy of the Lyman alpha lines, which is a line-integrated measurement. The plasma in tokamaks is characterized by a high density of electrons and ions and a relatively low concentration of neutral hydrogen atoms, which could make direct measurement of density seemingly impossible at first. We propose a laser-based method that allows for accurate measurement of both the spatial and absolute magnitude of the neutral D/H with minimal knowledge of the radial profiles of electron temperatures and densities. This relies on the fact that the neutral spectral profile can have a larger peak than the electron spectral profile and thus make the neutral density signal resolvable. In practice, this method can be co-located with Thomson scattering systems and is referred to as laser Rayleigh scattering (LRS). More specifically, we assess and evaluate the LRS method for two test cases: in the midplane radii of the National Spherical Torus Experiment (NSTX), and in the small angle slot divertor configuration of DIII-D. Preliminary simulations and calculations will determine the feasibility of LRS in the presence of incoherent Thomson scattering under neutral densities ranging from $ 10^{13} $ to $ 10^{21} \ \text{m}^{-3} $. Wavelength dependence of LRS will be evaluated to determine the boost in the signal and photon generation capability.	physics.app-ph	physics	Feasibility Study of a Laser-Based Approach for Diagnosing Deuterium Neutrals in the Edge of Fusion Devices David Feng1, Ahmed Diallo2, and Mikhail N. Shneider1,2 1Department of Mechanical & Aerospace Engineering, Princeton University, Princeton, 2Princeton Plasma Physics Laboratory, Princeton, NJ 08540 NJ 08544 December 10, 2019 1 Motivation and Goal In magnetically-confined plasmas of tokamaks, neutral deuterium/hydrogen (D/H) atoms play a role in energy, momentum, and particle balance, as well as the stabilization of plasma turbulence. One key important fusion performance parameter is the pedestal density. Understanding the pedestal density formation is critical for the development of predictive model of future fusion devices. Typically, mea- surements of the neutrals are obtained using optical emission spectroscopy of the Lyman alpha lines, which is a line-integrated measurement. The plasma in tokamaks is characterized by a high density of electrons and ions and a relatively low concentration of neutral hydrogen atoms, which could make direct measurement of density seemingly impossible at first. We propose a laser-based method that allows for accurate measurement of both the spatial and abso- lute magnitude of the neutral D/H with minimal knowledge of the radial profiles of electron temperatures and densities. This relies on the fact that the neutral spectral profile can have a larger peak than the electron spectral profile and thus make the neutral density signal resolvable. In practice, this method can be co-located with Thomson scattering systems and is referred to as laser Rayleigh scattering (LRS). More specifically, we assess and evaluate the LRS method for two test cases: in the midplane radii of the National Spherical Torus Experiment (NSTX), and in the small angle slot divertor configuration of DIII-D. Preliminary simulations and calculations will determine the feasibility of LRS in the presence of incoherent Thomson scattering under neutral densities ranging from 1013 to 1021 m−3. Wavelength de- pendence of LRS will be evaluated to determine the boost in the signal and photon generation capability. 2 The Rayleigh Scattering Diagnostic: A Brief Overview LRS is the elastic scattering of light by neutral atoms and molecules much smaller than the wavelength In the classical sense, it is the electromagnetic radiation emitted by an of the incident laser light. oscillating dipole moment (the atom/molecule) that is induced by an incoming electric field (the laser): p = α E(r, t) (2.1) The polarizability α characterizes the electric response specific to an individual class of scatterers, such as atoms, molecules, and even ions. It is a tensor quantity that can be simplified and reduced depending on the shape and orientation of the molecule. If the molecule is far away from resonance, such that the α is not affected by the frequency of the external electric field, it is called the static polarizability. α2 mean = 1 9 (αxx + αyy + αzz)2 (2.2) 1 For linear diatomic molecules, the square of effective polarizability averaged over all molecular orien- tations is [5]: where α2 mean = 45α2 + 10γ2 45 α2 = 1 9 (αxx + 2αyy)2 γ2 = (αxx − αyy)2 (2.3) (2.4) (2.5) X is the axis of symmetry of the molecule, Y and Z are normal to it, and the polarizabilities in the Y and Z directions are equal. If the scatterer is stationary, the emitted radiation is phase-locked to the incident light and the result is a coherent sum of the field plus all other nearby fields. The radiation cancels in all but the forward direction. However, molecular motion in the gas leads to microscopic fluctuations in the gas density that randomize the phase of the emitted radiation and causes it to be incoherent in all but the forward direction. Away from the forward direction, the rapidly changing interference averages out to become proportional to the number of scatterers in the gas (i.e. the number density). The average power of this emitted radiation (i.e. the total scattered average power) is given by: P = nV ILσRS (2.6) where N is the number density of the gas, V is the volume of scatterers, IL is the initial laser intensity, and σRS is the Rayleigh scattering (RS) cross-section of a single scatterer (or species). A more thorough [5]. When LRS is used for diagnostics (e.g. single or description of LRS can be found in Miles et al. multiple photodetectors), it is more practical to cast Equation 2.6 in differential form per unit solid angle Ω: where dΩσRS is the differential RS cross-section. It can be related to α via: dΩP = nV ILdΩσRS dΩσRS = α2 Vk4 L sin2 φ (2.7) (2.8) where αV = α/4πε0, kL is the laser wavevector, and φ is the azimuthal angle of the dipole radiation pattern. The total RS power is therefore: PRS = ηnV ILZ∆Ω dΩσRS dΩ (2.9) after multiplying by the appropriate optical, collection, and detector efficiencies η. It is evident that LRS can be used to determine aggregate number densities of the gas-phase system, including those of near-wall tokamak devices. Note that for multiple species present within the gas-phase system, σRS becomes the weighted cross-section by mole fraction Xi: For species of interest in nuclear fusion devices (i.e. typically species generated in the outer midplane regions of tokamaks), the table of polarizabilities is given by: σRS =Xi Xiσi (2.10) 2 Species Mean polarizability (A3) Ref. Note H 0.6668 H2 0.7894 H + 2 0.4696 [7] [4] [2] D 0.7587 CCCDB D2 0.7954 D + 2 0.4552 [4] [2] Calculation compared with previous data in close agreement. Calculation for the rotational quantum no. j = 0 in the electronic ground state. Calculation for the 1Se v = 0 state. Calculation via DFT and the basis set aug-cc-pVTZ. No other data exists in the literature as far as the authors know. Calculation for the rotational quantum no. j = 0 in the electronic ground state. Calculation for the 1Se v = 0 state. Table 1: Static mean polarizabilities for the species considered in the preliminary study, along with their references. CCCDB stands for the computational chemistry comparison and benchmark database operated by the National Institute for Standards and Technology (NIST). One major disadvantage with LRS is that background (BG) scattering from windows, surfaces, and Mie scattering can override the LRS signal due to typically being orders of magnitude stronger. The Mie and dust scattering is random and can be removed via post-processing, but the window and surface reflections will prevent the LRS from being visible, and if the incident laser beam or sheet is too close to a surface, then the desired image may not be properly recorded. However, since the BG scattering occurs at the laser wavelength and is very narrow in linewidth, it can be properly attenuated using an iodine filter, which has absorption bands at 532 nm. The combination of a very narrow linewidth laser and an atomic/molecular notch filter is known as filtered Rayleigh scattering (FRS), and is described in Miles et al. [6]. Another disadvantage particular to plasma environments is the presence of Thomson scattering (TS). If the laser wavelength is shorter than the Debye length of the plasma, the scattering from electrons is randomly distributed and the scattering is said to be incoherent. For most plasma conditions and mea- surement geometries, the electron scattering is incoherent. The TS cross-section is typically a few orders of magnitude greater than the RS cross-section for atmospheric and hydrogenic species. However, the TS spectrum is also extremely broad compared to the RS spectrum, often known as the Rayleigh-Brillouin scattering (RBS) spectrum. The TS spectrum is often a few nanometers wide, whereas the RBS spec- trum is only a few to tens of GHz wide. Peak values of the signal close to the laser frequency produced by the neutral species can therefore be resolved and detected, and will be the focus of the next section. Electrons and neutrals co-exist in the same environment inside tokamaks, so the LRS will also be coupled to the laser Thomson scattering (LTS). The total scattering cross-section of the electron is given exactly by: σTS = 8π 3 (cid:18) e2 4πε0mc2(cid:19)2 (2.11) which leads to a value approximately equal to 6.65 × 10−29 m2. In comparison, hydrogen has a total RS cross-section of about 7.25 × 10−32 m2 at a wavelength of 532 nm. The total cross-section for electrons is three orders of magnitude stronger than that of H, and via Eqn. 2.6 it would initially seem that four orders of magnitude more number density would be required for the RS of H to become compa- rable at the very least, e.g. an electron number density of 1015 m−3 gives a H number density of 1018 m−3. 3 To overcome this fundamental limitation, consider the scattered spectra. Near the laser frequency, at a given electron temperature the peak of the RS spectrum can be equal to or higher for comparable electron and neutral number densities. In the following study, we assume that the RS scattering remains in the kinetic collisionless regime, determined through the y parameter: y = 1 ksλMFP = nσc ks ∝ P · λL T (2.12) where ν is the laser frequency and ks = 4π sin(θ/2)/λL is the scattering wavevector. When y ≪ 1, the RS is in the collisionless kinetic regime, i.e. the Boltzmann equation with no collisional term is used to describe the gas density perturbation, which is approximately Gaussian in nature. Therefore we model the neutral species' spectra as a Gaussian lineshape: where ∆νth is the thermal velocity: gRS(ν) = ks/2π ∆vth · e(ν−ν0)2/∆v2 th ∆vth =s 2kBT µmol (2.13) (2.14) Note that the lineshape function is normalized such that R gRS(ν) dν = 1. The electron or TS lineshape is modeled using the Salpeter approximation, where the plasma spectrum takes a relatively simple form with both an ion and electron component: gTS(ω) = 1 − 2π ks(cid:12)(cid:12)(cid:12) 2 fi(cid:18) ω ks(cid:19) (2.15) 2 fe(cid:18) ω ks(cid:19) + χe ε (cid:12)(cid:12)(cid:12) 2πZ ks (cid:12)(cid:12)(cid:12) χe ε (cid:12)(cid:12)(cid:12) where ω is the angular frequency, fe, fi are the normalized one-dimensional electron and ion velocity distribution functions, χe, χi are the electron and ion susceptibilities, ε is the dielectric function, and Z is the charge of the ion. Typically, the ion component may be neglected. The approximation assumes a Maxwellian distribution function for the thermal velocities of both the electron and the ion. Through fe,i and χe,i it is also slightly dependent on the plasma scattering parameter α: α = 1 ksλDeb (2.16) where λDeb = pεkBTe/(nee2) is the Debye length of the plasma. For α ≪ 1, we have incoherent scattering and the lineshape reflects the distributions of the electrons according to Eqn. 2.15. 3 Preliminary Studies: Midplane NSTX Consider the inferred deuterium atom and molecule midplane density profiles for the National Spher- ical Torus Experiment (NSTX) in Figure 2 [8]. The region between R = 1.4-1.5 m marks the transition region where the molecular densities drop off sharply but the atomic density tails off slower as a result of dissociation due to collisions with ions and electrons. 4 Figure 1: Midplane density profiles for atomic and molecular deuterium for NSTX shots 139412 and 142214. Electron temperature profiles are also provided for reference. Figure is obtained from [8]. To determine the capability of distinguishing LRS from LTS, relevant neutral densities in this transi- tion region were modeled for various electron densities near the laser frequency, i.e. near the peak of the RS spectra. All the following spectra and data in this section will be for the laser wavelength λ = 532 nm. Consider the midplane slice at R = 1.4 m. Although computational data, derived from ab initio methods, for the polarizability of atomic deuterium (D) exists, it is not reliable and uncorroborated with experimental data or other literature sources; furthermore the same database offers two values which are highly disparate in value. Therefore H will be used in lieu of D. Figure 2: TS spectra at ne = 1 × 1015 m−3 for multiple electron temperatures. The integrated spectra is normalized to unity. Figure 2 demonstrates how the TS spectrum changes with electron temperature. As the tempera- 5 ture rises, the spectrum broadens out because the Maxwellian distributions of the electron (assuming thermodynamic equilibrium) have an inverse proportionality to the thermal speed: f =(cid:18) 1 πv2 th(cid:19)1/2 exp (−v2/v2 th) where vth =r 2kBTe me (3.1) (3.2) Therefore as the electron temperature Te grows, the distribution function flattens out because of the negative exponential dependence. More importantly, the peak of the TS spectral distribution also drops in order to preserve unity of the spectral integration, which is crucial to observing and resolving the LRS as follows. Figure 3: Multiple RS spectra (solid lines) in the presence of TS plasma backgrounds (dashed lines) at ne = 1 × 1015 m−3 for multiple electron temperatures for H. The peak at nneu = 5 × 1014 m−3 is greater than the TS background, which can be detected if the TS is filtered out (see Figure 7). Figure 3 shows RS spectra for H for up to nneu = 5 × 1014m−3 in the presence of TS background plasma with electron density ne = 1 × 1015m−3 and electron temperature Te = 200 − 400 eV. All spectra are scaled with respect to ne: gsc(ν) = Ineu IBG gRS(ν − ν0; T ; θ) (3.3) where Ineu/IBG is the scaling factor for the RS spectra. As Te increases, the BG decreases because the TS spectrum is getting broader, and thus the peak must correspondingly flatten if the lineshape is to remain normalized. Note that the TS spectrum is significantly much broader than the RS spectrum, nm compared to pm, respectively. At nneu = 5 × 1014m−3, the peak is considerably above the BG, and it is clear that values below nneu = 5 × 1014m−3 will be buried beneath it. Figure 4-5 show the remaining profiles for the considered species at electron temperatures and neutral densities corresponding to R = 1.4-1.5 m up for ne = 1-5 × 1017 m−3. Note that for H and the molecular cations, the RS peak is unresolvable because their polarizabilities (and therefore their cross-sections) are lower relative to the other molecules', which results in a lower scaling factor according to Eqn. 3.3. 6 Figure 4: RS spectra (solid lines) for multiple species at neutral densities ranging from 1 × 1013 m−3 to 5 × 1017 m−3 in the presence of plasma backgrounds (dashed lines) at ne = 1 × 1017 m−3. Each spectra corresponds to the species listed above it. Figure 5: RS spectra (solid lines) for multiple species at neutral densities ranging from 1 × 1013 m−3 to 5 × 1017 m−3 in the presence of plasma backgrounds (dashed lines) at ne = 5 × 1017 m−3. The volumetric photon emission Φ can also be calculated as follows: where the units are photons per unit volume per unit time. Figure 6 shows Φ for the experimental configuration EL = 0.5 J, τP = 10 ns, λ = 532 nm, at right angles for both the azimuthal and polar Φ = P hν · V (3.4) 7 angle, a distance to the photodetector 1 m, and radius of solid angle 1 cm, over neutral densities from 1014 to 1017 m−3. The power is integrated over the spectral domain shown in the previous figures. 1016 ) c e s 3 m ( / s n o t o h P 1014 1012 1010 0 H H2 H2+ D2 D2+ 1 2 3 4 Neutral density / m-3 5 1017 Figure 6: Volumetric photon emission of multiple species ranging from neutral densities 1 × 1013 m−3 to 5 × 1017 m−3 for ne = 1 × 1017 m−3. For a beam waist of wL = 1 mm, this gives an area A = (πw2 L)/4 ≈ 7.854 × 10−7 m2 and a Rayleigh range zR = (πw2 L)/λ ≈ 0.059 m, and thus an interrogation volume of V = A · 2zR ≈ 9.276 × 10−6 m3. For the neutral density on Figure 4 (approximately ≈ 1017 m−3), this gives a range of about 1.616 × 1010 to 4.859 × 1010 photons per second; for τp = 8 ns, this is approximately 3.89 × 102 photons per pulse. For neutral densities below it, it becomes extremely challenging to generate photons. In comparison, an electron density of about 1 × 1017 m−3 leads to a photon emission of 5.329 × 1013 photons per second, or 4.263 × 105 photons per pulse for the same ns pulse, which is three orders higher than for the largest neutral density in Figure 6. When integrating across the entire spectral domain, it would seem that Φ from the TS is greater than that of the RS, a threshold below which would make the RS signal undetectable. The idea would then be to reduce the bandwidth and focus over a few tens of GHz according to the linewidth of the RS. Consider a bandpass filter narrow enough to eliminate the transmission across the entire spectrum except for the domain ν = [−1, 1] THz as shown in Figure 7. 8 10-13 . U A . / e d u t i l p m A 6 5 4 3 2 1 0 Bandpass filter Rayleigh scattering Thomson scattering -500 0 500 Detuning / GHz 1 0.8 0.6 0.4 0.2 0 i i n o s s m s n a r T Figure 7: Representation of bandpass cutoff (orange line) of TS (dashed line), leaving primarily the RS (yellow line). The fraction of the scattered power is the integrated signal inside the bandpass region. By integrating over this region only, for some number densities we are able to retrieve a greater fraction of the photons coming from the RS than the TS scattering, and thus a detectable RS signal to the photodetector. The result is shown in Figure 8. ) c e s 3 m ( / s n o t o h P 7 6 5 4 3 2 1 0 1016 0 50 100 150 200 250 300 350 400 Electron temperature / eV Figure 8: Filtered-out TS powers based on the bandpass filter region (∆ν = ±1 THz) in Figure 7. Electron density is ne = 1 × 1017m−3. At electron temperatures above 50 eV, the filtered-out TS photon emission is reduced to the same order of magnitude as nneu ≈ 5 × 1017m−3 and lower in magnitude for the cases of H, H2, and D2. If the bandpass domain could be reduced by one order of magnitude further (e.g. ν = [−100, 100] GHz, or approximately λ = [−100, 100] pm) then the TS photon emission can be further reduced by one or- der of magnitude. It will be a challenge as this requirement pushes the limits of bandpass filter technology. 9 4 Preliminary Studies: Small Angle Slot Divertor The concept of a divertor to efficiently dissipate heat from tokamaks is critical for high performance (H-mode) operation. The divertor is situated along the magnetic field lines of the tokamak and "divert" the plasma away from other walls which could be eroded by its interaction with the high temperature plasmas. To achieve near-zero erosion at the divertor surface, the divertor plasma temperature must be kept cool by diffusing gas at the build-up regions, or the slots, in the divertor. An accurate dtermina- tion of the neutral profile in the divertor will provide key pararemeters for an accurate modeling of the radiation in future fusion devices. The following preliminary results will focus on the neutral density conditions of the small angle slot (SAS) divertor proposed in Guo et al. [1]. The most important part of the SAS divertor is its small angle at the outboard side end of the divertor slot, where electron temperatures must be kept sufficiently cool. These electron temperatures exhibit a close correlation to the D2 neutral density distributions as shown in Figure 9. Figure 9: Te and D2 distributions in the SAS, slant, and flat-end divertors. Figure is obtained from [1]. Neutral densities are significantly higher than those in the outer midplane of the NSTX by up to six orders of magnitude. Such densities are practical for LRS and can be easily commensurate, assuming the majority of the TS background can be filtered out. Figure 10 shows the spectra of D2 for λ = 532 nm, ne = 1.4 × 1019 m−3 and Te = 10, 30, 50 eV. Spectra are all calibrated to the electron density. As shown, the D2 spectra are well above the TS background. 10 Figure 10: RS spectra (solid lines) for D2 at neutral densities ranging from 1 × 1020 m3 to 1 × 1021 m3. The TS background spectra (dashed lines) are much lower than the RS spectra over the given spectral region. 1020 ) c e s 3 m ( / s n o t o h P 1019 1018 1 D2 2 3 4 5 6 7 8 9 Neutral density / m-3 10 1020 Figure 11: Volumetric photon emission for the neutral densities of Figure 10. Figure 11 shows the corresponding volumetric photon emission for EL = 0.5 J, τp = 8 ns, and wL = 1 mm. A neutral density of nneu = 1 × 1021 m−3 gives 9.359 × 1013 photons per second (assuming V = A · b where b = 2zR is the confocal parameter), and thus 7.488 × 105 photons per pulse. In comparison, an electron density of ne = 1.4 × 1019 gives 7.461 × 1015 photons per second without filtering, 3.626 × 1013 photons per second with a ∆ν = [−100, 100] GHz bandpass filter. A neutral density measurement can therefore be achieved in the SAS divertor concept provided the TS can be filtered out at the wings. LRS results may also be mixed for the slant and flat slot divertors in Figure 9 since their electron densities and neutral densities each does not vary by more than one order of magnitude from the SAS configuration. 5 Other Important Diagnostic Considerations From Eqn. 2.8, we can significantly increase the RS power by decreasing the wavelength of the laser via the cross-section: 11 dΩσRS ∼ k4 ∼ 1 λ4 (5.1) Consider the harmonic generation of an Nd:YAG laser: 1064 (fundamental), 532 (second harmonic), and 266 nm (fourth harmonic). The increase in RS power from 1064 to 532 nm is 16x, and from 532 to 266 is 16x, resulting in an overall 256x increase in total RS cross-section from the fundamental down to the fourth harmonic. However, the amount of energy per pulse is also decreased, typically by 70-80% of the fundamental at the fourth harmonic. Figure 12 compares spectra of D2 for the three wavelengths at the conditions for R = 1.4 m in the NSTX with ne = 1015 m−3. At λ = 1064 nm, the RBS spectra is completely buried underneath the TS spectra. At λ = 532 nm, the spectrum at nneu = 1 × 1014 m−3 barely peaks above the TS background, and at λ = 266 nm the RBS spectra is significantly above it. Figure 12: RS spectra (solid lines in TS background spectra (dashed lines) for D2 and ne = 1015 m−3. The photon volumetric emission Φ for the neutrals corresponding to Figure 12 and electron is shown in Figure 13. The energy per pulse for λ = 1064, 532, 266 nm is E = 1, 0.5, 0.3 J, respectively. The width of the filter to suppress the TS is 100 GHz centered at the laser center. Note that Φneu at λ = 266 nm is about two orders of magnitude greater than that of 1064 nm, while Φe suffers because it is only dependent on the pulse energy and not the wavelength. Using the parameters λ = 266 nm, wL = 1 mm, and thus V = A · b ≈ 1.855 × 10−5 m3, we obtain about 2.994 × 108 photons per second for nneu = 5 × 1014m−3; for a pulse width of 8 ns, this corresponds to about 2.4 photons per pulse. Calculations were also done for wL = 10 mm which resulted in about 100 times more photons per pulse. A larger beam parameter is therefore more advantageous and desirable assuming the entire confocal length region can be captured, although it is still a very small number of photons to collect, making the NSTX midplane region a highly challenging if not impossible task to do LRS. 12 1013 ) c e s 3 m ( / i i n o s s m e n o o h P t 1012 1011 0 1015 1014 1064 nm 532 nm 266 nm ) c e s 3 m ( / i i n o s s m e n o o h P t 1064 nm 532 nm 266 nm 1 2 3 4 Neutral density / m-3 5 1014 1013 200 250 300 350 400 Electron temperature / eV Figure 13: Volumetric photon emission for Ds (left) and electron (right) at multiple wavelengths of an Nd:YAG laser. Neutral density and electron temperature ranges mirror those of the NSTX midplane profiles (see Figure 1). Similarly, Φ is shown for divertor conditions in Figure 14. The neutral densities are higher by an order of magnitude of about 6-7, and thus a greater photon emission by approximately the same order of magnitude. This results in about 4.791 × 106 photons per pulse at the divertor. Even for interrogation lengths which are restricted due to detector size or collection optics, they would pose a decrease of at most one to two orders of magnitude. 1020 ) c e s 3 m 1019 1064 nm 532 nm 266 nm 1020 ) c e s 3 m 1019 1064 nm 532 nm 266 nm i i ( / n o s s m e n o t o h P 1018 1017 0 2 4 6 8 Neutral density / m-3 10 1020 i i ( / n o s s m e n o t o h P 1018 1017 0 50 100 150 200 250 300 350 400 Electron temperature / eV Figure 14: Volumetric photon emission for Ds (left) and electron (right) at multiple wavelengths of an Nd:YAG laser. Neutral density and electron temperature ranges mirror those of the SAS divertor (see Figure 9). Below is a table that summarizes the photons generated for the various neutral densities for D2. Note that the number of photons per pulse is directly proportional to the neutral density. 13 Neutral density (m−3) Photons per pulse 1014 1017 1019 1020 1021 0.479 4.791 × 104 4.791 × 106 4.791 × 107 4.791 × 108 Table 2: Summary of photons generated for various neutral number densities from D2. The conditions used are EL = 0.3 mJ, τp = 8 ns, wL = 1 cm, λ = 266 nm. Length is taken as the confocal beam parameter. The main consideration for wavelength will be the laser source. An Nd:YAG pulsed laser serves as the ideal laser system for this detection scheme due to its high peak powers in the visible and robust, long- term performance. Furthermore, it can be seeded or injection-locked for greater spectral purity of the laser wavelength; this is necessary for filtered Rayleigh scattering, which can reduce the highly narrow and much typically much stronger (than the RS) background scattering at the central wavelength. Overall, these aspects have made the Nd:YAG laser the workhorse of many LRS diagnostics and experiments, evidenced by the literature on LRS and its variations. However, output of the fourth harmonic (λL = 266 nm) is a different consideration, and modern commercial Nd:YAG lasers provide only around 100-250 mJ of fourth harmonic generation (see Spectraphysics Quanta-Ray Pro series and Surelite II series). Fourth harmonic lasers with up to 400-500 mJ have also been demonstrated in the laboratory [9, 10]. Although the energy change in comparison to wavelength is less consequential considering it is only around one order of magnitude, it is important for ensuring there are enough photons to collect in the measurement. 6 Two-color Scattering Approach Despite the reduction in laser energy/intensity due to nonlinear generation, it is clear that a multi- color approach towards shorter wavelengths offers an advantage for LRS in plasma environments, both for enhancing the LRS detection signal and distinguishing between LRS and LTS, if there is ever a need to measure both densities. The way to do this would be via a two-color scattering approach, which has been done for a laser spark shockwave in [3], and is described as follows. In some cases, it is not possible to determine from the measurement alone whether the photodetector signal primarily comes from LRS or LTS. However, as noted previously, the TS cross-section is indepen- dent of λL whereas the RS cross-section has a λ−4 L dependence. Comparison of two different wavelengths provides a way to discriminate between the two, providing there are sufficient photons for scattering from each. This is known as two-color scattering. The total signal for one color is: S(λ) = nneuσRS(λ) + neσTS n0σRS(λ) (6.1) All quantities are known except for nneu, ne. Two colors give two equations with two unknowns, thereby providing a determinate system. The denominator of Eqn. 6.1 is the reference signal, typically in atmospheric conditions. 7 Summary and Discussion Summary: Using laser Rayleigh scattering for neutral measurements in tokamaks has long been known to be challenging. Simulations show that the Rayleigh peak can be resolved under relatively com- parable electron densities from 1015 to 1019 m−3 and low to moderate electron temperatures (consistent with tokamak divertor temperatures). Calculations for NSTX midplane profiles (nneu = 1013-1017 m−3) imply photon generation per pulse to be virtually zero at λL = 532 nm and nonzero yet still "counting photons" at 266 nm (fourth harmonic of Nd:YAG laser). Moreover, for the same calculations at the divertor region (nneu = 1020-1021 m−3), we obtain approximately 106 photons per pulse. Going down 14 to the ultraviolet regime is crucial for generating more RS photons since the configuration at standard frequency-doubled Nd:YAG wavelengths is starved at the neutral density ranges of up to 1021 m−3. Narrow bandpass filters that can limit passing regions of up to [−1, 1] THz will need to be implemented, and possibly even narrower (hundreds of GHz, or hundreds of pm), in order to filter out at least one to two orders of magnitude of Thomson scattering. Lastly, provided the RS power is significant enough, a two-color scattering approach can be implemented to separate the RS from the TS. Two-color scattering is critical to this diagnostic effort of resolving neutral densities directly for the first time in tokamaks. Discussion: While we show that in the midplane, the Rayleigh scattering photon count is too low to resolve the neutral densities, it is conceivable to utilize this approach to get to the neutral penetration into the edge/pedestal of tokamaks. More specifically, it is well-known that the neutral opacity of ITER is still unknown. This diagnostic could enable the study of the pedestal density formation under gas fuelling (to be in the field of view of diagnostic resulting into a doubly active diagnostics) in present day tokamaks, which could be used to unravel the degree of neutral opacity for future devices such as ITER. Acknowlegment: The authors are grateful to Prof. R.B. Miles for his interest in the work and valuable comments. This work was funded under the fast track 2019 LDRD project "Feasibility Study of a Laser-Based Approach for Diagnosing Deuterium Neutrals in the Edge of Fusion Devices." References [1] H. Y. Guo, C. F. Sang, P. C. Stangeby, L. L. Lao, T. S. Taylor, and D. M. Thomas. Small angle slot divertor concept for long pulse advanced tokamaks. Nuclear Fusion, 57(4):044001, February 2017. [2] L. Hilico, N. Billy, B. Gr´emaud, and D. Delande. Polarizabilities, light shifts and two-photon transition probabilities betweenJ= 0 states of the H2$\mathplus$and D2$\mathplus$molecular ions. Journal of Physics B: Atomic, Molecular and Optical Physics, 34(3):491 -- 507, January 2001. [3] Christopher M. Limbach. Characterization of Nanosecond, Femtosecond, and Dual Pulse Laser Energy Deposition In Air For Flow Control and Diagnostic Applications. PhD thesis, Princeton University, 2015. [4] Yu. Ya. Milenko, L. V. Karnatsevich, and V. S. Kogan. On temperature dependence of the polar- izability of H2 and D2 molecules. Physica, 60(1):90 -- 96, July 1972. [5] Richard B. Miles, Walter R. Lempert, and Joseph N. Forkey. Laser Rayleigh scattering. Measurement Science and Technology, 12(5):R33, 2001. [6] Richard B. Miles, Azer P. Yalin, Zhen Tang, Sohail H. Zaidi, and Joseph N. Forkey. Flow field imag- ing through sharp-edged atomic and molecular 'notch' filters. Measurement Science and Technology, 12(4):442, 2001. [7] H. R. Sadeghpour and A. Dalgarno. Rayleigh and Raman scattering by hydrogen and caesium. Journal of Physics B: Atomic, Molecular and Optical Physics, 25(22):4801 -- 4809, November 1992. [8] D. P. Stotler, F. Scotti, R. E. Bell, A. Diallo, B. P. LeBlanc, M. Podest`a, A. L. Roquemore, and P. W. Ross. Midplane neutral density profiles in the National Spherical Torus Experiment. Physics of Plasmas, 22(8):082506, August 2015. [9] Yuji Suzuki, Shingo Ono, Hidetoshi Murakami, Toshimasa Kozeki, Hideyuki Ohtake, Nobuhiko Sarukura, Genta Masada, Hiroyuki Shiraishi, and Ichiro Sekine. 0.43 J, 10 Hz Fourth Harmonic Generation of Nd:YAG Laser Using Large Li2b4o7 Crystals. Japanese Journal of Applied Physics, 41(7B):L823, July 2002. [10] Zhichao Wang, Feng Yang, Guochun Zhang, Yong Bo, Shanshan Liu, Shiyong Xie, Yiting Xu, Nan Zong, Fangqin Li, Biaolong Liu, Jialin Xu, Qinjun Peng, Jingyuan Zhang, Dafu Cui, Yicheng Wu, and Zuyan Xu. High-power ultraviolet 278 nm laser from fourth-harmonic generation of a Nd:YAG laser in CsB3O5. Optics Letters, 37(12):2403 -- 2405, June 2012. 15 This figure "Figure_RS-TS-BPfilter.png" is available in "png"(cid:10) format from: http://arxiv.org/ps/1912.03601v1 This figure "Guo2017_Fig4_Te,D2distributions.jpg" is available in "jpg"(cid:10) format from: http://arxiv.org/ps/1912.03601v1 This figure "Stotler_Fig5_MidplaneDensityProfiles.PNG" is available in "PNG"(cid:10) format from: http://arxiv.org/ps/1912.03601v1
1709.04628	1	1709	2017-09-14T06:24:21	Design of an underwater acoustic bend by pentamode metafluid	[ "physics.app-ph" ]	We design an impedance matching underwater acoustic bend with pentamode microstructure. The proposed bend is assembled by pentamode lattice. The effective density and compressive mod- ulus of each unit cell can be tuned simultaneously, which are modulated to guarantee both the bending effect and high transmission. The standard deviations of transmitted phase are calculated to quantitatively evaluate the degree of the distortion of the transmitted wavefront, while the trans- mission is calculated to appraise the degree of acoustic impedance matching. The low standard deviations and high transmission indicate that the designed bend has a nice broadband bending effect and is impedance-matched to water. This design has potential applications in underwater communication and underwater detection.	physics.app-ph	physics	Design of an underwater acoustic bend by pentamode metafluid Zhaoyong Sun,1, 2, ∗ Han Jia,1, 2, 3, † Yi Chen,4, ‡ Zhen Wang,1, 2, § and Jun Yang1, 2, 3, ¶ 1Key Laboratory of Noise and Vibration Research, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, People's Republic of China 2University of Chinese Academy of Sciences, Beijing 100049, Peoples Republic of China 3State Key Laboratory of Acoustics, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, Peoples Republic of China 4School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, People's Republic of China We design an impedance matching underwater acoustic bend with pentamode microstructure. The proposed bend is assembled by pentamode lattice. The effective density and compressive mod- ulus of each unit cell can be tuned simultaneously, which are modulated to guarantee both the bending effect and high transmission. The standard deviations of transmitted phase are calculated to quantitatively evaluate the degree of the distortion of the transmitted wavefront, while the trans- mission is calculated to appraise the degree of acoustic impedance matching. The low standard deviations and high transmission indicate that the designed bend has a nice broadband bending effect and is impedance-matched to water. This design has potential applications in underwater communication and underwater detection. 7 1 0 2 p e S 4 1 ] h p - p p a . s c i s y h p [ 1 v 8 2 6 4 0 . 9 0 7 1 : v i X r a ∗ [email protected] † [email protected] ‡ [email protected] § [email protected] ¶ [email protected] I. INTRODUCTION 2 Acoustic metamaterials can exhibit many abnormal acoustic properties, such as negative density, negative modulus[14, 15, 17], anisotropic density[5, 27] and anisotropic modulus[10, 24]. These remarkable properties provide possibilities to manipulate sonic waves in ways that are impossible in ordinary materials. With these abnormal proper- ties, acoustic metamaterials can be used in acoustic cloaking[1, 4, 7, 23], acoustic subwavelength imaging[6, 8, 11, 22], rainbow trapping[9, 19, 31] and so on. Bending the wave without distorting its wavefront is one of these interesting applications. In recent years, some interesting work about acoustic bend has been reported. L.Wu etal. employed a two-dimensional graded sonic crystal to realize a graded index medium based acoustic bending waveguide[29]. Y. Wang etal. proposed an anisotropic metamaterial with only one component of the mass density tensor near zero to control the sound wave propagation and realized perfect bending waveguides numerically[28]. W. Lu et al. have designed and fabricated a broadband acoustic right angle bend by perforated panels which can tune the effective density[16]. In our previous work[30], we used perforated plates with side pipes to design an acoustic bend working in air. The effective bulk modulus is tuned by the pipes, while the effective density is adjusted by the perforated panels simultaneously. This enables the bend to match acoustic impedance to air. Acoustic bend also have important applications in underwater communication and underwater detection. However, perforated panels and the side pipes can not be used in water, since it is difficult to find a rigid material in water. Therefore, it is necessary to consider other approaches. An appropriate option is pentamode material(PM) [18]. PM is a kind of special solid structure with tunable effective modulus and density that only guarantees the longitudinal wave to propagate in it, and can be designed to have anisotropic modulus[10, 13]. This makes it have the similar acoustic property with fluid[21] and the advantage for designing impedance matching underwater devices[25, 26]. In this article, we use 2-dimensional(2D) version of the pentamode material(PM) lattice to design an underwa- ter acoustic bend. The required acoustic parameters are obtained by theoretical calculation. From the calculation, the modulus of the bend is proportional to the radial position, while the density is inversely proportional to the radial position. This property is the key point to protect the wavefront from being distorted and ensure the acoustic impedance matched to water. The unit cells are designed by a homogenization method[3], and their effective parame- ters coincide with the required parameters well. The latticed pentamode bend(PMB) is assembled by these unit cells. The simulated results are demonstrated and show the broadband bending effect and high transmission of the latticed PMB. II. THEORY OF THE ACOUSTIC BEND Figure 1 illustrates the outline of the 2D acoustic bend. In the proposed model, an acoustic bend (domain B) FIG. 1. (Color online) A schematic view of the acoustic bend with waveguides. Domain B is the acoustic bend, while domain A and C are a horizontal and a vertical waveguide correspondingly. connects a vertical waveguide (domain A) with a horizontal waveguide (domain C). The inner and outer radii are denoted as r1 and r2. These two waveguides are filled with the background fluid with density ρ0 and bulk modulus . The wave is emitted from the bottom of the vertical waveguide, and K0, which give the sound velocity c0 = assumed to travel with the velocity cθ = cθ(r) along the azimuthal direction in the bend. ρ0 (cid:113) K0 CBAIncident waveThe phase measure line Lereθ3 For an ideal acoustic bend, two conditions have to be satisfied in order to keep the nice bending effect and high transmission: one is that the phase of the wave does not vary along the radial direction, and the other is that all energy of the incident wave can pass through the bend. The former implies that the travel time of the wave along different curves r = ri is the same, while the latter signifies that the acoustic impedance of the bend matches to the fluid in the waveguides. These two conditions can immediately give the following equations: riθ cθ(ri) = rjθ cθ(rj) , ρB(r)cθ(r) = ρ0c0 (1) where ri and rj are two arbitrary radial coordinates, ρB(r) is the density of the bend. Equation (1) shows that velocity cθ(r) is proportional to the radial coordinate, while the density ρB(r) is inversely proportional to the radial coordinate. This gives the acoustic parameters distributions of the bend as follows: ρB(r) = bρ0 r , K = rK0 b (2) where K is compressive modulus of the bend B(K = ρBc2 and modulus of the bend. θ), and b is a constant that can tune the required density Since Eq.(1) and Eq.(2) are derived by the analysis in azimuthal direction, the acoustic parameters along the radial direction does not influence the bending effect. In the work of the acoustic bend in air [30], we used the unit cells with anisotropic density and isotropic modulus to design the bend structure. The azimuthal density and the bulk modulus satisfy Eq.(2), which makes the impedance match to air. Here, we use 2D PMs with anisotropic modulus and isotropic density to design an underwater PM bend(PMB). The acoustic parameters of the PMB are normalized to the corresponding values of water, and shown as follows ρ = b r , Kθ = r b , Kr = Kθ 2 , b = 0.6m (3) with the normalized azimuthal compressive modulus Kθ, normalized radial compressive modulus Kr and the normal- ized density ρ. FIG. 2. (Color online) (a) Microstructure of 2D PMs, with a highlighted unit cell. (b) Band diagram along the N − Γ − M direction. The first Brillouin zone is shown in the center of the figure. III. DESIGN OF THE UNIT CELLS FOR THE LATTICED PENTAMODE MATERIAL BEND The elastic tensor of the 2D PM that used in the underwater bend can be characterized as follows in the polar (4) coordinates: C =  C11 C12 C12 C22 0 0 0 0 C33  = K0  Kr Krθ Krθ Kθ  0 0 0 Grθ 0 4 where K0 is the bulk modulus of the background fluid. The background fluid is water with density ρ0 = 1000 kg/m3 and bulk modulus K0 = 2.25 GPa. For a perfect PM, the matrix elements in Eq.(4) have to satisfy the condition KrKθ = K 2 rθ and Grθ = 0 so that the modes pertaining shear deformations vanish and only the compressive one is reserved. In this sense, only longitudinal wave can propagate in the perfect PM[21]. In practical case, the shear modulus can not be zero because of the engineering limit and the demand for structural stability[10, 13]. Thus, the 2D practical PM needs KrrKθθ ≈ K 2 rθ and Grθ ≈ 0 to minimize the shear modulus. The model of 2D PM is designed as shown in Fig.2. The geometric parameters of the unit cell are characterized by length of the horizontal struts h, length of the oblique struts l, strut thickness t, angle β and the block size (ω, s). In order to demonstrate the fluid-like property of the structure, we choose a typical unit cell and calculate the band structure using Bloch-Floquet analysis[12] in COMSOL Multiphysics. The geometric parameters of the unit cell are set as l = h = 13.4 mm, β = 60◦, t = 0.7 mm, ω = 10 mm and s = 0.77 mm. The substrate of the latticed PM is chosen as aluminum with density ρAl = 2700 kg/m3, Youngs modulus EAl = 69 GPa and Poisson's ratio ν = 0.33. Figure 2 depicts the first four bands. The black line describes the shear wave mode, while the red line represents the longitudinal wave mode. The two blue lines correspond to flexural modes[2], which exist at high frequency range in our model. FIG. 3. (Color online) (a) Profiles of continuously varying (solid lines) acoustic parameters of the perfect PMB and layered (discrete symbols) objective acoustic parameters of the designed PMB. (b) The geometric parameters of the required unit cells. This is because the blocks on the oblique struts can prevent the struts from being bended. It is easy to see that at a large frequency domain (gray part), only the longitudinal modes exist in the structure, which imply the fluid- like property of the PM lattice. At the quasi-static regime[20], the effective density of the PM lattice equals to the volume average of the unit cell mass. Hence the effective density is immediately obtained as ρ = 1.004 (normalized to the density of water) from the geometric parameters and the substrate material. The longitudinal wave velocity along the azimuthal and radial direction can be obtained from the band diagram, and read as cLθ = 1508 m/s and cLr = 1499 m/s. Thus the normalized azimuthal and radial compressive modulus are Kθ = ρρ0c2 Lθ/K0 = 1.015 and Kr = ρρ0cLr/K0 = 1.003. It can be seen that the effective property of this unit cell coincide with that of water well. Therefore, for any given unit cell, the effective compressive modulus and density can be estimated with the band diagram. The geometry of the unit cell has an important impact on the effective acoustic parameters in such a way: angle β mainly determines the anisotropy of effective modulus along with the length ratio η = h l ; strut thickness t primarily contributes to the effective stiffness; The block size (ω, s) can adjust the effective density almost without influencing the effective modulus. Thus, the required acoustic property in Eq.(3) can be realized by adjusting the geometric parameters of the unit cells and retrieving their bands. The required acoustic parameters in Eq.(3) are depicted by solid lines in Fig.3a. According to Eq.(3), they are symmetric in the azimuthal direction and continuous in the radial direction. Thus domain B in Fig.1 is divided into 125 same sectors along circumferential direction. The continuous parameters has to be discretized with layered approximation. The first layer start from the location at the inner radius r1 = 743.8mm. The objective acoustic parameters in the first layer is Kr(r1), Kθ(r1), ρ(r1). And the mean task is to find the right unit cell that has effective parameters identical to the objective ones. This can be realized by retrieving the bands of different unit cells and calculating their effective densities and modulus. The structure of layers with microstructure is shown as in 0.80.91.01.11.2Location r(m)0.000.250.500.751.001.251.501.752.00Normalized Parameters(a)ρeffKeffθKeffrGeffrθρKθKr0.80.91.01.11.2Location r(m)0246810Geometric Parameters(b)l (mm)h (mm)w (mm)s (mm)5t (mm)5 FIG. 4. (Color online) (a) The microstructure of the first five layers in a sector. (b) Top view of the latticed 90◦ PMB. Fig.4a. The ending of the first layer is the starting of the second layer. The last layer ends at r2 = 1201.0 mm. The parameters of the unit cells in a sector is shown in Fig.3b, while their corresponding effective acoustic parameters are shown as discrete symbols in Fig.3a. It can be seen that the effective parameters of the unit cells coincide with the required ones well. We can also observe that the effective shear modulus of the structure is exceedingly small comparing with the compressive modulus. The bend structure is obtained by assembling the unit cells, shown as in Fig.4b. The maximal period is almost 15 mm which makes the unit cell at least 10 times smaller than the wavelength for frequencies under 8 kHz in water. Consequently, the layer can be regard as a homogeneous medium. And the effective acoustic parameters are expected to ensure the latticed PMB for smooth bending effect. IV. SIMULATION RESULTS (Color online) Acoustic pressure field of the 90◦ latticed PMB (a) and empty tube (b) for a plane wave incident from FIG. 5. the bottom of the vertical waveguide at 7.5 kHz. The performance of the 90◦ latticed PMB is simulated using finite element solver COMSOL Multiphysics. The incident and transmitted acoustic pressure fields of the latticed PMB are shown in Fig.5a. A plane wave at 7.5 kHz is emitted from the bottom of the vertical waveguide. It shows that the wavefront in the horizontal waveguide almost keeps the same shape with that in the vertical waveguide. For comparison, an empty tube that has the same shape with latticed PMB is also simulated. The acoustic pressure fields are shown as in Fig.5b. The wavefront in the horizontal waveguide in Fig.5b is severely distorted. Comparing Fig.5a with Fig.5b, it can be concluded that the latticed PMB can bend the wave smoothly at 7.5 kHz. 6 (Color online) (a) The standard deviations of a 90◦ empty tube(black diamonds), a 90◦ perfect PMB (red circles) and FIG. 6. a 90◦ latticed PMB. (b) The transmission of the 90◦ latticed PMB. In order to evaluate the deformation of the transmitted wavefront quantitatively, we calculate the standard devia- tion(SD) of the phases along the center line L(shown as dashed line in Fig.1). The standard deviation of the phase σ(φ) is expressed as follows: (cid:115) (cid:0)φi − φ(cid:1)2 n σ(φ) = Σn i (5) where φi is the phase distribution along line L, φ is the average value of all the phases, and n is the number of the total phase data. The value of the SD describes the degree of that the wavefront is distorted, i.e., a large SD means a serious distortion. Figure 6a shows the SDs of the latticed PMB, empty tube and perfect PMB at the frequencies ranging from 4 kHz to 8 kHz. It can be seen that the SDs of the perfect PMB are near to zero, which means the FIG. 7. (Color online) (a) Acoustic pressure field of a 120◦ latticed PMB. (b) The standard deviations for a 120◦ empty tube(black diamonds), a 120◦ perfect PMB (red circles) and a 120◦ latticed PMB. (c) The transmission of the 120◦ latticed PMB. transmitted wavefront is almost not twisted. The SDs of the latticed PMB coincide with that of the perfect PMB very well. Thus, the latticed PMB is as effective as the perfect PMB at a broadband frequency domain. 45678Freq(kHz)0.00.51.01.52.02.53.0Standard Deviation(a)90ࢪEmp(y∘T)be90ࢪPerfec(∘PMB90ࢪLa((iced∘PMB45678Freq( Hz)0.20.40.60.81.0Transmission Coefficient(b)90ࢪLa((iced∘PMBWe also calculated the transmission coefficient of the PMB to evaluate the degree of impedance matching. The transmission coefficient of 90◦ latticed PMB is shown in Fig.6b. It can be observed that the transmission coefficient is larger than 0.95. This implies that the acoustic impedance matches to water well. Thus, the 90◦ latticed PMB has been verified to have a nice bending effect and be acoustic impedance matched to water. Since the latticed PMB is impedance matching in the azimuthal direction, the azimuthal length of the PMB will not influence the bending effect and transmission. Hence the latticed PMB can work at a wide bend angle range. A 120◦ latticed PMB is also designed and simulated. The simulated results are shown in Fig.7. The methods and evaluation for the 120◦ latticed PMB is same with that of the 90◦ one. It is obvious that the 120◦ latticed PMB exhibits the bending effect as good as the 90◦ latticed PMB. 7 V. CONCLUSION In this article, we have designed and simulated a latticed pentamode material acoustic bend. The required acoustic parameters that can keep the good bending effect and high transmission are obtained by a theoretical calculation. The latticed PMB is achieved by layered approximation with the 2D hexagonal PM unit cells that have the effective parameters according with the required ones. The unit cells are obtained by retrieving the bands and calculating their effective acoustic parameters. The standard deviation of the transmitted phase is calculated to give a accurate analysis for the wave front, which shows the latticed PMB can keep the wavefront almost invariant. And the high transmission confirms that the latticed PMB is impedance-matched to water well. We hope this work can contribute to the researches on underwater communication and underwater detection. The authors sincerely acknowledge the financial support of the Youth Innovation Promotion Association CAS (Grant No. 2017029) and the National Natural Science Foundation of China (Grant No. 11304351, 1177021304). ACKNOWLEDGMENTS [1] Yafeng Bi, Han Jia, Wenjia Lu, Peifeng Ji, and Jun Yang. Design and demonstration of an underwater acoustic carpet cloak. Sci. Rep., 7(1):705, 2017. [2] Xuan Cai, Lei Wang, Zhigao Zhao, Aiguo Zhao, Xiangdong Zhang, Tao Wu, and Hong Chen. The mechanical and acoustic properties of two-dimensional pentamode metamaterials with different structural parameters. Appl. Phys. Lett., 109(13):131904, 2016. [3] Yi Chen, Xiaoning Liu, and Gengkai Hu. Latticed pentamode acoustic cloak. Sci. Rep., 5:15745, October 2015. [4] Yi Chen, Mingye Zheng, Xiaoning Liu, Yafeng Bi, Zhaoyong Sun, Ping Xiang, Jun Yang, and Gengkai Hu. Broadband solid cloak for underwater acoustics. Phys. Rev. B., 95(18):180104, 2017. [5] Johan Christensen and F Javier Garc´ıa de Abajo. Anisotropic metamaterials for full control of acoustic waves. Phys. Rev. Lett., 108(12):124301, 2012. [6] Ke Deng, Yiqun Ding, Zhaojian He, Heping Zhao, Jing Shi, and Zhengyou Liu. Theoretical study of subwavelength imaging by acoustic metamaterial slabs. J. Appl. Phys., 105(12):124909, 2009. [7] Wenlin Hu, Yuxian Fan, Peifeng Ji, and Jun Yang. An experimental acoustic cloak for generating virtual images. J. Appl. Phys., 113(2):45, 2013. [8] Han Jia, Manzhu Ke, Rui Hao, Yangtao Ye, Fengming Liu, and Zhengyou Liu. Subwavelength imaging by a simple planar acoustic superlens. Appl. Phys. Lett., 97(17):173507, 2010. [9] Han Jia, Minghui Lu, Xu Ni, Ming Bao, and Xiaodong Li. Spatial separation of spoof surface acoustic waves on the graded groove grating. J. Appl. Phys., 116(12):124504, 2014. [10] Muamer Kadic, Tiemo Buckmann, Robert Schittny, and Martin Wegener. On anisotropic versions of three-dimensional pentamode metamaterials. New J. Phys., 15(2):023029, 2013. [11] Nad`ege Kaina, Fabrice Lemoult, Mathias Fink, and Geoffroy Lerosey. Negative refractive index and acoustic superlens from multiple scattering in single negative metamaterials. Nature, 525(7567):77–81, 2015. [12] AA Kutsenko, AJ Nagy, X Su, AL Shuvalov, and AN Norris. Wave propagation and homogenization in 2d and 3d lattices: a semi-analytical approach. Q. J. Mech. Appl. Math., 70(2):131–151, 2017. [13] Christopher N Layman, Christina J Naify, Theodore P Martin, David C Calvo, and Gregory J Orris. Highly anisotropic elements for acoustic pentamode applications. Phys. Rev. Lett., 111(2):024302, 2013. [14] Sam Hyeon Lee, Choon Mahn Park, Yong Mun Seo, Zhi Guo Wang, and Chul Koo Kim. Composite acoustic medium with simultaneously negative density and modulus. Phys. Rev. Lett., 104(5):054301, 2010. [15] Zhengyou Liu, Xixiang Zhang, Yiwei Mao, YY Zhu, Zhiyu Yang, Che Ting Chan, and Ping Sheng. Locally resonant sonic materials. Science, 289(5485):1734–1736, 2000. 8 [16] Wenjia Lu, Han Jia, Yafeng Bi, Yuzhen Yang, and Jun Yang. Design and demonstration of an acoustic right-angle bend. J. Acoust. Soc. Am., 142(1):84–89, July 2017. [17] Jun Mei, Zhengyou Liu, Weijia Wen, and Ping Sheng. Effective dynamic mass density of composites. Phys. Rev. B., 76(13):134205, 2007. [18] Graeme W Milton and Andrej V Cherkaev. Which elasticity tensors are realizable? J. Eng. Mat. Tech., 117(4):483–493, 1995. [19] Xu Ni, Ying Wu, Ze-Guo Chen, Li-Yang Zheng, Ye-Long Xu, Priyanka Nayar, Xiao-Ping Liu, Ming-Hui Lu, and Yan-Feng Chen. Acoustic rainbow trapping by coiling up space. Sci. Rep., 4, 2014. [20] AN Norris. Mechanics of elastic networks. Proc. R. Soc. A, 470(2172):20140522–20140522, 2014. [21] Andrew N Norris. Acoustic metafluids. J. Acoust. Soc. Am., 125(2):839–849, 2009. [22] Jong Jin Park, Choon Mahn Park, KJB Lee, and Sam H Lee. Acoustic superlens using membrane-based metamaterials. Appl. Phys. Lett., 106(5):051901, 2015. [23] JB Pendry and Jensen Li. An acoustic metafluid: realizing a broadband acoustic cloak. New J. Phys., 10(11):115032, 2008. [24] Chen Shen, Jun Xu, Nicholas X Fang, and Yun Jing. Anisotropic complementary acoustic metamaterial for canceling out aberrating layers. Phys. Rev. X, 4(4):041033, 2014. [25] Xiaoshi Su, Andrew N Norris, Colby W Cushing, Michael R Haberman, and Preston S Wilson. Broadband focusing of underwater sound using a transparent pentamode lens. J. Acoust. Soc. Am., 141(6):4408–4417, 2017. [26] Ye Tian, Qi Wei, Ying Cheng, Zheng Xu, and Xiaojun Liu. Broadband manipulation of acoustic wavefronts by pentamode metasurface. Appl. Phys. Lett., 107(22):221906, 2015. [27] Daniel Torrent and Jos´e S´anchez-Dehesa. Anisotropic mass density by radially periodic fluid structures. Phys. Rev. Lett., 105(17):174301, 2010. [28] Yang-Yang Wang, Er-Liang Ding, Xiao-Zhou Liu, and Xiu-Fen Gong. An acoustic bending waveguide designed by anisotropic density-near-zero metamaterial. Chin. Phys. B, 25(12):124305, 2016. [29] Liang-Yu Wu and Lien-Wen Chen. An acoustic bending waveguide designed by graded sonic crystals. J. Appl. Phys., 110(11):114507, 2011. [30] Yuzhen Yang, Han Jia, Wenjia Lu, Zhaoyong Sun, and Jun Yang. perforated plates and side pipes. J. Appl. Phys., 122(5):054502, 2017. Impedance-matching acoustic bend composed of [31] Chen Zhou, Baoguo Yuan, Ying Cheng, and Xiaojun Liu. Precise rainbow trapping for low-frequency acoustic waves with micro mie resonance-based structures. Appl. Phys. Lett., 108(6):063501, 2016.
1904.04594	1	1904	2019-04-09T11:17:05	Tracking-free Determination of Microparticle Motion from Image Variance	[ "physics.app-ph", "cond-mat.soft", "eess.IV", "physics.data-an", "physics.flu-dyn" ]	In this work, we use the standard deviation of image pixel intensity to analyse the speed, direction and surface-interaction of microparticles in fluid. First, we present an analytical model for estimating the total variance in the image space for directed or diffusive motion of microparticles and show that this measure is correlated to the density and speed of the particles. The analytical model was found to have good agreement with numerical simulations for low particle density. Then, using only the local image variance we obtain the magnitude and direction of the particle velocity in a rectangular microfluidic channel, closely matching the theoretical profile. Further, we also demonstrate the application of this method as a probe for particle-surface interactions by extracting the differences in distribution and time-evolution of image variance from mobile microparticles adhering to different surfaces. We believe that the image variance based method described here presents an addition to the suite of tracking-free techniques such as Differential Dynamic Microscopy (DDM) to extract motility parameters from video data.	physics.app-ph	physics	Tracking-free Determination of Microparticle Motion from Image Variance Harish Sasikumar 1, and Manoj M. Varma 1 , 2 , * 1Center for Nano Science and Engineering, Indian Institute of Science, Bangalore, 560012, India 2Robert Bosch Center for Cyber Physical Systems, Indian Institute of Science, Bangalore, 560012, India Abstract In this work, we use the standard deviation of image pixel intensity to analyse the speed, direction and surface-interaction of microparticles in fluid. First, we present an analytical model for estimating the total variance in the image space for directed or diffusive motion of microparticles and show that this measure is correlated to the density and speed of the particles. The analytical model was found to have good agreement with numerical simulations for low particle density. Then, using only the local image variance we obtain the magnitude and direction of the particle velocity in a rectangular microfluidic channel, closely matching the theoretical profile. Further, we also demonstrate the application of this method as a probe for particle-surface interactions by extracting the differences in distribution and time-evolution of image variance from mobile microparticles adhering to different surfaces. We believe that the image variance based method described here presents an addition to the suite of tracking-free techniques such as Differential Dynamic Microscopy (DDM) to extract motility parameters from video data. Introduction Ability to quantify motion at microscale is of fundamental importance in understanding several phenomena involving Brownian and phoretic motion of particles as well as their interactions with materials they come in contact with. A straight-forward method for these studies is particle-tracking, where the positions of individual particles are tracked over time followed by computational analysis of particle trajectories, for instance, as in micro particle velocimetry [1]. These techniques generally involve the use of fluorescent or other tracer particles and computational extraction of particle position as a function of time. Often elaborate imaging techniques and autocorrelation of the images are used to estimate the particle velocities [2]. Though information of particle dynamics is present in its entirety in these methods, there are two aspects which motivates development of tracking-free methods to characterize micro-scale motion. Firstly, tracking data is voluminous and often one is only interested in summary statistics - central tendencies, measures of dispersions etc. of particle locations, which represent a significant compression of information contained in the raw particle trajectory data. Secondly, tracking is computationally intensive and prone to errors and is particularly challenging in crowded environments. Therefore, obtaining summary statistics of motion by tracking-free methods is desirable. Among tracking-free methods, Differential Dynamic Microscopy (DDM) has emerged as a popular technique for analysing the dynamics of particles -- both inanimate [3, 4] and animate * Author to whom correspondence should be addressed. Electronic mail: [email protected] [5, 6]. Apart from saving time and effort, this method avoids the errors associated with particle tracking and provides a reliable method to estimate the average dynamics of microscopic particles [7]. DDM studies can work well even at the optical resolution limits and reveal the finer workings of many microscopic systems. For instance, usefulness of DDM in differentiating directed motion and random diffusion of sub-diffraction particles in crowded and noisy environment has made it a suitable procedure for studying intracellular transport mechanism [8]. In conventional DDM, the analysis is often done in k-space by assuming the isotropy of the image (or differential image). This allows one to take azimuthal averages in the Fourier domain to isolate the different Fourier components. There are only a few deviations to this general approach which are reported; for instance, Differential Variance Analysis (DVA) [9] has been used to monitor dynamic spatial heterogeneities in amorphous materials. There are also methods, such as the one reported by Mathias et al. [10], to quantify the alignment of anisotropic colloids to external fields by extracting the directionality using DDM. In this case, a more generalized form of DDM is derived, avoiding the azimuthal averaging, maintaining the vector nature of the wave vector. It is to be noted that even here, it is the global orientational order parameter of the system which is extracted. Anisotropy in the differential image data, arising from spatial variations in the speed or direction of motion is common and is important to be captured. Micro particle image velocimetry (mPIV) techniques have been applied to get the speed and direction of micro-scale flows [11]. This technique requires the autocorrelation of image fields to estimate the spatial shifts and can be computationally expensive [12]. In this paper, we show that the variance (or standard deviation) of the pixel intensity, which can be easily extracted from video data, can be used to obtain velocity fields. This approach does not require calculations of correlations or transforming data to fourier domain. By dividing the video data spatially and temporally, one can obtain the speed and direction of micro-particles in the system. The inherent heterogeneities of the system are also captured due to this spatial compartmentalization. In addition to motility, one can also use pixel variance to gather information about the nature of interaction of particle with surfaces in contact with it. Studying these interactions is useful in developing advanced surfaces like polyelectrolyte (PE) brushes, where often techniques like optical tweezers and AFM colloidal probes are used in this regard [13]. These surface interactions also affect the Brownian motion of micro particles in contact with the surfaces resulting in variations in the number of mobile particles, their velocity distributions as well as the time evolution of their velocities. Hence, the statistics of the movements of microparticles can in turn be used in studying the properties of surfaces. Recently such fluctuations were used for sensing the position and orientation of single, bound viruses [14]. The paper is organized in the following manner. In the first section, we present theoretical and numerical studies to clarify the process of using pixel variance to estimate motility parameters for directed as well as diffusive motion. We then show an example of using pixel variance to obtain the velocity profile from video recording of flow in microchannels with rectangular cross-section. Finally, we show how temporal evolution of the pixel variance informs one about the nature of interaction between a microparticle and the surface it contacts. Theory and Simulation Theoretical Model Consider an imaging plane of area 𝑎 × 𝑏 pixels with p particles at random locations. Each pixel in the area has a value xi,j, addressed by its location (i, j), which is an element in the set G = {(1,1), (1,2), … . (1, b), (2,1), … , (a, b)} with a cardinality ab. For simplicity, we assume monochrome images with a dimension scaling factor of γ (usually measured in the units μm pixel ), acquired at a frame rate of 𝑓(usually measured in the units frames second ). The acquisition is done for a total time of T seconds, which corresponds to N (= Tf) frames. In the analytical model, we assume that the particles are sparsely distributed so that during the observation time, their paths do not collide or cross over. Further, each particle is assumed to be occupying only one pixel in the observation area. (These two assumptions are removed in the simulation model). The particle is considered to be opaque which makes the occupied pixel to have a lower value (l), than that of an unoccupied pixel (h). For obtaining an analytical model for the linear motion, we assume a particle velocity of v which creates a scaled particle velocity vs in the images, given by 𝑣𝑠 = 𝑣 𝛾𝑓 = 1 𝑛 (1) where n is the number of time frames for which the particle appears stationary before moving on to the next pixel along its direction of motion. It is to be noted that vs has unit assume that the total number of frames captured is an integral multiple of n, i.e 𝑁 = 𝑚𝑛, where m is a positive integer. In effect, m is the total displacement of the particle (in unit of pixels) during the total observation time. When observed for N time frames, the pixels along the trajectories of particles change, whereas the value of the remaining pixels remain constant (h). Hence, the average value of a pixels in the imaging plane is given by, . We frames pixels = { (𝑥𝑖,𝑗) 𝑥𝑝𝑎𝑡ℎ = ℎ − 𝛥 𝑁𝑣𝑠𝑐 ; 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∈ 𝑆𝑖,𝑗 𝑥𝑏𝑙𝑎𝑛𝑘 = ℎ ; 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∉ 𝑆𝑖,𝑗 (2) Where Si,j is the set of pixels where the particle has occupied during the observation and Δ is the difference in pixel values, given as ℎ − 𝑙. The above expression indicates a velocity dependent histogram of the averages of pixels. Similarly, the standard deviation of the pixels in the imaging plane also has a velocity dependent histogram given by, 𝜎(𝑥𝑖,𝑗) = { 𝛥 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∈ 𝑆𝑖,𝑗 √𝑁𝑣𝑠 0 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∉ 𝑆𝑖,𝑗 (3) Summing up the standard deviations of all the pixels in the imaging plane, we can define and calculate a global parameter S given by, 𝑆 = ∑ 𝜎(𝑥𝑖,𝑗) = 𝑝𝛥√𝑁𝑣𝑠 (𝑖,𝑗) ∈ 𝐺 (4) Note that, here p is the number of particles. Though the equations developed so far provide an insight into the correlation between pixel standard deviations and velocity of the particles, they can be insufficient for predicting experimental outcomes. It is also worth noting that for particles with uniform velocity, S is independent of the area of simulation and is linearly dependent on the number of particles. However, in real experiments, these equations are inaccurate because the particles often occupy multiple pixels and their trajectories can intersect. Hence, a simulation model was developed which takes care of longer simulation times with trajectory crossings, and multi-pixel particles. For modelling systems with circularly symmetric larger (multi-pixel) particles, the images have to be considered as airy functions as they are the point spread function of a diffraction-limited particle. However, for mathematical convenience, we use the Gaussian distribution to approximate this function [15]. (𝐺) = 𝐴𝑒− 𝑥𝑖,𝑗 (𝑖−𝑖𝑟)2+(𝑗−𝑗𝑟)2 2𝑅2 (5) Where xi,j is the value of the pixel at (i, j) for the Gaussian distribution with center at (ir, jr) and the variance R. See Section 2 of supplementary information (SI) for the comparison of multi-pixel simulation model and experimentally observed micro particle. For the case where the total displacement of the multipixel particle is larger than the particle dimension (𝑁𝑣𝑠 ≫ (G) 𝑅), the average pixel value (xi,j ) is given by (𝐺) 𝑥𝑖,𝑗 = 𝐴√𝜋 𝑖2 4𝑅2 𝑒− 𝑅 𝑁 (6) 2 G ) Similarly, the mean square displacement (xi,j individual pixels (σ2 (xi,j)) can be derived as, can be calculated, using which the variance of 𝜎2(𝑥𝑖,𝑗) = 𝐴2 (√ 𝜋 2 𝑅 𝑁 − 𝜋 𝑅2 𝑁2) 𝑒− 𝑖2 4𝑅2 (7) And the sum of the standard deviations of all pixels due to p particles are given by 𝑆(𝐺) = 𝑝𝜎 = 𝐴𝑝𝑣 𝛾𝑓 √√ 𝜋3 2 𝑁𝑅3 (1 − √2𝜋 𝑅 𝑁 ) (8) Please refer to supplementary information (SI) section 3.1 for detailed derivation of Eq. 8. Simulation Model The simulation parameters were chosen to closely match the experimental conditions, namely, the size of the observation area and resolution of the imaging camera. For representative purposes, FIG. 1 (a) to (d) shows a simulation area with a dimension of 85 pixel × 85 pixel. For modelling uniformly illuminated background with opaque microparticles, a high value (h = 1) background was populated by pixels occupied by microparticles at lower values. To verify the theoretical approximations, initially the particles were simulated as single pixel model, given by 𝑥𝑖,𝑗 = { 1 ; 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∉ {𝑝𝑖,𝑗} 0 ; 𝑓𝑜𝑟 𝑥𝑖,𝑗 ∈ {𝑝𝑖,𝑗} (9) where the pi,j is the set of central locations of the particles. FIG. 1(a) shows an area of simulation of 85 pixel × 85 pixel with 10 randomly distributed particles represented by single pixel model. (b) shows an area of simulation of 85 pixel × 85 pixel with 10 particles represented by multipixel model with R = 1. Periodic boundary conditions were applied to all the four boundaries of the simulation area, so that the number of particles is conserved despite the free movement of particles. After a fixed amount of time (represented by a simulation time of N time frames), the variance of each pixel is calculated as 𝜎2(𝑥𝑖,𝑗) = 1 𝑁 𝑁 ∑(𝑥𝑖,𝑗(𝑛) − 𝑥𝑖,𝑗) 𝑛=1 2 (10) where xi,j(n) is the value of the pixel at the frame n and xi,j is the time average value of the pixel given by 𝑥𝑖,𝑗 = 1 𝑁 𝑁 ∑ 𝑥𝑖,𝑗(𝑛) 𝑛=1 (11) FIG. 1 (c) and (d) are the respective variances plots for single pixel and mutli-pixel models where the particles are simulated to move downwards by 60 contiguous positions from the locations in FIG. 1 (a) and (b). The periodic boundary condition is evident from the variance plots, where the particle trajectories ending at the bottom of the simulation area is restarted from the top. It is also apparent from these plots that the standard deviation plots align along the direction of the particle movement; vertical in this case. A global indicator of the particle movement, S, defined in equation 4 as the pixel standard deviation, was calculated. The simulations were done on a rectangular area of 1200 pixels high and 1600 pixels wide, a typical value closer to the image dimensions obtained from our camera. FIG. 1 (e) shows the comparison of the evolution of total standard deviation as predicted by the theory, the single- pixel (SP) model and the multi-pixel (MP) model. It can be seen that the theoretical predictions closely match the simulations at lower particle densities and shorter simulation times. At higher particle densities or when the simulation is done for a longer time, particle trajectories invariably cross and the total standard deviation predicted from the theoretical model overestimates the pixel variance as it assumes that the particle trajectories are mutually exclusive. With increasing particle number, the chance of trajectory crossings increases, resulting in larger overestimation of the pixel variance as seen in FIG. 1 (e). Similarly, the multi-pixel situation (i.e. when a particle covers multiple pixels in the image) exacerbates the effect of trajectory crossings and results in larger deviation between theory and simulation. FIG. 1. (a) and (b) are representative simulation area of 85 pixel × 85 pixel using respectively, single pixel (SP) and Multi pixel (MP) models for 10 randomly placed particles. (c) and (d) are the standard deviation plots after the particles have moved 60 pixels downwards. (e) The evolution of total standard deviation, S as the simulation time, N increases for (P=) 10, 100 and 1000 number of particles in the 1200 pixel × 1600 pixel simulation area. R is taken as 1.5. In case of two-dimensional Brownian movement, a two-dimensional random walk scenario was simulated i.e. at every time-step in the simulation, the subsequent positions of each of the particles were decided by random experiments of 4 equiprobable outcomes. Corresponding to each of these four outcomes, the particles were made to move in one of the four directions- up, down, left or right. FIG. 2 (a), (b), (c) and (d) are the standard deviation plot of such a simulation with initial configuration as in FIG. 1 (b) and for simulation times, N =10, 100, 300 and 600, respectively. FIG. 2 (e) is the evolution of total standard deviation, (similar to FIG. 1 (e)) for a system of 10, 100 and 1000 particles undergoing Brownian motion. For this, the position of the particle is updated at each time frame by translating it to one of the adjacent pixels, randomly. It can be seen that, for both single pixel and Gaussian models, S for Brownian motion stays lower than those in directed motion. In addition, even the slopes of the curves are lower than that of directed motion, making it apparent that the increase in pixel standard deviation is slower for systems with particles undergoing Brownian motion. FIG. 2. Particles represented by Multipixel (MP) model undergoing Brownian motion for total time frames of N= (a) 10, (b) 100, (c) 200 and (d) 600 frames. (e) Time evolution of total standard deviation (S) for particles (derived theoretically (Th) and using simulation (Simul)) undergoing Brownian motion compared with the linear motion (Lin) of single pixel (SP) and multipixel (MP) particles. The particles are assumed to be moving by 1 pixel in every time frame. Experiments, Observations and Results The movement of microparticles in a straight, laminar flow can be assumed as a good model system with directed motion. Hence, in order to verify the theoretical model, velocity profiles and flow directions were created using microfluidic devices in Polydimethylsiloxane (PDMS)[16]. The devices were fabricated using standard replica molding[17], with masters created using SU8 on Silicon (Refer Section 4 in SI text). A constant flow was maintained in the devices using a syringe pump. The device was observed using an optical microscope (Olympus, BX51M) with a 20x objective. The devices were imaged at 10 frames per second using a CCD monochrome camera (XM10) with an exposure time of 10 ms. The microparticles used in the experiments were polystyrene beads with diameter of 1 micron dispersed in an aqueous solution. These were procured from Sigma-Aldrich and used at a dilution corresponding to a number density of about 106 mm-3. Further details of the fabrication process, experimental setup and raw materials are given in SI section 4. Characterizing velocity profile of the particles As shown in the schematic in FIG. 3 (a), a rectangular microfluidic channel (h = w = 220 μm) was created. The entire PDMS device was attached on a glass slide coated with gold. The pressure difference between either ends of the channel create a flow in the positive z direction with a flow profile varying along x and y. FIG. 3 (b) shows a 1200 pixel high and 1600 pixel wide microscopic image of this device. Videos of the flow was acquired (supplementary information as video1) which was later analyzed using ImageJ [18], an open- source software. The standard deviations of the pixels for a time period of 1.5 seconds (𝑁=15 frames) were calculated (Refer FIG. 3 (c)) for an X-Z plane close to the bottom surface. The total standard deviation (summed along z direction) at each of the x position was calculated using the formula 1200 𝑆(𝑗) = ∑ 𝜎(𝑥𝑖,𝑗) 𝑖=1 (12) The velocity of the particles in each of the x position is then calculated using equation 8. 𝑣(𝑗) = 𝛾𝑓𝑆(𝑗) 𝐴𝑝𝐶 ; 𝑤ℎ𝑒𝑟𝑒 𝐶 = √√ 𝜋3 2 𝑁𝑅3 (1 − √2𝜋 𝑅 𝑁 ) (13) 𝛾 and 𝑓, being the dimension scaling factor and frame rate, are the properties of the imaging system. They were found to be 1.39 𝜇𝑚/𝑝𝑖𝑥𝑒𝑙 and 10 frames/second, respectively. 𝐴 and 𝑅 are the height and standard deviation of the 2D Gaussian produced by the particle on the imaging system. They were calculated from the averages of 5 particles and were found to be 33.58 and 3.5 respectively. As 𝑆(𝑗) is summed along the 𝑖 direction, 𝑝 is the total number of particles (summed along 𝑖 directions) in each of the j positions. The estimate of this was found by dividing the number of particles with the width (in pixels) of the channel. The value of 𝑝 was found to be 1.9. 𝑁 is the number of time frames used in the calculations and is 15. The complete set of parameters used for computation are tabulated in Section 3.2 of SI text. FIG. 3 (d) shows the z-averaged velocity profile of the micro channel. As expected, a faster flow was observed along the centre of the microfluidic channels than the outer regions. In order to check the validity of the variance based determination of speeds, a limited number of particles were tracked and their speeds were determined from the tracked trajectories. These are shown as the red data points in FIG. 3(d). The close match between the magnitude of speed obtained using the variance method and single particle tracking reveals the validity of our method in extracting accurate motion parameters from recorded data. Further, we also compared the analytical solution to the Navier-Stokes equation for a pressure-driven steady state flow in the rectangular channel considered [19]. Velocity profile along z direction at regions close to the bottom surface (y = h/10) is given by ∞ 𝑣𝑧(𝑥) = ∑ 𝑛,𝑜𝑑𝑑 1 𝑛3 [1 − 𝑐𝑜𝑠ℎ 𝑛𝜋 𝑐𝑜𝑠ℎ 𝑛𝜋 𝑥 ℎ 𝑤 2ℎ ] 𝑠𝑖𝑛 𝑛𝜋 10 (14) The first 100 non-zero terms in the above expression was used to plot the normalized Fourier sum approximation in FIG. 3 (d) and shows a qualitatively similar profile as obtained experimentally. FIG. 3. (a) Schematic of the single microchannel device. (b) Microscopic image of the fabricated device. (c) Standard deviations of the pixels for a time period of 1.5 seconds (15 frames). The imaging is done on an area of 1600 pixels wide and 1200 pixels long (d) Velocity profile along the x direction of the microfluidic channel. The black circular markers indicate the velocity estimated by from the pixel standard deviation (SD). The red square markers show the velocity estimated by particle tracking. Blue, broken line is the normalized velocity profile as predicted by the Fourier Sum approximations[19]. Characterizing direction of particles For creating flows at varying directions, a second device with structure as shown in FIG. 4 (a) was made. It consisted of two microfluidic channels (h = w = 50 μm) as in the previous device. The micro channels were then merged into a wider region (W' = 1 mm) so that the flow direction changes as shown in video [SI video2]. Figure 4 (b) shows the microscope image of the second device. As mentioned in the previous section, the experiments consisted of observing, analyzing and extracting information from the images. Initially, a standard deviation plot was generated as in the previous case. For extracting the local direction of the particles, we subdivided the imaging area into a 25×25 array of equal sizes. In each subdivision, the prominent direction was found out by the directionality plugin of ImageJ[20] using the scale space approach to the directional analysis [21]. FIG. 4 (c) is the directionality plot generated by replacing each of the 25×25 elements of the array with quivers at the angles extracted from the analysis and with standard deviation plot superimposed on it. FIG. 4 (d) is the magnified image for the regions marked with broken lines in FIG. 4 (c). The change is the flow direction as the particles enter the wider region is apparent from this image. FIG. 4 (e) is the average direction of particles at selected x positions. It can be seen that the particles move in z direction, at an angle closer to π 2 with respect to XY plane before reaching the wide opening (z=0 to z=470). Inside the wider opening (z>470), the angles spread out towards 0 showing an orthogonal component of the flow, in addition to the initial flow parallel to the z-axis. FIG. 4. (a) Schematic of the microchannel device opening to a wider area. (b) Microscopic image of the fabricated device. (c) Standard deviations of the pixels for a time period of 10 seconds (100 frames) superimposed by the quiver plot showing the direction. (d) Shows the magnified image for the regions marked with broken lines in (c) (e) Direction of particle movement as a function of Z and averaged along X direction. Characterizing particle-surface interactions From the theoretical model, it is understood that the total standard deviation (S) in a region is proportional to the density and velocity of the mobile particles. These parameters are in turn dependent on the net interaction force between micro particle and surfaces in contact. Hence, it can be hypothesized that the pixel standard deviation has a higher value for regions on a smooth substrate with lesser interaction (like Silicon) than that on rougher surfaces, such as charged polyelectrolyte layers, with higher particle-substrate interaction. Thus, we can use the pixel standard deviation to study the variations in particle motion and thereby the surface interactions. The following experiment is an illustration of such a potential application. FIG. 5 (a) shows the schematic of the device used for this study. It consists of a silicon substrate, partially coated with a thin (total thickness of about 10 nm), 5 bilayer system of polyelectrolyte (PE) consisting of Polyallylamine hydrochloride (PAH) and Poly acrylic acid (PAA). It is generally represented as (PAH/PAA)5. A fluid chamber is then made from a thin sheet of poly vinyl chloride (thickness of about 100 μm) with a hole (of diameter 3mm) in the centre [22]. The hole is aligned to the PE-Silicon interface. Micro beads in water medium (as in the previous experiments) were introduced into the chamber using a micropipette. The entire fluid column remains water tight with a cover slip attached at the top. FIG. 5 (b) shows the microscope image of the PE-Silicon interface in the fabricated device. From the initial observations it was apparent that there was larger number of mobile particles over the bare silicon region than the PE region. The comparison of pixel standard deviations in the two regions revealed more qualitative and quantitative understanding of the effects of the surface interactions. For studying the velocity distribution of mobile particles, 14 regions of size 40 pixel × 40 pixel were chosen, 7 each on PE and silicon. Each of these regions contains the entire trajectory of one of the mobile particles for a time duration of 30 seconds (300 timeframes). The histogram of the pixel standard deviations was then calculated for each region. Average of seven of these histograms, were calculated and plotted in FIG. 5 (c) (details are given in SI text, Section 5). It can be seen that an average pixel on the PE region has a lesser standard deviation as well as a lesser span in standard deviation (and correspondingly, particle velocity). This shows that the mobile microparticles on Si move faster on average than those on PE. Another interesting observation is obtained by comparing the time evolution of total standard deviation of pixel on PE and on bare Si. It has to be remembered that the sum of the pixel standard deviations in a region is proportional to the velocity and number of mobile particles in that region. For comparing the time evolution of total standard deviation of pixel on PE and on bare Si, regions of dimensions 1200 pixel high and 600 pixel wide were chosen on either surfaces (Refer FIG. 5 (b)). To avoid the irregularities at the interface, the remaining central region (1200 pixel high and 400 pixel wide) was neglected. The total standard deviation for a time duration of 30 seconds were calculated in every 2 minutes. Portions from these observations are given in SI video3. FIG. 5 (d) shows the evolution of total pixel standard deviation (𝑆) at regions on either surface. Fitting the evolution of total standard deviations to exponential decay models, a faster decay time constant (13.3 minutes) was extracted for PE than that for Silicon (22.2 minutes) surface. This points towards a higher damping of mobile particles on rough PE than those on Silicon as expected from comparison of surface roughness between these two structures shown in SI section 6. It is seen that the PE surface is significantly rougher indicating a larger surface interaction with the particles and consequently greater damping of its motion as revealed by the temporal evolution of the pixel variance. FIG. 5. (a) schematic of the device used to compare microparticle-surface interacions on Polyelectrolyte (PE) multilayer and on bare silicon. (b) Microscope image of the PE multilayer -- bare silicon interface. (c) histogram of pixel standard deviations on an average 40 pixel × 40 pixel region of particle trajectories on PE and on Si. (d) The time evolution of total standard deviation (for 300 frames) of pixels in regions of 1200 pixel × 600 pixels on PE multilayer and on Si. Conclusions To summarize, we have developed a simple, tracking-free method using variations in image space to analyse the speed, direction and surface-interaction of micro particles in various microfluidic systems. Velocity profiles closely matching theoretical estimates were obtained using this technique. This technique can be used in calculating flow patterns without any additions to the standard microscope setup. This could be a novel method in analysing the movements of microorganisms or microparticles without resorting to single particle tracking. Further, we also demonstrated the potential use of this method in probing microparticle-surface interactions which can be explored in future for sensing applications. We intend to explore the application of this technique to probe bacterial motility, especially to extract local structures within bacterial flows such as localized vortices. It is hoped that the image variance based method we described here presents an addition to the suite of tracking-free techniques such as Differential Dynamic Microscopy (DDM) to extract motility parameters from video data. References [1] S. T. Wereley and C. D. Meinhart, "Recent Advances in Micro-Particle Image Velocimetry," Annual Review of Fluid Mechanics, vol. 42, pp. 557-576, 2010. [2] [3] [4] [5] [6] [7] R. Lindken, M. Rossi, S. Grosse, and J. Westerweel, "Micro-Particle Image Velocimetry (microPIV): recent developments, applications, and guidelines," Lab Chip, vol. 9, pp. 2551-67, Sep 7 2009. N. Shokeen, C. Issa, and A. Mukhopadhyay, "Comparison of nanoparticle diffusion using fluorescence correlation spectroscopy and differential dynamic microscopy within concentrated polymer solutions," Applied Physics Letters, vol. 111, p. 263703, 2017. R. Cerbino and V. Trappe, "Differential dynamic microscopy: probing wave vector dependent dynamics with a microscope," Physical review letters, vol. 100, p. 188102, 2008. L. G. Wilson, V. A. Martinez, J. Schwarz-Linek, J. Tailleur, P. N. Pusey, W. C. Poon, and G. Bryant, "Differential dynamic microscopy of bacterial motility," Phys Rev Lett, vol. 106, p. 018101, Jan 7 2011. V. A. Martinez, R. Besseling, O. A. Croze, J. Tailleur, M. Reufer, J. Schwarz-Linek, L. G. Wilson, M. A. Bees, and W. C. Poon, "Differential dynamic microscopy: a high- throughput method for characterizing the motility of microorganisms," Biophys J, vol. 103, pp. 1637-47, Oct 17 2012. R. Cerbino and P. Cicuta, "Perspective: Differential dynamic microscopy extracts multi-scale activity in complex fluids and biological systems," J Chem Phys, vol. 147, p. 110901, Sep 21 2017. [8] M. Drechsler, F. Giavazzi, R. Cerbino, and I. M. Palacios, "Active diffusion and advection in Drosophila oocytes result from the interplay of actin and microtubules," Nat Commun, vol. 8, p. 1520, Nov 15 2017. [9] R. Pastore, G. Pesce, and M. Caggioni, "Differential Variance Analysis: a direct method to quantify and visualize dynamic heterogeneities," Sci Rep, vol. 7, p. 43496, Mar 14 2017. [10] M. Reufer, V. A. Martinez, P. Schurtenberger, and W. C. Poon, "Differential dynamic microscopy for anisotropic colloidal dynamics," Langmuir, vol. 28, pp. 4618-24, Mar 13 2012. [11] M. Raffel, C. E. Willert, F. Scarano, C. J. Kähler, S. T. Wereley, and J. Kompenhans, Particle image velocimetry: a practical guide, Chapters 10,16: Springer, 2018. [12] M. Raffel, C. E. Willert, F. Scarano, C. J. Kähler, S. T. Wereley, and J. Kompenhans, Particle image velocimetry: a practical guide, Chapters 4,5,6,7: Springer, 2018. [13] A. Drechsler, A. Synytska, P. Uhlmann, M. M. Elmahdy, M. Stamm, and F. Kremer, "Interaction Forces between Microsized Silica Particles and Weak Polyelectrolyte Brushes at Varying pH and Salt Concentration," Langmuir, vol. 26, pp. 6400-6410, May 4 2010. [14] P. Kukura, H. Ewers, C. Müller, A. Renn, A. Helenius, and V. Sandoghdar, "High- speed nanoscopic tracking of the position and orientation of a single virus," Nature methods, vol. 6, p. 923, 2009. [15] R. J. Adrian and C.-S. Yao, "Pulsed laser technique application to liquid and gaseous flows and the scattering power of seed materials," Applied Optics, vol. 24, pp. 44-52, 1985/01/01 1985. [16] N.-T. Nguyen and S. T. Wereley, Fundamentals and applications of microfluidics, 2nd ed. Boston: Artech House, 2006. [17] Y. Xia and G. M. Whitesides, "Soft lithography," Annual review of materials science, vol. 28, pp. 153-184, 1998. [18] J. Schindelin, C. T. Rueden, M. C. Hiner, and K. W. Eliceiri, "The ImageJ ecosystem: An open platform for biomedical image analysis," Molecular Reproduction and Development, vol. 82, pp. 518-529, 2015. [19] H. Bruus, Theoretical microfluidics vol. 18: Oxford university press Oxford, 2008. [20] J.-Y. Tinevez, J. Schindelin, J. Eglinger, and M. Hiner. Directionality plugin in ImageJ. Available: https://imagej.net/Directionality [21] Z. Q. Liu, "Scale space approach to directional analysis of images," Appl Opt, vol. 30, pp. 1369-73, Apr 10 1991. [22] K. Basu and K. Baishya, "Brownian motion: Theory and experiment a simple classroom measurement of the diffusion coefficient," Resonance, vol. 8, pp. 71-80, 2003. Supplementary Information Tracking-free Determination of Microparticle Motion from Image Variance Harish Sasikumar 1, and Manoj M. Varma 1 , 2 , 1 1Center for Nano Science and Engineering, Indian Institute of Science, Bangalore, 560012, India 2Robert Bosch Center for Cyber Physical Systems, Indian Institute of Science, Bangalore, 560012, India 1 Useful Formulae ∞ ∫ 𝑒− −∞ 𝑥2−𝐴𝑥 𝑅2 𝑑𝑥 = 𝑅√𝜋 e 𝐴2 4𝑅2 ∞ ∫ 𝑒− −∞ 𝑥2 𝑅2𝑑𝑥 = R√π (1) (2) 2 Image of a micro particle: Comparing Experimental observation and Simulation model 2.1 Theory Though the point spread function of a diffraction-limited case is an Airy distribution, for mathematical conveniences, it can be approximated to a Gaussian distribution1. Figure 1 Comparing Gaussian Plot and particle profile (a) Typical 10-micron polystyrene bead imaged with a 10x objective. The observation area is 22 × 22 pixels and the profile is smoothened, inverted and normalized. Using curve fitting, 𝜎 was found to be ~3.2 (b) Normalized Gaussian plot with 𝜎 = 3.2 and center at (11.5,11.5) For smoothening the image of micro particle, before fitting to Gaussian, the code given in 2 is used with a window size of 3 pixels. 1 Author to whom correspondence should be addressed. Electronic mail: [email protected] 3 Linear motion of multi-pixel particle with Gaussian profile 3.1 Theory Consider the image of a particle with centre at (𝑖𝑟, 𝑗𝑟) and standard deviation 𝑅, creating an intensity distribution in the image plane as follows (G) = Ae− xi,j (i−ir)2+(j−jr)2 2R2 (3) As shown in Figure 2(a), the particle moves from its initial position to the final position (along j axis) in 𝑁 time frames. The centre of the particle transverses m pixels during this time. Figure 2(b) shows the corresponding standard deviation obtained in the image plane. (Note that irrespective of the representation of the particle -- as dark entity in a bright field or bright entity in a dark field- the standard deviation will be the same) Figure 2 Multi-pixel particle performing linear motion (a) Particle at its initial, final and an in-between position (A=1). (b) a typical standard deviation plot of the linear motion. The value of a pixel at the 𝑖𝑡ℎ position along the centre of the particle (at time 𝑛) is given by xi,j,n = Ae− i2 2R2 At any other time-instance, k frames ahead, it can be given as xi,j,n+k = Ae− (i−k)2+k2 2R2 (4) (5) Hence the average value of a pixel, away from the initial or final positions can be approximated as, (G) xi,j = A N ∞ ∑ e− k=−∞ (i−k)2+k2 2R2 = A N ∞ −i2 2R2 ∑ e− k2−ik R2 e− k=−∞ (6) Note that, even when the sum is composed of infinite number of terms, the dying nature of the function makes the sum finite. Approximating the sum as integral, and using equation 1, the above equation can be simplified into Similarly, the average of the squares of the pixel value can be calculated as (G) xi,j = √π i2 4R2 e− AR N (7) 2 = xi,j A2 N ∞ ∑ e− k=−∞ (i−k)2+k2 R2 = A2 N ∞ −i2 R2 ∑ e−2 k2−ik R2 e− k=−∞ Which can be simplified (by replacing 𝑅 with 𝑅/√2 in equation 7), to, 2 = √ xi,j π 2 A2R N i2 2R2 e− (8) (9) Thus, the variance of the pixel at position (𝑖, 𝑗) can be derived using equations 7 and 9 as, 2 = xi,j σi,j 2 − (xi,j) 2 = A2 (√ π 2 R N − π R2 N2) e− i2 2R2 (10) And for a particle which has moved m pixels, the total standard deviation is given (using the approximation in equation 2) by ∞ σ = m ∑ σi,j = mA√(√ i=−∞ π 2 R N − π R2 N2) R√π (11) Note that, 𝑚 arises due to the length of the variance plot and the summation, due to the width of the particle. For 𝑝 particles with non-intersecting trajectories, the total standard deviation is given by, S = pσ = Apm√√ π3 2 R3 N (1 − √2π R N ) (12) 𝑚, the length of the variance plot arises due to the movement of particle and hence is related to the apparent velocity (𝑣𝑠 [pixel/frame]) of the particle and the total time frames 𝑁, by the following relation 𝑚 = 𝑁𝑣𝑠 The apparent velocity of the particle in the image space is related to the actual velocity (𝑣) by the scaling factors of the video imaging system: the dimension scaling factor 𝛾 [ 𝜇𝑚 𝑝𝑖𝑥𝑒𝑙 ] and the frame acquisition rate 𝑓 [ 𝑓𝑟𝑎𝑚𝑒𝑠 𝑠𝑒𝑐𝑜𝑛𝑑 ]. Hence, 𝑆 can be written as, 𝑣𝑠 = 𝑣 𝛾𝑓 S = Apv γf √√ π3 2 NR3 (1 − √2π R N ) (13) 3.2 Simulation Using equation 13, the velocity of the particles in a rectangular channel can be calculated from the total standard deviation (𝑆) calculated by summing in the vertical direction. 𝑣 = 𝛾𝑓𝑆 𝐴𝑝𝐶 ; 𝐶 = √√ π3 2 NR3 (1 − √2π R N ) Values used for the simulation Parameter Definition 𝛾 𝑓 𝐴 𝑅 𝑝 𝑁 dimension scaling factor Frame rate of image acquisition Height of the Gaussian (image of the particle) Standard deviation of Gaussian (image of the particle) Number of particles per unit pixel (along the width) Number of time frames Value or estimate 1.39 𝜇𝑚/𝑝𝑖𝑥𝑒𝑙 (Microscope specification) 10 frames/ second 33.58 (average obtained from the images of 5 particles) 3.5 (average obtained from the images of 5 particles) Total particles / width of the channel in pixels = 289 / 152 = 1.90 15 (1.5 seconds of data at 10 frames per second) 4 Soft-Lithography for the fabrication of microfluidic devices The processes for soft-lithography were adapted from 3. The steps were as following • PDMS elastomer and curing agent (SYLGARD 184, from Dow Corning) were taken in the weight ratio 10:1 in a clean, disposable cup. They were mixed thoroughly using a spatula for ~5 minutes and the mixture is kept in a low-pressure chamber (desiccator connected to a vacuum pump) for ~1 hour to degas out the bubbles. • The degassed mixture was poured onto the mold (master fabricated on silicon substrate) kept in a Teflon Petri dish and once again degassed. It was then cured at 90°C for ~3 hours over a hot plate. After curing, the PDMS structures were allowed to cool down to room temperature. • The PDMS structures were then peeled-off carefully from the mold. A biopsy punch of was used to drill holes at the outlet and inlet points. • The freshly prepared PDMS surfaces were exposed to Oxygen plasma for ~30 seconds using a 150 W RF plasma generator and with atmospheric air. The exposed PDMS surface were kept onto a gold coated glass slide and a small pressure was applied using a plastic tweezer. Further the bonded samples are baked at ~80°C for 20 minutes. 5 Histograms of Static Particle and Dynamic Particles on Si and PE The histograms were thresholded for pixel values above 3.5 to reject the inherent noises in the imaging system, including the CMOS sensors. Refer Figure 3 where the standard deviation of a 40 pixel × 40 pixel region involving a static microparticle is given. Figure 3. Standard deviation of a static particle Standard deviation (resulting from the noises in the imaging system) as of a 40 pixel × 40 pixel region involving a static microparticle. 14 regions, each containing the trajectory of single mobile particle for a time duration of 30 seconds (300 timeframes) were chosen. They were of size 40 pixel × 40 pixel and were chosen such that 7 were on Si and 7 were on silicon. The histogram of the pixel standard deviations were then calculated for each region. Average of seven of these histograms, were calculated and plotted. Figure 4 (a) and (b) show the histograms of particles on Si and PE regions respectively. Figure 4 Pixel standard deviation produced by mobile particles 14 regions of size 40 pixel × 40 pixel were chosen, 7 each on PE and silicon. Each of these regions contains the entire trajectory of one of the mobile particles for a time duration of 30 seconds (300 timeframes). (a) and (b) are the pixel standard deviations of these regions in Si and PE, respectively. (c) is the colorbar of the standard deviation. 6 Comparison of Surface Roughness -- Si v/s Polyelectrolyte For the comparison of surface roughness, surface profile of patterned PEM (coated on Si sample) was obtained using an Atomic Force Microscope (AFM). A typical interface of the circular PEM pattern and Si is given in Figure 5. Figure 5. Comparison of roughness in PE and Si regions Uniform regions of size 5 𝜇𝑚 × 5 𝜇𝑚 were taken on the PEM and Si surfaces and the root mean square average of height deviation (𝑅𝑞) was calculated. 𝑅𝑞 on PE and Si regions were found to be 2.80 nm and 1.44 nm respectively. References 1 2 3 Ronald J. Adrian and Chung-Sheng Yao, Applied optics 24 (1), 44 (1985). Greg Reeves, (Mathworks, File Exchange, 2009). Dong Qin, Younan Xia, and George M. Whitesides, Nature Protocols 5, 491 (2010); Younan Xia and George M Whitesides, Angewandte Chemie International Edition 37 (5), 550 (1998); Xueye Chen and Lei Zhang, Sensors and Actuators B: Chemical 254, 648 (2018).
1912.06940	1	1912	2019-12-14T22:17:46	Tunable telecom to mid-infrared optical parametric oscillation via microring-based $\chi^{(3)}$ nonlinearities	[ "physics.app-ph", "physics.optics" ]	Optical parametric oscillation (OPO) with far-shifted frequency sidebands has attracted significant interests in precision spectroscopy and quantum information processing. Microresonator based OPO sources hold the advantages of miniaturized footprint and versatile dispersion engineering. Here we demonstrate large-frequency-shifted $\chi^{(3)}$-based OPO from crystalline aluminum nitride microrings pumped at $\sim$2 $\mu$m in the normal dispersion regime. OPO in the telecom and mid-infrared bands with a frequency separation of 65.5 THz is achieved. The OPO frequency can be agilely tuned in the ranges of 10, 1 and 0.1 THz respectively by tailoring the microring dimensions, shifting the pump wavelength, and controlling the chip temperature. At high intracavity pump powers, the OPO sidebands further evolve into localized frequency comb lines. Such telecom to mid-infrared OPO with flexible wavelength tunability will lead to enhanced chip-scale light sources.	physics.app-ph	physics	Letter Optics Letters 1 9 1 0 2 c e D 4 1 ] h p - p p a . s c i s y h p [ 1 v 0 4 9 6 0 . 2 1 9 1 : v i X r a Tunable telecom to mid-infrared optical parametric oscillation via microring-based χ(3) nonlinearities YULONG TANG1,#,†, ZHENG GONG1,#, XIANGWEN LIU1, AND HONG X. TANG1,* 1Department of Electrical Engineering, Yale University, New Haven, Connecticut 06520, USA *Corresponding author: [email protected] †Present address: School of Physics and Astronomy, Key Laboratory for Laser Plasmas (MOE), Collaborative Innovation Center of IFSA,Shanghai Jiao Tong University, Shanghai 200240, China; email:[email protected]. #These authors contributed equally Compiled December 17, 2019 Optical parametric oscillation (OPO) with far-shifted frequency sidebands has attracted significant interests in precision spectroscopy and quantum information processing. Microresonator based OPO sources hold the advantages of miniaturized footprint and versatile dispersion engineering. Here we demonstrate large- frequency-shifted χ(3)-based OPO from crystalline alu- minum nitride microrings pumped at ∼2 µm in the nor- mal dispersion regime. OPO in the telecom and mid- infrared bands with a frequency separation of 65.5 THz is achieved. The OPO frequency can be agilely tuned in the ranges of 10, 1 and 0.1 THz respectively by tailor- ing the microring dimensions, shifting the pump wave- length, and controlling the chip temperature. At high intracavity pump powers, the OPO sidebands further evolve into localized frequency comb lines. Such tele- com to mid-infrared OPO with flexible wavelength tun- ability will lead to enhanced chip-scale light sources. © 2019 Optical Society of America OCIS codes: mixing; (140.3948) Microcavity devices. (190.0190) Nonlinear optics; (190.4223) Nonlinear wave http://dx.doi.org/10.1364/ao.XX.XXXXXX Broadband and widely-tunable coherent light sources are essen- tial tools for various applications [1]. While the available emis- sion spectra of conventional lasers are set by their gain media, nonlinear optical frequency conversions help generate coherent light ranging from ultraviolet to mid-infrared (mid-IR) regions [2]. Among these nonlinear optical processes, χ(3)-based optical parametric oscillation (OPO) features the generation of coherent frequency components with two sidebands equally separated from the pump and has attracted significant interest for its high conversion efficiency and large sideband frequency tunability [3, 4]. Such OPOs have been extensively studied in silica fiber cavities [5, 6], but with limited bandwidth due to inconvenient dispersion engineering and high pump power requirement. On the other hand, optical microresonators, possessing high quality-factors (Qs) as well as small mode volumes, have proven as excellent platforms for ultra-low threshold OPO generation [7, 8]. Meanwhile, the versatile and precise dispersion engineer- ing in microresonators could facilitate large-frequency-shifted or octave-spanning OPOs in the desired spectrum regions [4]. For instance, by tailoring higher-order dispersion (even orders) while pumping in the normal dispersion regime, OPO side- bands extending to mid-IR regimes of 2 and 2.4 um have been achieved in silica and magnesium fluoride microresonators with corresponding frequency shifts of 85 and 140 THz, respectively [9, 10]. Such broadband OPO sources could enable applications in chemical and molecule detection owing to fluent character- istic vibrational and rotational transitions of many important molecules lie in the mid-IR regime (2.5-20 µm) [11]. Apart from whispering gallery microcavities, nanophotonic microring resonators based on silicon nitride (Si3N4) platforms have also been employed to generate low threshold OPO [12] in the telecom band [13], near infrared [3], and even down to the visible wavelength region [14]. However, to the best of our knowledge, widely-tunable OPO sidebands that extend to the mid-IR regime have not been demonstrated in a nanophotonic platform. Recently, aluminum nitride (AlN) has emerged as an appealing material for integrated nonlinear photonics thanks to its strong quadratic (χ(2)) and cubic (χ(3)) susceptibilities [15 -- 17]. Single-crystalline AlN, in particular, has shown improved device performance [18, 19] and has stimulated progresses including frequency comb generation [20, 21], second harmonic conversion [22], ultraviolet supercontinuum combs [23] and efficient OPO sources [8]. While the extent of these advances spans from the ultraviolet to near-infrared regimes, the broad transparency of AlN (from 200 nm to 10 µm) [15, 24] also allows for nonlinear frequency conversion into the mid-IR region. In this letter, we demonstrate large frequency-shifted χ(3) OPOs in crystalline AlN microrings pumped at ∼2 µm. By tailoring the microring dimensions, desirable phase-match con- ditions can be obtained for OPO generation at the telecom and mid-IR regions with a frequency separation of 65.5 THz. We then show that the OPO frequency sidebands can be tuned by ∼10 THz over a varied microring width of 200 nm, which is con- sistent with our numerical simulations. Further fine-tuning of the OPO sideband frequencies is realized by varying the pump Letter Optics Letters 2 wavelength and controlling the chip temperature with tuning scales of ∼1 THz and ∼0.1 THz, respectively. Our microring resonators are made out of a 1-µm thick epi- taxial AlN film grown on a c-plane sapphire substrate via metal organic chemical vapor deposition [25]. The microring patterns are defined by 100 kV electron beam lithography and trans- ferred into the AlN film using a two-step dry etching process [20, 26]. At last, the devices are encapsulated within ∼1.5-µm thick silicon dioxide (SiO2) via plasma enhanced chemical vapor deposition. Figure 1(a) shows the scanning electron microscope (SEM) image of the fabricated straight waveguide-coupled mi- croring resonator. And Fig. 1(b) schematically illustrates the OPO generation process in a Kerr microcavity, where the pump field inside the cavity gives rise to oscillations at two new fre- quency components based on 2ωp = ωs + ωi. Such process can be enhanced if ωs and ωi overlap the microring resonances. Fig. 1. (a) SEM image of the AlN microring with a radius R=50 µm. (b) Schematic of the χ(3)-based OPO sideband gen- eration in the microcavities with ωp, ωs and ωi respectively being the pump, signal and idler frequency. (c,d) Simulated TE00 -- mode integrated dispersion (blue) and phase matching curves (magenta) for the microing width W=3.2, 3.4 µm and R=50 µm using the finite element method, where the crossing- zero points of the magenta curve indicate the phase-matched modes for effective OPO. For the device design, we fix the microring radius at 50 µm while tailoring the microring width to adjust OPO phase- matching conditions. After considering up to the 4th order dis- persion, the cavity modes frequencies can be expanded around the pump angular frequency ω0 as [10] ωµ = ω0 + D1µ + (1/2!)D2µ2 + (1/3!)D3µ3 + (1/4!)D4µ4, where µ is the mode number relative to the pump mode and Di denotes the ith or- der dispersion coefficient at ω0. Meanwhile, the OPO side- bands are determined by the phase-matching condition [10]: ω0+µ + ω0−µ − 2ω0 = D2µ2 + (1/12)D4µ4 (cid:39) 0. To achieve widely-separated OPOs, we tailor the microring geometries to have D2 < 0 and D4 > 0 so that the phase-matching curve of (ω0+µ + ω0−µ − 2ω0)/2π can be bent to have cross-zero points far away from the pump. Figures 1(c, d) show the simulated Dint (blue) and phase-matching curves (magenta) for the TE00 mode in the microrings with widths of W=3.2 and 3.4 µm, re- spectively. For the 3.2-µm-wide micoring, the phase matching points are at ∼1.71 and ∼2.31 µm, corresponding to µ ≈ ±52 with D2/(2π) ≈-12 MHz and D4/(2π) ≈59 kHz; while the 3.4-µm-wide microring exhibits phase-matching points at ∼1.61 and ∼2.51 µm, corresponding to µ ≈ ±73 with D2/(2π) ≈- 26 MHz and D4/(2π) ≈57 kHz. The larger frequency-shifted phase-matching points from the pump in wider microrings is mainly due to the decrease of D2 at similar D4. The experimental setup is schematically presented in Fig. 2(a). The pump source was a narrow-linewidth tunable seed laser (1940-1980 nm) amplified with a thulium-doped fiber amplifier (TDFA). A fiber-based polarization controller (FPC) is used to control the input light polarization before launching it into the waveguide with a mid-IR aspheric lens pair. And the output signal, collected with another aspheric lens pair, is sent into an optical spectrum analyzer (Yokogawa AQ6376, 1500-3400 µm). The total transmission efficiency of the chip is ∼10%. Figures 2(b- e) present the measured transmission spectra of the microrings (radius of R=50 µm, widths of W=3.2, 3.4 µm), where the free spectral range (FSR) is ∼450 GHz and the extracted loaded Q- factors are ∼480 k and ∼570 k around the pump wavelengths, respectively. Along with experimental investigation, numerical simula- tions are also carried out to study the OPO process under the dispersion profiles in Figs.1(c, d) based on the coupled mode equations [27]: ∂Aµ ∂t = − κ0 + κext ig ∑ + δµ0 2 Aµ(cid:48) Aµ(cid:48)(cid:48) A(cid:63) µ(cid:48),µ(cid:48)(cid:48),µ(cid:48)(cid:48)(cid:48) √ κextSine−i(ωp−ω0)t+ µ(cid:48)(cid:48)(cid:48) e−i(ωµ(cid:48) +ωµ(cid:48)(cid:48)−ωµ(cid:48)(cid:48)(cid:48)−ωµ)t. (1) Here, Aµ is the amplitude of the mode with a relative mode number µ, Aµ2 describe the photon number of each mode, and t is the time. κ0 and κext denote the intrinsic decay and coupling rates of the cavity, respectively. Sin = Pin/¯hω0 is the amplitude of the pump power, and δµ0 is the Kronecker delta. The summation term includes all µ(cid:48), µ(cid:48)(cid:48), µ(cid:48)(cid:48)(cid:48) satisfying the relation µ = µ(cid:48) + µ(cid:48)(cid:48) − µ(cid:48)(cid:48)(cid:48). The cubic Kerr nonlinearity of the system is described by the nonlinear coupling coefficient: √ g = 2cn2 ¯hω0 n02Ve f f , (2) √ where n0 is the refractive index, n2 is the Kerr nonlinear refrac- tive index, c is the light speed in vacuum, ¯h is the reduced Planck constant, and Ve f f is the effective cavity mode volume. The am- plitude of the output power from the cavity Sout = Pout/¯hω0 can be obtained by: κext ∑ Aµe−i(ωµ−ωp)t. Sout = Sin − √ In simulation, similar parameters as in the experiment are adopted, and normalized pump-resonance detuning 2(ωp − ω0)/(κ0 + κext) is scanned to trigger OPO. Figures 3(a, b) show the measured OPO spectra from the microrings with an on-chip pump power of 375 mW. For the 3.2-µm-wide microring, the OPO sidebands show up at 1711.6 (3) -50-2502550-50-2502550 (GHz) Dint Phase matchDint/2(GHz)W=3.2m 16001800200022002400-50-2502550-50-2502550 (GHz)Dint/2(GHz)Wavelength (nm) Dint Phase matchW=3.4m20μm Pump ωp (b) (c) (a) ω Optical frequency Power Signal ωs Idler ωi Resonances (b) (c) (d) Letter Optics Letters 3 tuned by varying the pump wavelength within the D2<0 regime. Figure 4 illustrates how the OPO sidebands generate in the 3.4-µm-wide microring when blue-shifting the pump from one resonance to another. As can be seen, by decreasing the pump wavelength, the signal (idler) sideband shifts to the red (blue) spectral region, resulting in a widened spacing between the signal/idler modes. In this case, the signal (idler) frequency can be shifted by ∼3 (∼2.5) THz in total as the pump is tuned by ∼2.1 THz. Such OPO frequencies tuning effect is also numerically simulated and plotted as the star dots in Fig. 4, which trend the experimental results. Fig. 2. (a) The experimental setup. ECDL: external-cavity diode laser; TDFA: thulium-doped fiber amplifier; FPC: fiber polarization controller; OSA: optical spectrum analyzer. (b,c) TE00 mode transmission spectra for microrings W=3.2, 3.4 µm and R=50 µm. (d,e) The zoom-in views of the resonances at ∼1965.97 and ∼1962.43 nm in (b,c) respectively; dots show the measured data while the curve is Lorentz fitting. and 2312 nm, which agrees well with the cross-zero points of the phase-matching curve in Fig. 1(c), as well as the simulated side- bands in Fig. 3(c). The total frequency shift from the signal (at 2312 nm) to the idler (at 1711.6 nm) is over 45.5 THz (around 600 nm). Upon increasing the microring width to 3.4 µm, extended OPO frequency shift is obtained which is also consistent with the simulation in Fig.3 (d). In this case, the two OPO sidebands are observed to appear at 1618.2 and 2503 nm with a total frequency separation of over 65.5 THz (>880 nm), which is 20 THz broader than that in the 3.2-µm-wide device. These results indicate that frequency tuning of the OPO sidebands over ∼10 THz can be achieved in an AlN nanophotonic platform by adjusting the microring width of 200 nm. Fig. 4. OPO sidebands tuning in the microring of W=3.4 µm and R=50 µm by tuning the pump laser to different reso- nances. Star dots show the simulated OPO sideband positions under the same pump wavelengths as the experiments. Fig. 5. Temperature tuning of OPO signal and idler sidebands in the microring of W=3.4 µm and R=50 µm. The pump wave- length is adjusted accordingly as the temperature is varied. νp0, νi0 and νs0 are the pump frequency and that of the two generated OPO sidebands when the chip's temperature is at 24 °C, corresponding to wavelengths of 1962.8, 1618.2, and 2493.8 nm. The error bar results from the OSA's resolution (∼0.1 nm). While the above OPO frequency tuning mechanisms (Figs. 3 and 4) provide discrete and relative large tuning steps, quasi- continuous frequency tuning can also be achieved by utilizing the thermal-optic effect in the AlN microrings [20]. For this purpose, we mounted an external heater (precision of 0.5 °C) below the chip for effective temperature tuning and recorded the temperature-dependent OPO spectra simultaneously. The observed tuning of the signal/idler frequencies under fixed Fig. 3. Measured OPO spectra from the microrings of R=50 µm, W=3.2 µm (a) and W=3.4 µm (b). (c,d) Simulated OPO spectra corresponding to (a,b) under the Dint and phase- matching curves shown in Fig.1(c,d). Apart from controlling the dispersion profile via tailoring the microring geometries, the OPO sideband frequencies can also be ECDL TDFA FPC OSA Micro-ring Lenses Lenses (a) 196019651970197519801985-6-4-202 Transmission (dB)1962.421962.431962.441962.450.20.40.60.81.01.2 1940194519501955196019651970-6-4-202 1965.951965.961965.971965.980.40.60.81.0 Wavelength (nm)Q~480kW=3.4 mW=3.2 mQ~570kTransmission (a.u.)Wavelength (nm)(b) (c) (d) (e) 160018002000220024002600 1711.6nm1967nm2312nm160018002000220024002600 160018002000220024002600 160018002000220024002600 W=3.4mW=3.4mW=3.2mWavelength (nm)Intensity (dBm, 20dBm/scale)(d)(a)(b)(c)ExperimentSimulationWavelength (nm)1962.7nm1618.2nm2503nmW=3.2m 160018002000220024002600 1974.2nm1637.8nm2484.4nm 1626.0nm1968.4nm2493.6nm Intensity (20dBm/scale)2493.6nm1962.7nm1618.1nm Wavelength (nm)1606.5nm1957.0nm2503.0nm 2030405060708090020406080100120140 Signal shift Idler shift Pump shiftns0-ns/ni0-ni/np0-np(GHz)Temperature (Deg.)Letter Optics Letters 4 pump power are summarized in Fig. 5. By varying the heater temperature from 24 to 90 °C, the signal and idler sidebands red- shift by 82.9 GHz (1.72 nm) and 137.4 GHz (1.2 nm), respectively. Or equivalently, the OPO sideband frequencies can be tuned at a step of ∼0.75 GHz (0.013 nm) for the signal and ∼1.25 GHz (0.009 nm) for the idler, which is limited by the 0.5 °C precision of the temperature controller and if desired can be further improved by embedding the device in a ovenized heater. By further tuning the pump into the resonances, single-FSR- spaced four-wave mixing (FWM) comb lines are subsequently generated around the pump and the primary OPO sidebands. These comb spectra are shown in Figs. 6(a,b). Note that the 3.2- µm-wide microring generates stronger comb lines that emerge above the OSA noise floor over a full 800 nm wavelength span, as shown in Fig. 6(a), while the comb lines from the 3.4-µm-wide microring (Fig. 6(b)) are clustered around the signal, pump and idler which are widely separated. Figures 6(c,d) are the simu- lated comb lines for the two microrings, showing good agree- ment with experimental observations. In our experiment and simulation, the involved comb lines are found to be in incoher- ent/noisy states. Similar localized frequency comb generation has been observed in crystalline MgF2 microresonators [28]. Fig. 6. Experimentally measured spectra of comb structures in the microring with R=50 µm and W=3.2 µm (a) and W=3.4 µm (b). (c,d) show the corresponding simulated comb spectra under on-chip pump power of 300 mW. In conclusion, we have realized widely-frequency-shifted (from ∼45.5 to ∼65.5 THz) OPO sources in integrated crystalline AlN microrings pumped at ∼ 2µm, with sidebands extending to the telecom band and the mid-IR region. The experimental re- sults agree well with the numerical simulations. The dispersion engineering via tailoring the microring width modifies the OPO phase-matching condition and gives rise to frequency tuning of the OPO sidebands with offset up to 10-THz. In addition, the OPO sideband frequencies can also be fine-tuned in the 1- THz and 0.1-THz scales by shifting the pump wavelength and applying temperature control, respectively. We believe such wide-spanning on-chip OPO sources with flexible wavelength tunability would enable new capabilities for spectroscopy sens- ing and nonlinear frequency conversion based on integrated nanophotonic circuits. ACKNOWLEDGEMENT Yulong Tang acknowledges China Scholarship Council for the visiting research in Yale University. FUNDING INFORMATION This work is supported by DARPA SCOUT (W31P4Q-15-1-0006). H.X. Tang acknowledges partial support from DARPA's ACES programs as part of the Draper-NIST collaboration (HR0011-16- C-0118), an AFOSR MURI grant (FA9550-15-1-0029), a LPS/ARO grant (W911NF-14-1-0563), a NSF EFRI grant (EFMA-1640959) and Packard Foundation. DISCLOSURES The authors declare no conflicts of interest. REFERENCES 1. 2. 3. Q. Li, M. Davanco, and K. Srinivasan, Nat. Photon. 10, 406 (2016). 4. F. Duarte, Tunable Laser Applications (CRC Press, 2008). N. Savage, Nat. Photon. 4, 124 (2010). N. Sayson, T. Bi, V. Ng, H. Pham, L. Trainor, H. Schwefel, S. Coen, M. Erkintalo, and S. Murdoch, Nat. Photon. 13, 701 (2019). 5. G. K. L. Wong, S. G. Murdoch, R. Leonhardt, J. D. Harvey, and V. Marie, 6. 7. 8. 9. Opt. Express 15, 2947 (2007). F. Bessin, F. Copie, M. Conforti, A. Kudlinski, and A. Mussot, Opt. Lett. 42, 3730 (2017). T. Kippenberg, S. Spillane, and K. J. Vahala, Phy. Rev. Lett. 93, 083904 (2004). A. W. Bruch, X. Liu, J. B. Surya, C.-L. Zou, and H. X. Tang, Optica. 6, 1361 (2019). N. L. B. Sayson, K. E. Webb, S. Coen, M. Erkintalo, and S. G. Murdoch, Opt. Lett. 42, 5190 (2017). 10. S. Fujii, S. Tanaka, M. Fuchida, H. Amano, Y. Hayama, R. Suzuki, Y. Kakinuma, and T. Tanabe, Opt. Lett. 44, 3146 (2019). 11. C. Lecaplain, C. Javerzac-Galy, M. L. Gorodetsky, and T. J. Kippenberg, Nat. Commun. 7, 13383 (2016). 12. X. Ji, F. A. S. Barbosa, S. P. Roberts, A. Dutt, J. Cardenas, Y. Okawachi, A. Bryant, A. L. Gaeta, and M. Lipson, Optica. 4, 619 (2017). 13. S. Ramelow, A. Farsi, Z. Vernon, S. Clemmen, X. Ji, J. Sipe, M. Liscidini, M. Lipson, and A. Gaeta, Phys. Rev. Lett. 122, 153906 (2019). 14. X. Lu, G. Moille, Q. Li, D. Westly, A. Singh, A. Rao, S. Yu, T. Briles, S. Papp, and K. Srinivasan, Nat. Photon. 13, 593 (2019). 15. C. Xiong, W. H. P. Pernice, and H. X. Tang, Nano Lett. 12, 3562 (2012). 16. X. Guo, C.-L. Zou, and H. X. Tang, Optica. 3, 1126 (2016). 17. J. B. Surya, X. Guo, C.-L. Zou, and H. X. Tang, Optica. 5, 103 (2018). 18. X. Liu, C. Sun, B. Xiong, L. Wang, J. Wang, Y. Han, Z. Hao, H. Li, Y. Luo, J. Yan, T. Wei, Y. Zhang, and J. Wang, Opt. Express 25, 587 (2017). 19. X. Liu, A. W. Bruch, Z. Gong, J. Lu, J. B. Surya, L. Zhang, J. Wang, J. Yan, and H. X. Tang, Optica 5, 1279 (2018). 20. Z. Gong, A. Bruch, M. Shen, X. Guo, H. Jung, L. Fan, X. Liu, L. Zhang, J. Wang, J. Li, J. Yan, and H. X. Tang, Opt. Lett. 43, 4366 (2018). 21. X. Liu, C. Sun, B. Xiong, L. Wang, J. Wang, Y. Han, Z. Hao, H. Li, Y. Luo, J. Yan, T. Wei, Y. Zhang, and J. Wang, ACS Photonics 5, 1943 (2018). 22. A. W. Bruch, X. Liu, X. Guo, J. B. Surya, Z. Gong, L. Zhang, J. Wang, J. Yan, and H. X. Tang, Appl. Phys. Lett. 113, 131102 (2018). 23. X. Liu, A. Bruch, J. Lu, Z. Gong, J. Surya, L. Zhang, J. Wang, J. Yan, and H. Tang, Appl. Phys. Lett. 10, 2971 (2019). 24. P. T. Lin, H. Jung, L. C. Kimerling, A. Agarwal, and H. X. Tang, Laser Photonics Rev. 8, L23 (2014). 25. Y. Jianchang, W. Junxi, L. Naixin, L. Zhe, R. Jun, and L. Jinmin, J. Semicond. 30, 103001 (2009). 26. L. Fan, K. Y. Fong, M. Poot, and H. Tang, Nat. Commun. 6, 5850 (2015). 27. T. Herr, V. Brasch, J. D. Jost, C. Y.Wang, N. M. Kondratiev, M. L. Gorodetsky, and T. J. Kippenberg, Nat. Photon. 8, 145 (2014). 28. N. L. B. Sayson, H. Pham, K. E. Webb, V. Ng, L. S. Trainor, H. G. L. Schwefel, S. Coen, M. Erkintalo, and S. G. Murdoch, Opt. Lett. 43, 4180 (2018). 160018002000220024002600-60-40-200 160018002000220024002600-60-40-200 160018002000220024002600-60-40-200 1967.28nm 160018002000220024002600-60-40-200W=3.4mW=3.4mW=3.2mWavelength (nm)Intensity (dBm)(d)(a)(b)(c)ExperimentSimulationWavelength (nm)1962.7nmW=3.2mOptics Letters 5 22. A. W. Bruch, X. Liu, X. Guo, J. B. Surya, Z. Gong, L. Zhang, J. Wang, J. Yan, and H. X. Tang, "17000%/w second-harmonic conversion ef- ficiency in single-crystalline aluminum nitride microresonators," Appl. Phys. Lett. 113, 131102 (2018). 23. X. Liu, A. Bruch, J. Lu, Z. Gong, J. Surya, L. Zhang, J. Wang, J. Yan, and H. Tang, "17000%/w second-harmonic conversion efficiency in single-crystalline aluminum nitride microresonators," Appl. Phys. Lett. 10, 2971 (2019). 24. P. T. Lin, H. Jung, L. C. Kimerling, A. Agarwal, and H. X. Tang, "Low- loss aluminium nitride thin film for mid-infrared microphotonics," Laser Photonics Rev. 8, L23 -- L28 (2014). 25. Y. Jianchang, W. Junxi, L. Naixin, L. Zhe, R. Jun, and L. Jinmin, "High quality AlGaN grown on a high temperature AIN template by MOCVD," J. Semicond. 30, 103001 (2009). 26. L. Fan, K. Y. Fong, M. Poot, and H. Tang, "Cascaded optical trans- parency in multimode-cavity optomechanical systems," Nat. Commun. 6, 5850 (2015). 27. T. Herr, V. Brasch, J. D. Jost, C. Y.Wang, N. M. Kondratiev, M. L. Gorodetsky, and T. J. Kippenberg, "Temporal solitons in optical mi- croresonators," Nat. Photon. 8, 145 -- 152 (2014). 28. N. L. B. Sayson, H. Pham, K. E. Webb, V. Ng, L. S. Trainor, H. G. L. Schwefel, S. Coen, M. Erkintalo, and S. G. Murdoch, "Origins of clus- tered frequency combs in kerr microresonators," Opt. Lett. 43, 4180 -- 4183 (2018). Letter FULL REFERENCES 1. 2. F. Duarte, Tunable Laser Applications (CRC Press, 2008). N. Savage, "Optical parametric oscillators," Nat. Photon. 4, 124 -- 125 (2010). 3. Q. Li, M. Davanco, and K. Srinivasan, "Efficient and low-noise single- photon-level frequency conversion interfaces using silicon nanophoton- ics," Nat. Photon. 10, 406 -- 415 (2016). N. Sayson, T. Bi, V. Ng, H. Pham, L. Trainor, H. Schwefel, S. Coen, M. Erkintalo, and S. Murdoch, "Octave-spanning tunable parametric oscillation in crystalline kerr microresonators," Nat. Photon. 13, 701 -- 706 (2019). 4. 6. 5. G. K. L. Wong, S. G. Murdoch, R. Leonhardt, J. D. Harvey, and V. Marie, "High-conversion-efficiency widely-tunable all-fiber optical parametric oscillator," Opt. Express 15, 2947 -- 2952 (2007). F. Bessin, F. Copie, M. Conforti, A. Kudlinski, and A. Mussot, "Modu- lation instability in the weak normal dispersion region of passive fiber ring cavities," Opt. Lett. 42, 3730 -- 3733 (2017). T. Kippenberg, S. Spillane, and K. J. Vahala, "Kerr-nonlinearity optical parametric oscillation in an ultrahigh-q toroid microcavity," Phy. Rev. Lett. 93, 083904 (2004). A. W. Bruch, X. Liu, J. B. Surya, C.-L. Zou, and H. X. Tang, "On-chip χ (2) microring optical parametric oscillator," Optica. 6, 1361 -- 1366 (2019). N. L. B. Sayson, K. E. Webb, S. Coen, M. Erkintalo, and S. G. Murdoch, "Widely tunable optical parametric oscillation in a kerr microresonator," Opt. Lett. 42, 5190 -- 5193 (2017). 9. 7. 8. 10. S. Fujii, S. Tanaka, M. Fuchida, H. Amano, Y. Hayama, R. Suzuki, Y. Kakinuma, and T. Tanabe, "Octave-wide phase-matched four-wave mixing in dispersion-engineered crystalline microresonators," Opt. Lett. 44, 3146 -- 3149 (2019). 11. C. Lecaplain, C. Javerzac-Galy, M. L. Gorodetsky, and T. J. Kippen- berg, "Mid-infrared ultra-high-q resonators based on fluoride crystalline materials," Nat. Commun. 7, 13383 (2016). 12. X. Ji, F. A. S. Barbosa, S. P. Roberts, A. Dutt, J. Cardenas, Y. Okawachi, A. Bryant, A. L. Gaeta, and M. Lipson, "Ultra-low-loss on-chip res- onators with sub-milliwatt parametric oscillation threshold," Optica. 4, 619 -- 624 (2017). 13. S. Ramelow, A. Farsi, Z. Vernon, S. Clemmen, X. Ji, J. Sipe, M. Liscidini, M. Lipson, and A. Gaeta, "Strong nonlinear coupling in a si3n4 ring resonator," Phys. Rev. Lett. 122, 153906 (2019). 14. X. Lu, G. Moille, Q. Li, D. Westly, A. Singh, A. Rao, S. Yu, T. Briles, S. Papp, and K. Srinivasan, "Efficient telecom-to-visible spectral trans- lation through ultralow power nonlinear nanophotonics," Nat. Photon. 13, 593 -- 601 (2019). 15. C. Xiong, W. H. P. Pernice, and H. X. Tang, "Low-loss, silicon integrated, aluminum nitride photonic circuits and their use for electro-optic signal processing," Nano Lett. 12, 3562 -- 3568 (2012). 16. X. Guo, C.-L. Zou, and H. X. Tang, "Second-harmonic generation in aluminum nitride microrings with 2500%/w conversion efficiency," Optica. 3, 1126 -- 1131 (2016). 17. J. B. Surya, X. Guo, C.-L. Zou, and H. X. Tang, "Efficient third-harmonic generation in composite aluminum nitride/silicon nitride microrings," Optica. 5, 103 -- 108 (2018). 18. X. Liu, C. Sun, B. Xiong, L. Wang, J. Wang, Y. Han, Z. Hao, H. Li, Y. Luo, J. Yan, T. Wei, Y. Zhang, and J. Wang, "Aluminum nitride-on- sapphire platform for integrated high-q microresonators," Opt. Express 25, 587 -- 594 (2017). 19. X. Liu, A. W. Bruch, Z. Gong, J. Lu, J. B. Surya, L. Zhang, J. Wang, J. Yan, and H. X. Tang, "Ultra-high-q uv microring resonators based on a single-crystalline aln platform," Optica 5, 1279 -- 1282 (2018). 20. Z. Gong, A. Bruch, M. Shen, X. Guo, H. Jung, L. Fan, X. Liu, L. Zhang, J. Wang, J. Li, J. Yan, and H. X. Tang, "High-fidelity cavity soliton generation in crystalline aln micro-ring resonators," Opt. Lett. 43, 4366 -- 4369 (2018). 21. X. Liu, C. Sun, B. Xiong, L. Wang, J. Wang, Y. Han, Z. Hao, H. Li, Y. Luo, J. Yan, T. Wei, Y. Zhang, and J. Wang, "Integrated high-q crystalline aln microresonators for broadband kerr and raman frequency combs," ACS Photonics 5, 1943 -- 1950 (2018).
1801.09761	1	1801	2018-01-22T21:43:40	Degradation Kinetics of Inverted Perovskite Solar Cells	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We explore the degradation behaviour under continuous illumination and direct oxygen exposure of inverted unencapsulated formamidinium(FA)0.83Cs0.17Pb(I0.8Br0.2)3, CH3NH3PbI3, and CH3NH3PbI3-xClx perovskite solar cells. We continuously test the devices in-situ and in-operando with current-voltage sweeps, transient photocurrent, and transient photovoltage measurements, and find that degradation in the CH3NH3PbI3-xClx solar cells due to oxygen exposure occurs over shorter timescales than FA0.83Cs0.17Pb(I0.8Br0.2)3 mixed-cation devices. We attribute these oxygen-induced losses in the power conversion efficiencies to the formation of electron traps within the perovskite photoactive layer. Our results highlight that the formamidinium-caesium mixed-cation perovskites are much less sensitive to oxygen-induced degradation than the methylammonium-based perovskite cells, and that further improvements in perovskite solar cell stability should focus on the mitigation of trap generation during ageing.	physics.app-ph	physics	Degradation Kinetics of Inverted Perovskite Solar Cells Mejd Alsari1, Andrew J. Pearson1, Jacob Tse-Wei Wang2,3, Zhiping Wang2, Augusto Montisci4, Neil C. Greenham1, Henry J. Snaith2, Samuele Lilliu5,6, Richard H. Friend1 1 Cavendish Laboratory, University of Cambridge, CB30HE Cambridge, UK 2 Clarendon Laboratory, Department of Physics, University of Oxford, OX1 3PU Oxford, UK 3 CSIRO Energy, Mayfield West, NSW 2304, Australia 4 University of Cagliari, Dept. of Electrical and Electronic Engineering, 09123 Cagliari, Italy 5 Department of Physics and Astronomy, University of Sheffield, S3 7RH Sheffield, UK 6 The UAE Centre for Crystallography, UAE * Correspondence to: [email protected], [email protected] Abstract We explore the degradation behaviour under continuous illumination and direct oxygen exposure of inverted unencapsulated formamidinium(FA)0.83Cs0.17Pb(I0.8Br0.2)3, CH3NH3PbI3, and CH3NH3PbI3-xClx perovskite solar cells. We continuously test the devices in-situ and in-operando with current-voltage sweeps, transient photocurrent, and transient photovoltage measurements, and find that degradation in the CH3NH3PbI3-xClx solar cells due to oxygen exposure occurs over shorter timescales than FA0.83Cs0.17Pb(I0.8Br0.2)3 mixed-cation devices. We attribute these oxygen-induced losses in the power conversion efficiencies to the formation of electron traps within the perovskite photoactive layer. Our results highlight that the formamidinium-caesium mixed-cation perovskites are much less sensitive to oxygen-induced degradation than the methylammonium- based perovskite cells, and that further improvements in perovskite solar cell stability should focus on the mitigation of trap generation during ageing. Keywords Perovskite, Stability, Inverted, Oxygen, Solar Cell The relatively high power conversion efficiency (PCE)[1] of perovskite solar cells (PSCs) combined with their potential for low-cost production[2] and their outstanding opto-electronic properties such as band-gap tuneability,[3] long charge diffusion length[4], low recombination rates,[5] and photon recycling,[6, 7] would make these devices ready for the PV market, although long-term stability remains a concern.[8] PSCs degradation can take place in the light-absorbing perovskite layer and/or in any intermediate layers, which can degrade due to their intrinsic structural instability and/or due to external factors, such as oxygen, moisture, heat, electrical bias, and mechanical stress.[9] Research into the degradation mechanisms of PSCs has so far predominantly focussed on regular n-i-p architectures. Inverted p-i-n devices can potentially outclass the n-i-p stack, both in terms of efficiency and stability, provided that stable n-type materials can be identified.[8] In this work, we explore the degradation kinetics of unencapsulated inverted p-i-n PSCs employing the benchmark CH3NH3PbI3 and CH3NH3PbI3-xClx, and a more thermally durable alternative FA0.83Cs0.17Pb(I0.8Br0.2)3 perovskites as the photoactive layers.[10, 11] Recently, we investigated the degradation kinetics of unencapsulated regular CH3NH3PbI3-xClx (MAPIC) PSCs under continuous illumination in dry N2 (stabilization phase) and N2:O2 (stress phase) atmospheres.[12] Current-voltage (IV) sweeps, transient photocurrent (TPC) and transient photovoltage (TPV) measurements were continuously and sequentially acquired in-situ and in-operando. During the stress phase the PCE was exponentially lost over time due to the emergence of a space-charge within the device that impeded charge extraction and accelerated photo-oxidation of the perovskite layer.[12] Here, we use the same setup to age MAPIC, CH3NH3PbI3 (MAPI) and FA0.83Cs0.17Pb(I0.8Br0.2)3 (mixed-cation) PSCs. The intrinsic stability of MAPI is poor due to the volatility of the methylammonium (MA) cation.[13] As MA sublimates, the perovskite converts into PbI2-rich domains that lower the efficiency of charge generation and impede charge transport between perovskite grains, thus affecting the open-circuit voltage (Voc) and the short-circuit current (Jsc).[9] To overcome these issues, more structurally stable perovskites have been obtained by replacing the MA cation with complex cation mixtures.[14-16] The caesium/formamidinium (Cs/FA) combination has been used to fabricate structurally stable and band-gap tuneable FA0.83Cs0.17Pb(IxBr1-x)3 regular PSCs with relatively high PCEs .[11, 17, 18] Here we use the mixed-cation devices to provide a point of comparison between PSCs with active layers of differing intrinsic stability. In Figure 1 we show the evolution of the normalized figures-of-merit (FOM) extracted from reverse and forward IV sweeps (Figure S1-Figure S3) of the three inverted devices stressed under continuous simulated solar illumination (AM 1.5 G) in dry N2 and in dry N2 (99%): O2 (1%) atmospheres. All devices discussed here have the architecture FTO/PEDOT:PSS/Poly-TPD/Perovskite/PCBM/BCP/Au (see SI for Materials and Methods). Such a device structure results in a negligible hysteresis (Figure S4-Figure S6) compared to analogous regular n-i-p devices,[12] due to the good charge extraction properties of PCBM, and presumably fewer defects responsible for charge recombination at the perovskite charge extraction layer interface.[19] 2 Figure 1 Evolution of the figures-of-merit (FOM) of inverted MAPIC, MAPI, and mixed-cation PSCs under continuous illumination and dry N2 (Time < 0) and dry N2 (99%) with O2 (1%) (Time > 0). Solid lines correspond to the FOM extracted from the IV reverse sweeps (from 1 V to 0 V). Pale lines represent forward IV sweeps (from 0 V to 1 V). Metrics during stabilization (Time < 0) and stress (Time > 0) phases are normalized to the first recorded value of the reverse metrics during stabilization and stress phases, respectively. Normalized open-circuit voltage Voc (a), short-circuit current Jsc (b), fill factor FF (c) and power conversion efficiency PCE (d). During the stabilization phase (Time < 0) all devices undergo a reduction in PCE, with the MAPI PSC experiencing total failure within 20 hours. The loss in the PCE of the MAPIC and mixed-cation PSCs is mainly due to a reduction in Jsc, however for the MAPI device, the Voc also reduces. The superior stability of MAPIC over MAPI during the stabilization phase could be an effect of PbCl2 in the precursor solution resulting in a perovskite layer with improved morphology and/or lower defect density[20, 21], although the exact mechanism(s) for stability enhancement are not fully understood. During the stress phase (Time > 0) the MAPIC PSC completely degrades to ~5% of its initial PCE over 20 hours whilst the mixed-cation device retains ~70% of its initial PCE. The Voc of the mixed-cation PSC remains constant throughout both phases, suggesting that the perovskite remains stable and is not apparently affected by halide segregation. The MAPIC device turned into yellow colour at the end of the stress phase, consistent with the known mechanism for generation and reaction of superoxide (O2 crystal.[22] In general, for all the devices, most of the losses in the PCE are due to losses in the Jsc. Therefore, we postulate that photo-oxidization, or degradation of the charge extraction layers, or their interface with the perovskite could be playing a role with reducing the charge extraction efficiency. -), which subsequently decomposes the methylammonium halide within the perovskite To explore in detail the photocurrent loss mechanisms we consider the evolution in TPC and TPV traces measured in sequence with the IV scans during the stabilization and stress phases (see Figure S7-Figure S9 and Figure S13-Figure S18 for TPC and TPV data, respectively). From the TPC traces we identify five types of 3 photocurrent behaviour, which are represented in Figure 2a. Slow components (dominant in type 3 and 4) are typically attributed to charge trapping/de-trapping and recombination processes, while fast transients (dominant in type 1, 2 and 5) are compatible with timescales associated with charge carrier transport.[23-25] The TPC dataset was clustered with a Pattern Recognition Neural Network (PRNN), which is a software-based computing system that works similarly to biological nervous systems. Once trained to recognize certain patterns, PRNNs can output fuzzy or intermediate answers. Here a PRNN (Figure S10) is trained with the TPC dataset shown in Figure 2a to provide a qualitative description of the TPC shape evolution during ageing. In Figure 2b we plot the extracted charge from the photocurrent decay transients and indicate the TPC curve types evolution during ageing. At the beginning of the stabilization phase all devices behave according to type 1 with a fast transient when the LED is switched on/off, which is indicative of the relatively clean and efficient photocurrent generation behaviour of the as-fabricated PSCs.[12] Continuous operation in N2 induces changes in the TPC shape for all devices. The TPC of the MAPIC device transitions from type 1 to type 2 after ~6 hours of ageing. The photocurrent overshoot in type 2, observed in the first few μs of PSC illumination, may be attributed to the rapid formation of a transient diffusion gradient that enhances charge carrier recombination (reduces the photocurrent) before fading.[24, 26] The TPC of the MAPI PSC transitions from type 1 to type 4 after only ~3 hours of ageing, during which time the extracted charge from the photocurrent decays progressively reduces until the solar cell stops working. The TPC characteristics of the mixed-cation PSC immediately transitions from type 1 to type 5 after ~1 hour of ageing and maintains this behaviour until the last ~5 hours of the stress phase, when it goes back to type 1. Throughout ageing of the mixed-cation PSCs, the extracted charge experiences a negligible drop. In the MAPIC PSC, after ~1.5 hour exposure to oxygen the TPC transitions from type 2 to type 5 with a continuous decrease in the extracted charge. As the TPC traces further evolve from type 5 to type 3 the decay signal becomes negative, which is indicative of charge injection into the cell.[27] This observation and the photocurrent decay during the LED 'on' period could be explained by enhanced trap-assisted recombination and reduced charge de-trapping rate mechanisms.[28] While the charge density within the device increases due to continued photoexcitation, the competition between charge recombination and charge extraction in the PSC favours the former process to an extent that the steady-state photocurrent decreases. An increase in charge density within the PSC may also result in a space-charge that opposes the built-in field, resulting in a lower charge extraction efficiency.[12] For the mixed-cation PSC, the fact that the TPC shape does not seem to be influenced by the presence of oxygen indicates the superior stability of this device. We also observe that TPC type 4, seen during periods of severe photo-degradation, and type 3, which is dominant during the stress period, are both characterized by slow photocurrent decay transients (prolonged charge de-trapping and injection), compared to the other curves. TPV measurements provide complementary information on the generation/recombination kinetics of photo-generated charges in the small perturbation regime.[27] In our degraded solar cells the TPV decays are best 4 fitted with a double exponential function (see Figure S19-Figure S24).[20, 29] In Figure 2c we show that for the MAPIC device the fast time constant (T2) dominates during stabilization (a2 > a1). During the stress phase however the slow time constant (T1) increases and becomes dominant (a1 > a2) within ~5 hours before stabilizing. Although the origin of the slow and fast components are still under debate,[12, 29] we note that the double exponential behaviour is indicative of two populations of carriers that independently recombine.[20] In Figure 2d we show the evolution of the slow time constant T1 versus Voc during the stress phase. This trend is compared to the ideal behaviour of the same device prior to ageing obtained by measuring TPVs under different light intensities. The T1 vs Voc trend during the stress phase is non-linear with remarkably higher time constants compared to the ideal behaviour, suggesting that the time constants measured during the stress phase are likely to originate from trapped charges within the perovskite layer rather than free carriers. However, for the mixed- cation the dominant time constant is significantly lower (~1-2μs) and remains stable throughout ageing (Figure S22-Figure S24). This indicates that traps are not being generated in the perovskite layer and that the observed degradation might be due to degrading interlayers reducing current extraction and increasing the series resistance. Figure 2 Stability kinetics extracted from transient photocurrent (TPC) and transient photovoltage (TPV) measurements of MAPIC, MAPI, and mixed-cation PSCs under continuous illumination and dry N2 (Time < 0) and dry N2 (99%) with O2 (1%) (Time > 0). a, Offset plot of types of behaviour for normalized TPC curves. b, Extracted charge obtained by integrating the TPC curves after the LED is switched off. The indicated TPC types are limited by arrows. c, Extracted time and amplitude constants from the double exponential fits of the TPV transient decays (a1×exp(-x/T1)+ a2×exp(-x/T2), where x is the time (μs)) for the MAPIC device. d Slow photovoltage decay time constant (T1) versus Voc for the MAPIC device during the stress phase compared to reference values (black curve). 5 To further understand the recombination dynamics of the solar cells under stress conditions, we measured IV sweeps under variable light intensities (1-100 mW/cm2 AM1.5G) before stabilization, at the end of stabilization, and at the end of the stress phase. The Voc versus the natural logarithm of the light intensity shows a linear behaviour (Figure S25) and from its slope nkT/q we can extract the ideality factor n (Figure 3).[30, 31] For the mixed-cation PSC n ≈ 1 throughout ageing indicates bimolecular charge recombination.[32] For the MAPIC PSC the progressive increase of n from ~1.66 to 2.53 during the stabilization phase indicates an increase in Shockley- Reed-Hall trap-based recombination. Further, we examine the power law dependence of Jsc with light intensity (Jsc ∝ Iα) (Figure S26). The fitted alpha parameter (Figure 3) reduces throughout ageing for both MAPIC and the mixed-cation PSCs indicating the possible presence of trapped charges within the perovskite layer.[33] Figure 3 Ideality (n) and alpha factor extracted from Voc and Jsc vs Light Intensity, respectively. Measurements are performed before the beginning of the stabilization phase, at the end of the stabilization phase, and at the end of the stress phase. Full lines represent the extracted ideality factor, as indicated by the black arrow. Dotted lines represent the extracted alpha factor, as indicated by the black dotted arrow. Vertical lines indicate the 95% confidence intervals for n and alpha extracted from the fits. In summary, we investigated the operational stability kinetics of unencapsulated CH3NH3PbI3, CH3NH3PbI3-xClx and FA0.83Cs0.17Pb(I0.8Br0.2)3 inverted perovskite solar cells in the presence of light and dry oxygen using in-situ and in-operando IV, TPC and TPV measurements. We confirm the superior stability of the mixed-cation PSCs compared to the benchmark PSCs. The observed light- and oxygen-induced degradation in the MAI-based solar cells occurs over shorter timescales than the mixed-cation devices, and is dominated by a loss in photocurrent and charge extraction efficiency. We interpret this to the generation of electron traps, resulting in long-lived trapped charge and the build-up of space-charge within the perovskite absorber layer. Our findings provide important insights towards understanding the operation of perovskite solar cells, and suggest that focussing on mitigating trap generation during ageing will lead to further improvements in perovskite solar cell operation. 6 Acknowledgements M.A. acknowledges the Ministry of Presidential Affairs (UAE) for supporting her doctoral studies. A.J.P. acknowledges support from the EPSRC through the grant EP/M024873/1. J.T.-W.W. acknowledges the EPSRC funding. We thank Chris Amey and Ravichandran Shivanna for preliminary measurements. Supporting Information Supporting information is available at the following link: https://www.dropbox.com/s/g3dvufiwzqyhskb/StressPerovskite_v4.1_SI.pdf?dl=0 . References [1] [2] [3] [4] NREL, 2017. H. J. Snaith, J. Phys. Chem. Lett. 2013, 4. J. H. Noh, S. H. Im, J. H. Heo, T. N. Mandal, S. I. Seok, Nano letters 2013, 13. S. D. Stranks, G. E. Eperon, G. Grancini, C. Menelaou, M. J. Alcocer, T. Leijtens, L. M. Herz, A. Petrozza, H. J. Snaith, Science 2013, 342. H. Oga, A. Saeki, Y. Ogomi, S. Hayase, S. Seki, J. Am. Chem. Soc. 2014, 136. L. M. Pazos-Outón, M. Szumilo, R. Lamboll, J. M. Richter, M. Crespo-Quesada, M. Abdi-Jalebi, H. J. Beeson, M. Vrućinić, M. Alsari, H. J. Snaith, Science 2016, 351. M. Alsari, O. Bikondoa, J. Bishop, M. Abdi-Jalebi, L. Y. Ozer, M. Hampton, P. Thompson, M. T. Horantner, S. Mahesh, C. Greenland, J. E. Macdonald, G. Palmisano, H. J. Snaith, D. G. Lidzey, S. D. Stranks, R. H. Friend, S. Lilliu, Energy Environ. Sci. 2018. T. Leijtens, G. E. Eperon, N. K. Noel, S. N. Habisreutinger, A. Petrozza, H. J. Snaith, Adv. Energy Mater. 2015, 5. T. Leijtens, K. Bush, R. Cheacharoen, R. Beal, A. Bowring, M. D. McGehee, J. Mater. Chem. A 2017. J. T.-W. Wang, Z. Wang, S. Pathak, W. Zhang, D. W. deQuilettes, F. Wisnivesky-Rocca-Rivarola, J. Huang, P. K. Nayak, J. B. Patel, H. A. Mohd Yusof, Y. Vaynzof, R. Zhu, I. Ramirez, J. Zhang, C. Ducati, C. Grovenor, M. B. Johnston, D. S. Ginger, R. J. Nicholas, H. J. Snaith, Energy Environ. Sci. 2016, 9. D. P. McMeekin, G. Sadoughi, W. Rehman, G. E. Eperon, M. Saliba, M. T. Hörantner, A. Haghighirad, N. Sakai, L. Korte, B. Rech, Science 2016, 351. A. J. Pearson, G. E. Eperon, P. E. Hopkinson, S. N. Habisreutinger, J. T.-W. Wang, H. J. Snaith, N. C. Greenham, Adv. Energy Mater. 2016, 6. B. Conings, J. Drijkoningen, N. Gauquelin, A. Babayigit, J. D'Haen, L. D'Olieslaeger, A. Ethirajan, J. Verbeeck, J. Manca, E. Mosconi, Adv. Energy Mater. 2015, 5. [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] M. Saliba, T. Matsui, K. Domanski, J.-Y. Seo, A. Ummadisingu, S. M. Zakeeruddin, J.-P. Correa-Baena, [15] W. R. Tress, A. Abate, A. Hagfeldt, Science 2016, 354. H. Tan, A. Jain, O. Voznyy, X. Lan, F. P. G. de Arquer, J. Z. Fan, R. Quintero-Bermudez, M. Yuan, B. Zhang, Y. Zhao, Science 2017, 355. [16] M. Saliba, T. Matsui, J.-Y. Seo, K. Domanski, J.-P. Correa-Baena, M. K. Nazeeruddin, S. M. Zakeeruddin, W. Tress, A. Abate, A. Hagfeldt, Energy Environ. Sci. 2016, 9. Z. Wang, D. P. McMeekin, N. Sakai, S. van Reenen, K. Wojciechowski, J. B. Patel, M. B. Johnston, H. J. Snaith, Adv. Mater 2017, 29. Z. Wang, Q. Lin, F. P. Chmiel, N. Sakai, L. M. Herz, H. J. Snaith, Nat. Energy 2017, 2. S. van Reenen, M. Kemerink, H. J. Snaith, J. Phys. Chem. Lett. 2015, 6. V. Roiati, S. Colella, G. Lerario, L. De Marco, A. Rizzo, A. Listorti, G. Gigli, Energy Environ. Sci. 2014, 7. F. K. Aldibaja, L. Badia, E. Mas-Marza, R. S. Sanchez, E. M. Barea, I. Mora-Sero, J. Mater. Chem. A 2015, 3. 7 [17] [18] [19] [20] [21] [22] [23] N. Aristidou, I. Sanchez-Molina, T. Chotchuangchutchaval, M. Brown, L. Martinez, T. Rath, S. A. Haque, Angewandte Chemie International Edition 2015, 54. A. J. Pearson, P. E. Hopkinson, E. Couderc, K. Domanski, M. Abdi-Jalebi, N. C. Greenham, Organic Electronics 2016, 30. I. Hwang, C. R. McNeill, N. C. Greenham, J Appl Phys 2009, 106. Z. Li, F. Gao, N. C. Greenham, C. R. McNeill, Adv Funct Mater 2011, 21. [24] [25] [26] W. Tress, N. Marinova, T. Moehl, S. M. Zakeeruddin, M. K. Nazeeruddin, M. Gratzel, Energy Environ. Sci. 2015, 8. P. Calado, A. M. Telford, D. Bryant, X. Li, J. Nelson, B. C. O'Regan, P. R. F. Barnes, 2016, 7. Z. Li, W. Wang, N. C. Greenham, C. R. McNeill, Phys. Chem. Chem. Phys. 2014, 16. B. C. O'Regan, P. R. F. Barnes, X. Li, C. Law, E. Palomares, J. M. Marin-Beloqui, J. Am. Chem. Soc. 2015, 137. S. Shao, M. Abdu-Aguye, L. Qiu, L.-H. Lai, J. Liu, S. Adjokatse, F. Jahani, M. E. Kamminga, H. Gert, T. T. Palstra, Energy Environ. Sci. 2016, 9. J. Liu, G. Wang, Z. Song, X. He, K. Luo, Q. Ye, C. Liao, J. Mei, J. Mater. Chem. A 2017, 5. G. J. A. Wetzelaer, M. Scheepers, A. M. Sempere, C. Momblona, J. Ávila, H. J. Bolink, Adv. Mater 2015, 27. D. Zhao, M. Sexton, H.-Y. Park, G. Baure, J. C. Nino, F. So, Adv. Energy Mater. 2015, 5. [27] [28] [29] [30] [31] [32] [33] 8
1907.00767	1	1907	2019-06-18T05:37:53	Chipscale plasmonic modulators and switches based on metal-insulator-metal waveguides with Ge2Sb2Te5	[ "physics.app-ph", "physics.optics" ]	We introduce phase-change material Ge2Sb2Te5 (GST) into metal-insulator-metal (MIM) waveguide systems to realize chipscale plasmonic modulators and switches in the telecommunication band. Benefitting from the high contrast of optical properties between amorphous and crystalline GST, the three proposed structures can act as reconfigurable and non-volatile modulators and switches with excellent modulation depth 14 dB and fast response time in nanosecond, meanwhile possessing small footprints, simple frameworks and easy fabrication. This work provides new solutions to design active devices in MIM waveguide systems, and can find potential applications in more compact all-optical circuits for information processing and storage.	physics.app-ph	physics	Chipscale plasmonic modulators and switches based on metal-insulator-metal waveguides with Ge2Sb2Te5 Zhaojian Zhang 1, Junbo Yang 2, , Wei Bai 3, Yunxin Han 2, Xin He 2, Jingjing Zhang 1, Jie Huang 1, Dingbo Chen 1, Siyu Xu 1 and Wanlin Xie 1 1 College of Liberal Arts and Sciences, National University of Defense Technology, Changsha 410073, China 2 Center of Material Science, National University of Defense Technology, Changsha 410073, China 3 Institute of Optics and Electronics Chinese Academy of Sciences, Chengdu 610209, China Correspondence: [email protected] Abstract: We introduce phase-change material Ge2Sb2Te5 (GST) into metal-insulator-metal (MIM) waveguide systems to realize chipscale plasmonic modulators and switches in the telecommunication band. Benefitting from the high contrast of optical properties between amorphous and crystalline GST, the three proposed structures can act as reconfigurable and non-volatile modulators and switches with excellent modulation depth 14 dB and fast response time in nanosecond, meanwhile possessing small footprints, simple frameworks and easy fabrication. This work provides new solutions to design active devices in MIM waveguide systems, and can find potential applications in more compact all-optical circuits for information processing and storage. Keywords: metal-insulator-metal, phase-change material, plasmonic modulator, plasmon- induced transpareny 1. Introduction Facing the increasing demand for information quantity and transmission rate, today's electronic integrated devices will be unable to afford roles in future chipscale systems due to the fundamental limits [1]. As a promising solution, photonic integrated devices can break the blocks by using light as the carrier of information. To construct basic photonic circuits, dielectric waveguides (such as silicon waveguide) have been widely studied for decades [2,3]. However, the scale of a dielectric waveguide cannot go smaller than the working wavelength due to the diffraction limit, which restricts the integration and compactness of photonic integrated chips [1]. Recently, surface plasmon polariton (SPPs), which are the surface electromagnetic waves supported at the interface between metal and dielectric, have attracted more attention due to the capacity to break the diffraction limit [1]. SPPs also introduce plasmonic waveguides, which can deliver light in sub-wavelength scale [4,5]. Especially, plasmonic metal-insulator-metal (MIM) waveguides possess good confinement of light, low bending loss, easy fabrication and acceptable propagation length, therefore can be potential platforms for future more compact all-optical integrated circuits [6-9]. As the promising next-generation on-chip photonic system, MIM system has supported various chipscale passive devices such as filters [10,11], demultiplexers [12,13], sensors [14,15] and spacers [16,17]. However, active devices are also required in such system. Therefore, optical materials with tunable properties have been utilized to realize dynamically tunable functionalities. For example, nonlinear materials have been used to realize all-optical tunable filters, logic gates and diodes [18-20]. Lasers, switches and slow light enhancement can be achieved by integrating gain materials [21-23]. Two-dimension (2D) materials have contributed to the active manipulation of MIM waveguides as well [24-26]. Some novel optical concepts, such as epsilon-near-zero (ENZ) effects and non-PT-symmetry, have also been made possible in MIM systems assisted by optical materials [27,28]. Obviously, optical materials will play an important role in active systems, and it is essential to explore more outcomes from the combination between MIM waveguides and novel optical materials. Recently, phase-change materials (PCMs), such as Ge2Sb2Te5 (GST), have drawn more attention. GST possesses tremendously different optical properties from amorphous to crystalline states. Such phase transition can be triggered by thermal, electrical or optical schemes, has an ultrafast switching time in nanosecond (ns) or even sub-ns, and is usually reversible [29]. Up to now, GST has been regarded as a strong candidate for realizing reconfigurable and non-volatile all-optical devices for data processing and storage [30], and many applications have been proposed such as on-chip memory elements [31-33], all-optical switching [34-36] and reconfigurable metasurfaces [37-39]. Here, we introduce GST into MIM systems to realize chipscale plasmonic modulators and switches in shortwave infrared (SWIR) telecommunication regime. Combining metal-silicon- metal (MSM) waveguides and metal-GST-metal (MGM) resonators, we numerically propose three typical MIM structures, including end-coupled rectangular resonator, side-coupled stub resonator and two mutually coupled resonators, to investigate their performances on transmission modulations. Due to the phase transition of GST, the effective refractive index inside MGM resonators can be actively tuned with high contrast, consequently leading to the significant shifts of resonant wavelengths. Benefitting from this, the three proposed structures can act as reconfigurable and non-volatile modulators and switches with excellent modulation depth and fast response time, meanwhile possessing small footprints, simple frameworks and easy fabrication. This work provides new solutions to design active devices based on MIM systems, and can find potential applications in future on-chip all-optical circuits. 2. Structures, materials and methods The first proposed structure is depicted in Fig. 1, consisting of input and output MSM waveguides end-coupled with an MGM rectangular resonator. The geometric parameters are as follows: the width of waveguide w= 50 nm, the gap g= 20 nm, and the length of resonator L= 120 nm. Fig. 1 The schematic of the first structure. In Fig. 1, the grey area is silver (Ag), the dielectric constant of which is defined by the Drude model [10]: e m ¥= e 2 p w + ww g ( i ) (1) p m e is the dielectric constant of Ag, e¥ is the dielectric constant at infinite frequency, Where w is the bulk plasma frequency, g is the electron oscillation damping frequency, and w is the angular frequencies of incident waves. The parameters of Ag are as follows: e¥ =3.7, = 1.38×1016 Hz, and g = 2.73×1013 Hz. The waveguides are filled with silicon (Si) due to the low dispersion and loss in SWIR which is shown in red in Fig. 1, and the corresponding optical w p constant (n+ik) are presented in Fig. 2 [40]. GST possesses dramatically different optical properties between the amorphous (aGST) and crystalline state (cGST), the corresponding optical constants are also shown in Fig. 2, respectively [41]. We can see that there is a high contrast in refractive index between the two states, and cGST possesses much more intrinsic loss than aGST in SWIR. The effective dielectric constants of GST with different crystallinities are predicted by the Lorentz-Lorenz relation [42]: e e GST GST l ( , l ( , C ) 1 C ) 1 e = · C e cGST l l cGST ( ) 1 + ( ) 2 + - (1 e C e ) l l aGST aGST ( ) 1 + ( ) 2 (2) Here, e and aGST e cGST are dielectric constants of aGST and cGST, respectively. C is the crystallization fraction of GST from 0 to 1. Fig. 2 (a) The refractive index of amorphous GST, crystalline GST and silicon. (b) The extinction coefficient of amorphous GST, crystalline GST and silicon. In the MIM waveguide, only fundamental transverse-magnetic (TM) mode exists. The dispersion of fundamental mode is described as follows [6]: i e b e b m 2 e e 2 + tanh( b w e 2 2 k 2 0 m k 2 0 i k 2 0 i = ) 0 (3) i e is the dielectric constant of the insulator (Si or GST), Where represents the wavevector in free space, and b is the wavevector of SPPs inside the MIM waveguide. The effn of fundamental TM effective refractive index is defined as . The calculated kb= pl 2 effn k = 0 0 0 mode in the MSM waveguide as well as the MGM waveguide in different crystallization conditions are shown in Fig. 3(a) and (b). The profile of fundamental TM mode in the MSM - - - - · - - - - waveguide is presented in Fig. 3(c), and the mode in the MGM waveguide has the similar profile. 2D Finite-Difference Time-Domain (FDTD) method is applied to do the simulation, the mesh size is set as 3 nm to ensure the accuracy, and the boundary condition is set as perfectly matched layers (PMLs) to maintain convergence. Two power monitors are placed at input and output respectively, and the distance between the two monitors is d= 500 nm. The transmission is calculated following T= Pout/Pin. Fig. 3 (a) The real part of effective refractive index. (b) The imaginary part of effective refractive index. (c) The profile of fundamental TM mode in the MSM waveguide. 3. Results and discussions The transmission spectrum of the structure in Fig. 1 corresponding to different phase states of GST are shown in Fig. 4(a). In MIM system, such structure is a typical band-pass filter. The rectangular resonator acts as a Fabry -- Perot (F-P) resonator, only resonant wavelength can be coupled into this resonator from the input waveguide, and released to the output waveguide. The resonant condition is as follows [43]: j l = p ) = L m ) , n eff (4) Here, m is the mode number, which is an integer. j is the phase shift induced by the reflection at the insulator-metal interface. As shown in Fig. 4(a) in blue, the passed resonant wavelength 4 Re( 1, 2... (2 m is 1527 nm when GST is in amorphous state, the transmission of which is 44%. The corresponding magnitude of magnetic field H distribution is shown in Fig. 4(b), from which we can see that there exists a F-P mode with m= 1 in the rectangular resonator, and SPPs can pass through such structure. However, When GST comes to crystalline state, the transmission - peak will have a red shift due to the increase of Re( )effn as shown in Fig. 4(a). Meanwhile, the amplitude of the peak will decline caused by the rise of material loss. Benefitting from this, the transmission at 1527 nm can be actively tuned from 44% to 1.8%, the corresponding H distribution is shown in Fig. 4(c), showing that transmitted SPPs is suppressed. To evaluate the performance of transmission modulation, the modulation depth (MD) is defined as follows: = MD T -T min max T max 100% (5) Where maxT and minT are maximum and minimum transmission values, respectively. For convenience, we also provide modulation results in log scale: IL (Insertion loss ) = - 10lg(T) = - MD dB 10lg(T /T ) max min (6) The corresponding modulation results are shown in Table. 1. From the aGST to cGST, the MD can reach 96% (13.8 dB). Such modulation is much better than previous similar works based on MIM structures [44-48], meanwhile possesses comparable response time. Fig. 4 (a) The transmission spectrum corresponding to different phase states of GST. (b) The H distribution at 1527 nm in GST amorphous state. (c) The H distribution at 1527 nm in GST crystalline state. Table 1. The transmission as well as insertion loss in different GST states and the modulation depth. GST state T IL a 44% 3.6 dB 25% c 16% 8.0 dB 50% c 6.3% 75% c 3.1% c 1.8% MD 96% 12.0 dB 15.1 dB 17.4 dB 13.8 dB Next, we introduce the second structure, a main MSM waveguide side-coupled with an MGM stub resonator, as shown in Fig. 5. The geometric parameters are as follows: w= 50 nm and L= 200 nm. The other parameters are unchanged. · Fig. 5 The schematic of the second structure. Such structure is a typical band-stop filter in MIM system. The stub resonator can be seen as a semi F-P resonator, can trap the resonant wavelength and prevent it from passing, the resonant condition is as follows [49]: + - 1 (2 n j l = p ) 4 Re( n eff = L n ) , 0,1... (7) Where n is the mode number. The transmission spectrum corresponding to different phase states of GST are shown in Fig. 6(a). Due to the high contrast between the refractive index of aGST and cGST, a giant red shift of transmission dip happens. Consequently, the transmission at 1513 nm can be modulated from 2.1% (16.8 dB) to 57% (2.4 dB), leading to an excellent MD 96% (14.4 dB), the details are given in table. 2. The H distribution at 1513 nm corresponding to aGST and cGST are depicted in Fig. 6 (b) and (c), respectively, indicating n= 1. Fig. 6 (a) The transmission spectrum corresponding to different phase states of GST. (b) The H distribution at 1513 nm in GST amorphous state. (c) The H distribution at 1513 nm in GST crystalline state. Table 2. The transmission as well as insertion loss in different GST states and the modulation depth. GST state T IL a 2.1% 16.8 dB 25% c 19% 7.2 dB 50% c 41% 3.9 dB 75% c 52% 2.8 dB c 57% 2.4 dB MD 96% 14.4 dB The last structure is shown in Fig. 7, including a main MSM waveguide side-coupled with a stub resonator and a rectangular resonator. The geometric parameters are as follows: h= 45 nm, g= 15 nm, L= 110 nm and w= 50 nm. The other parameters are unchanged. Fig. 7 The schematic of the third structure. Such structure can produce plasmon-induced transparency (PIT) effect, which is a special case of Fano resonance. When this is only stub resonator, a transmission dip exists as shown in Fig. 8(a) in blue. However, when the rectangular resonator is coupled upside the stub resonator, a transmission peak will appear at the position of the original dip as given in Fig. 8(a) in green, which is called PIT. Such effect attributes to the near-field interaction between the mode in the stub (radiative mode) and the mode in the rectangle (subradiant mode). When the two modes have close resonant wavelengths, constructive interference will happen between the two modes, consequently leading to that transparent window [50]. Fig. 8 (a) The transmission spectrum corresponding to only stub as well as stub plus rectangle. (b) The transmission spectrum corresponding to aGST and cGST. (c) The H distribution at 1479 nm in aGST. (d) The H distribution at 1479 nm in cGST. Here, we utilize GST to tune the resonant wavelengths of the subradiant mode, and further control the interference to realize switching. When GST is amorphous, the resonant wavelengths of radiative and subradiant modes are close to each other, therefore leading to an induced transparent peak at 1479 nm as shown in Fig. 8 (b) in blue. However, when it comes to cGST, the resonant wavelength of subradiant mode is shifted away, leading to no interference i.e., a transmission dip at 1479 nm as shown in Fig. 8(b) in green. The corresponding H distributions are given in Fig. 8(c) and (d), respectively. The transmission at 1479 nm can be switched from 55% (2.6 dB) to 2% (17.0 dB), the MD can be up to 96% (14.4 dB), which is promising for all-optical switching. 4. Conclusion In summary, we investigate three structures to show that, GST can be efficiently utilized to construct the high-performance modulator and switch in the plasmonic MIM system, and it is the first time to introduce GST into MIM system so far as we know. Besides, this work focuses on telecommunication band, aiming to provide more solutions for on-chip photonic systems for information processing. Furthermore, the small footprint and easy fabrication of proposed structures can have advantages in applications on future more compact all-optical circuits. Acknowledgments This work is supported by the National Natural Science Foundation of China (61671455, 61805278), the Foundation of NUDT (ZK17-03-01), the Program for New Century Excellent Talents in University (NCET-12-0142), and the China Postdoctoral Science Foundation (2018M633704). Reference [1] Gramotnev, D. K., & Bozhevolnyi, S. I. (2010). Plasmonics beyond the diffraction limit. Nature photonics, 4(2), 83. [2] Marcatili, E. A. (1969). Dielectric rectangular waveguide and directional coupler for integrated optics. Bell System Technical Journal, 48(7), 2071-2102. [3] Deslandes, D., & Wu, K. (2001). Integrated microstrip and rectangular waveguide in planar form. IEEE Microwave and Wireless Components Letters, 11(2), 68-70. [4] Bozhevolnyi, S. I., Volkov, V. S., Devaux, E., Laluet, J. Y., & Ebbesen, T. W. (2006). Channel plasmon subwavelength waveguide components including interferometers and ring resonators. Nature, 440(7083), 508. [5] Fang, Y., & Sun, M. (2015). Nanoplasmonic waveguides: towards applications in integrated nanophotonic circuits. Light: Science & Applications, 4(6), e294. [6] Dionne, J. A., Sweatlock, L. A., Atwater, H. A., & Polman, A. (2006). Plasmon slot waveguides: Towards chip- scale propagation with subwavelength-scale localization. Physical Review B, 73(3), 035407. [7] Neutens, P., Van Dorpe, P., De Vlaminck, I., Lagae, L., & Borghs, G. (2009). Electrical detection of confined gap plasmons in metal -- insulator -- metal waveguides. Nature Photonics, 3(5), 283. [8] Lu, H., Wang, G., & Liu, X. (2013). Manipulation of light in MIM plasmonic waveguide systems. Chinese Science Bulletin, 58(30), 3607-3616. [9] Kriesch, A., Burgos, S. P., Ploss, D., Pfeifer, H., Atwater, H. A., & Peschel, U. (2013). Functional plasmonic nanocircuits with low insertion and propagation losses. Nano letters, 13(9), 4539-4545. [10] Lu, H., Liu, X., Mao, D., Wang, L., & Gong, Y. (2010). Tunable band-pass plasmonic waveguide filters with nanodisk resonators. Optics Express, 18(17), 17922-17927. [11] Zhan, G., Liang, R., Liang, H., Luo, J., & Zhao, R. (2014). Asymmetric band-pass plasmonic nanodisk filter with mode inhibition and spectrally splitting capabilities. Optics express, 22(8), 9912-9919. [12] Wang, G., Lu, H., Liu, X., Mao, D., & Duan, L. (2011). Tunable multi-channel wavelength demultiplexer based on MIM plasmonic nanodisk resonators at telecommunication regime. Optics Express, 19(4), 3513-3518. [13] Nakayama, K., Tonooka, Y., Ota, M., Ishii, Y., & Fukuda, M. (2018). Passive plasmonic demultiplexers using multimode interference. Journal of Lightwave Technology, 36(10), 1979-1984. [14] Binfeng, Y., Guohua, H., Ruohu, Z., & Yiping, C. (2014). Design of a compact and high sensitive refractive index sensor base on metal-insulator-metal plasmonic Bragg grating. Optics Express, 22(23), 28662-28670. [15] Chen, L., Liu, Y., Yu, Z., Wu, D., Ma, R., Zhang, Y., & Ye, H. (2016). Numerical analysis of a near-infrared plasmonic refractive index sensor with high figure of merit based on a fillet cavity. Optics express, 24(9), 9975- 9983. [16] Wang, G., Lu, H., & Liu, X. (2012). Trapping of surface plasmon waves in graded grating waveguide system. Applied Physics Letters, 101(1), 013111. [17] Li, D., Du, K., Liang, S., Zhang, W., & Mei, T. (2016). Wide band dispersion less slow light in hetero-MIM plasmonic waveguide. Optics express, 24(20), 22432-22437. [18] Yang, X., Hu, X., Chai, Z., Lu, C., Yang, H., & Gong, Q. (2014). Tunable ultracompact chip-integrated multichannel filter based on plasmon-induced transparencies. Applied Physics Letters, 104(22), 221114. [19] Yang, X., Hu, X., Yang, H., & Gong, Q. (2017). Ultracompact all-optical logic gates based on nonlinear plasmonic nanocavities. Nanophotonics, 6(1), 365. [20] Chai, Z., Hu, X., Yang, H., & Gong, Q. (2017). Chip-integrated all-optical diode based on nonlinear plasmonic nanocavities covered with multicomponent nanocomposite. Nanophotonics, 6(1), 329. [21] Hill, M. T., Marell, M., Leong, E. S., Smalbrugge, B., Zhu, Y., Sun, M., ... & Nötzel, R. (2009). Lasing in metal-insulator-metal sub-wavelength plasmonic waveguides. Optics express, 17(13), 11107-11112. [22] Yu, Z., Veronis, G., Fan, S., & Brongersma, M. L. (2008). Gain-induced switching in metal-dielectric-metal plasmonic waveguides. Applied Physics Letters, 92(4), 041117. [23] Zhang, Z., Yang, J., He, X., Han, Y., Zhang, J., Huang, J., ... & Xu, S. (2018). Active Enhancement of Slow Light Based on Plasmon-Induced Transparency with Gain Materials. Materials, 11(6), 941. [24] Lu, H., Gan, X., Mao, D., & Zhao, J. (2017). Graphene-supported manipulation of surface plasmon polaritons in metallic nanowaveguides. Photonics Research, 5(3), 162-167. [25] Chen, Z., Zhou, R., Wu, L., Yang, S., & Liu, D. (2018). Surface plasmon characteristics based on graphene- cavity-coupled waveguide system. Solid State Communications, 280, 50-55. [26] Song, C., Wang, J., Liu, D., Hu, Z. D., & Zhang, F. (2018). Wavelength-sensitive PIT-like double-layer graphene-based metal -- dielectric -- metal waveguide. Applied optics, 57(33), 9770-9776. [27] Li, Y., & Argyropoulos, C. (2018). Tunable nonlinear coherent perfect absorption with epsilon-near-zero plasmonic waveguides. Optics letters, 43(8), 1806-1809. [28] Huang, Y., Min, C., & Veronis, G. (2016). Broadband near total light absorption in non-PT-symmetric waveguide-cavity systems. Optics express, 24(19), 22219-22231. [29] Wuttig, M., Bhaskaran, H., & Taubner, T. (2017). Phase-change materials for non-volatile photonic applications. Nature Photonics, 11(8), 465. [30] Youngblood, N., Ríos, C., Gemo, E., Feldmann, J., Cheng, Z., Baldycheva, A., ... & Bhaskaran, H. (2019). Tunable Volatility of Ge2Sb2Te5 in Integrated Photonics. Advanced Functional Materials, 29(11), 1807571. [31] Lankhorst, M. H., Ketelaars, B. W., & Wolters, R. A. (2005). Low-cost and nanoscale non-volatile memory concept for future silicon chips. Nature materials, 4(4), 347. [32] Ríos, C., Stegmaier, M., Hosseini, P., Wang, D., Scherer, T., Wright, C. D., ... & Pernice, W. H. (2015). Integrated all-photonic non-volatile multi-level memory. Nature Photonics, 9(11), 725. [33] Feldmann, J., Youngblood, N., Wright, C. D., Bhaskaran, H., & Pernice, W. H. P. (2019). All-optical spiking neurosynaptic networks with self-learning capabilities. Nature, 569(7755), 208. [34] Rudé, M., Pello, J., Simpson, R. E., Osmond, J., Roelkens, G., van der Tol, J. J., & Pruneri, V. (2013). Optical switching at 1.55 μ m in silicon racetrack resonators using phase change materials. Applied Physics Letters, 103(14), 141119. [35] Zheng, J., Khanolkar, A., Xu, P., Colburn, S., Deshmukh, S., Myers, J., ... & Boechler, N. (2018). GST-on- silicon hybrid nanophotonic integrated circuits: a non-volatile quasi-continuously reprogrammable platform. Optical Materials Express, 8(6), 1551-1561. [36] Rios, C., Stegmaier, M., Cheng, Z., Youngblood, N., Wright, C. D., Pernice, W. H., & Bhaskaran, H. (2018). Controlled switching of phase-change materials by evanescent-field coupling in integrated photonics. Optical Materials Express, 8(9), 2455-2470. [37] Wang, Q., Rogers, E. T., Gholipour, B., Wang, C. M., Yuan, G., Teng, J., & Zheludev, N. I. (2016). Optically reconfigurable metasurfaces and photonic devices based on phase change materials. Nature Photonics, 10(1), 60. [38] Gholipour, B., Zhang, J., MacDonald, K. F., Hewak, D. W., & Zheludev, N. I. (2013). An all‐optical, non‐ volatile, bidirectional, phase‐change meta‐switch. Advanced materials, 25(22), 3050-3054. [39] Chen, Y. G., Kao, T. S., Ng, B., Li, X., Luo, X. G., Luk'Yanchuk, B., ... & Hong, M. H. (2013). Hybrid phase- change plasmonic crystals for active tuning of lattice resonances. Optics express, 21(11), 13691-13698. [40] Palik, E. D. (Ed.). (1998). Handbook of optical constants of solids (Vol. 3). Academic press. [41] Chu, C. H., Tseng, M. L., Chen, J., Wu, P. C., Chen, Y. H., Wang, H. C., ... & Tsai, D. P. (2016). Active dielectric metasurface based on phase‐change medium. Laser & Photonics Reviews, 10(6), 986-994. [42] Tian, J., Luo, H., Yang, Y., Ding, F., Qu, Y., Zhao, D., ... & Bozhevolnyi, S. I. (2019). Active control of anapole states by structuring the phase-change alloy Ge 2 Sb 2 Te 5. Nature communications, 10(1), 396. [43] Bavil, M. A., Gao, L., & Sun, X. (2013). A compact nanoplasmonics filter and intersection structure based on utilizing a slot cavity and a Fabry -- Perot resonator. Plasmonics, 8(2), 631-636. [44] Cai, W., White, J. S., & Brongersma, M. L. (2009). Compact, high-speed and power-efficient electrooptic plasmonic modulators. Nano letters, 9(12), 4403-4411. [45] Min, C., & Veronis, G. (2009). Absorption switches in metal-dielectric-metal plasmonic waveguides. Optics Express, 17(13), 10757-10766. [46] Zhong, Z. J., Xu, Y., Lan, S., Dai, Q. F., & Wu, L. J. (2010). Sharp and asymmetric transmission response in metal-dielectric-metal plasmonic waveguides containing Kerr nonlinear media. Optics express, 18(1), 79-86. [47] Piao, X., Yu, S., & Park, N. (2012). Control of Fano asymmetry in plasmon induced transparency and its application to plasmonic waveguide modulator. Optics express, 20(17), 18994-18999. [48] Haddadpour, A., Nezhad, V. F., Yu, Z., & Veronis, G. (2016). Highly compact magneto-optical switches for metal-dielectric-metal plasmonic waveguides. Optics letters, 41(18), 4340-4343. [49] Lin, X. S., & Huang, X. G. (2008). Tooth-shaped plasmonic waveguide filters with nanometeric sizes. Optics letters, 33(23), 2874-2876. [50] Wang, T., Zhang, Y., Hong, Z., & Han, Z. (2014). Analogue of electromagnetically induced transparency in integrated plasmonics with radiative and subradiant resonators. Optics express, 22(18), 21529-21534.
1712.10245	1	1712	2017-12-29T14:53:02	Advanced data mining in field ion microscopy	[ "physics.app-ph" ]	Field ion microscopy (FIM) allows to image individual surface atoms by exploiting the effect of an intense electric field. Widespread use of atomic resolution imaging by FIM has been hampered by a lack of efficient image processing/data extraction tools. Recent advances in imaging and data mining techniques have renewed the interest in using FIM in conjunction with automated detection of atoms and lattice defects for materials characterization. After a brief overview of existing routines, we review the use of machine learning (ML) approaches for data extraction with the aim to catalyze new data-driven insights into high electrical field physics. Apart from exploring various supervised and unsupervised ML algorithms in this context, we also employ advanced image processing routines for data extraction from large sets of FIM images. The outcomes and limitations of such routines are discussed, and we conclude with the possible application of energy minimization schemes to the extracted point clouds as a way of improving the spatial resolution of FIM.	physics.app-ph	physics	Advanced data mining in field ion microscopy Shyam Katnagallua, Baptiste Gaulta,∗, Blazej Grabowskib, Jorg Neugebauerb, Dierk Raabea, Ali Nematollahib,∗ aDepartment of Microstructure Physics and Alloy Design, Max Planck Insitut fur bDepartment of Computational Material Science, Max Planck Insitut fur Eisenforschung. Eisenforschung Abstract Field ion microscopy (FIM) allows to image individual surface atoms by ex- ploiting the effect of an intense electric field. Widespread use of atomic resolution imaging by FIM has been hampered by a lack of efficient image processing/data extraction tools. Recent advances in imaging and data min- ing techniques have renewed the interest in using FIM in conjunction with automated detection of atoms and lattice defects for materials characteriza- tion. After a brief overview of existing routines, we review the use of machine learning (ML) approaches for data extraction with the aim to catalyze new data-driven insights into high electrical field physics. Apart from exploring various supervised and unsupervised ML algorithms in this context, we also employ advanced image processing routines for data extraction from large sets of FIM images. The outcomes and limitations of such routines are dis- cussed, and we conclude with the possible application of energy minimization schemes to the extracted point clouds as a way of improving the spatial res- olution of FIM. Keywords: FIM, Image Processing, Machine learning, Atomic resolution, Data Mining. ∗Corresponding Author Email addresses: [email protected] (Shyam Katnagallu), [email protected] (Baptiste Gault), [email protected] (Blazej Grabowski), [email protected] (Jorg Neugebauer), [email protected] (Dierk Raabe), [email protected] (Ali Nematollahi) Preprint submitted to Materials Characterization January 1, 2018 1. Introduction Field ion microscopy (FIM), invented in 1951 by Erwin Muller [1, 2, 3], is a high electric field technique which uniquely enables imaging of surfaces with atomic resolution. FIM is based on ionization of an imaging gas in the vicinity of a field-emitter tip as a consequence of the locally high electric field. The high electric field is achieved by applying a high voltage of a few kilovolts onto a very sharp needle-shaped specimen maintained at a temperature usually below 80 K. Specimens are either electropolished [4] or milled with a focused ion beam (FIB) [5] into a very sharp needle tip with an end radius below 100 nm. The advantage of using FIB for specimen preparation lies in its site specific application for extracting tips in microstructure regions of high interest such as across internal interfaces. An excellent review on using FIB for site specific specimen preparation can be found in reference [5]. Once the specimen is mounted an imaging gas is introduced. The in- troduced imaging gas gets attracted by the cold surface due to polarization forces. The gas atoms then thermally accommodate with the cold tip surface by performing a series of "hops". Surrounding the tip surface there exists a critical surface, where the maximum ionization occurs. This surface usu- ally lies around 1-4 A above the tip [6]. During the "hops", the ionization probability for the gas atoms can be considerable as they spend a signifi- cant amount of time in the critical surface. As a consequence an electron can tunnel from the imaging gas atom into the tip. The ionized gas atom is accelerated away from the positively biased tip and towards the detector, where gas ions contribute to image formation. The surface is the intersection of the crystalline lattice with the imposed end shape, often approximated as nearly-spherical. Owing to the discreteness of the atomic arrangement at these scales the tip apex curvature is, in reality, made of atomic scale crystallographic terrace features where some of the top, edge and corner atoms are naturally protruding, producing local field en- hancement. This means that these exposed atomic terrace positions are sites of magnified electrostatic field strength and also of aberrated field direction. The amount of gas atoms ionizing depends on such a local enhancement of the electric field. This variation in electric field strength across the surface atoms gives the final contrast in the FIM image. Overall the contrast in FIM also depends on the gas supply function and adsorption behavior [7, 8, 9]. Atomic resolution can be attained in some cases on certain high index facets where the surface field distribution is corrugated enough to give contrast in 2 the image. By collecting the gas ions on a phosphor screen, an image is formed that reveals the distribution of the electrostatic field near the surface and the current created by the number of incoming imaging gas ions. Historically, images were collected on a film in a dark room after sufficient exposure on the screen was achieved by field ionization [2]. Eventually, field ion microscopes were fitted with a stack of micro-channel plates (MCP) in- front of the phosphor screen. The image on the phosphor screen can be recorded with a high-resolution high-frame-rate camera. The MCPs act as a photo multiplier by creating an electron cascade in response to the gas ion impact. Another variant of FIM is referred to as eFIMTM, which is performed on a local electrode atom probe (LEAP) [10] using the delay-line detector used in the atom probe mode. The atomic resolution of FIM made it a popular technique for studying internal interfaces [11, 12, 13] and dislocations [14, 15, 16] at unprecedented atomic positioning resolution. When exposing the tip not only to the mini- mum field strength required for ionizing the imaging gas but also for evap- orating the tip atoms themselves continually, the method is rendered depth sensitive. This means that the specimen can be investigated tomographically along the tip longitudinal axis, which has led to the development of 3DFIM [17]. The emergence of atom probe tomography (APT), which is addition- ally able to characterize the chemical identity of the imaged atoms, lead to some decline in the usage of the FIM technique by the materials science community. Nevertheless 3DFIM offers the important advantage over APT of a significantly higher spatial resolution with 100 % positional detection efficiency in 3 dimensions in some cases. This high degree in positional accu- racy allows characterization even down to single point defects in a material, a feature not offered by any other device. This causes currently an increased interest in the 3DFIM technique. Ultimately, owing to all the advances in detector technology, 3DFIM is capable of producing large and accurate tomographic datasets containing in- formation on sequential atomic positions. These large datasets lead to a new tremendous challenge of how to manage the data. Presently, there is a lack of efficient data handling and data treatment algorithms to extract perti- nent information from these datasets in an (a) automated; (b) fast; (c) user- independent; (d) and error quantified manner. For instance, characterization of a volume of 0.001 µm3 (a typical sample size analyzed in 3DFIM) produces in the range of 2 × 1015 images (assuming a constant field evaporation rate and capture speed). Hence, there is a great need for efficient algorithms and 3 data mining routines to fully exploit the potential of 3DFIM. To this end, M. Dagan et al. [18] have proposed an atom by atom data extraction routine for reconstructing 3DFIM data. Building on their work we recently proposed a new method to extract atomic positions from 3DFIM datasets [19]. With this article, we focus our attention on using various modern image processing and machine learning algorithms for extracting information from 3DFIM. The developments in Artificial Intelligence (AI), especially in computer vision have been explosive. Modern machine learning algorithms enable a fully automated detection and classification of objects in a picture. We give an overview about these advanced data mining tools and how they can be utilized to extract the wealth of information from 3DFIM images. We have implemented some of these concepts within a set of routines in PythonTM (Python Software Foundation; Python Language Reference, ver- sion 2.7; Available at http://www.python.org) employing the SciPy package [20]. These routines allow us to extract the relevant information from a large number (order of several 10,000s) of FIM images in very short computational times (order of minutes). 2. Existing data extraction routines for 3D field ion microscopy A 3DFIM experiment produces a series of images of the continually field evaporating surface. A main challenge lies not in acquiring such large datasets but rather in analyzing them. The article by Vurpillot et al. [21] serves as an excellent review for the current state-of-the-art and the main issues associated with data extraction from large FIM datasets. We briefly review here the available analysis methods and also some recent develop- ments. Broadly speaking the analysis methods can be categorized into an atom by atom approach and a geometrical approach. The first approach towards advanced 3DFIM analysis was already de- veloped in the early 70s for characterizing radiation damage in metals at the atomic scale [22, 23]. In these early approaches FIM images were cap- tured on film which later were developed and manually analyzed individually. The captured FIM images were dissected atom-by-atom and the positions of atoms and defects were marked manually. Owing to the associated cumber- some analysis methods, systematic FIM studies of more complex atomic sce- narios remained an exception. Taking the additional disadvantage of 3DFIM of being insensitive to the chemical nature, APT became gradually the more dominant technique. Yet, FIM's ability to characterize atomistic defects 4 such as vacancies in three dimensions is still unparalleled with any other technique. In this context the drastic increase in computing power became an essential asset when Dagan et al. developed an automated method to reconstruct 3DFIM data atom by atom [21, 24]. The algorithm takes advan- tage of layer by layer evaporation and the atoms are identified based on a threshold intensity. The final coordinates are converted to real space based on theoretical nearest neighbor distances. This work lead to a rise in interest around the physics of image formation in FIM and also in the use of the associated computationally enhanced analysis techniques. The geometrical approach to 3DFIM atomic position reconstruction in- troduced by Vurpillot et al. consists in stacking the digital images obtained from a 3DFIM experiment [17]. The image stack is then corrected, assuming a known projection law, a specimen's geometry and a constant evaporation rate. The stacking approach does not provide atomistic precision but is rather used for investigation of segregation, clustering and fine scale precip- itation studies [25, 26, 27, 28, 29]. This method can also be used to identify crystallographic planes and dislocations which are hard to spot in a 2DFIM image. Both the atom by atom reconstruction approach and the geometrical method suffer from their own limitations. For instance, the atom-by-atom approach is limited to regions with atomic resolution, and thus a 3D recon- struction is only possible around certain high index facets. The geometrical approach looses the atomic positioning precision due to the simplistic geo- metrical assumptions of the tip shape. In the following sections we showcase how various data extraction methods can be employed to further improve the atom-by-atom analysis approach and recover as much positioning infor- mation as possible. In addition, the use of machine learning algorithms to extract the physics behind field ionization and evaporation is also explored. 3. Supervised and unsupervised machine learning Machine learning (ML) algorithms are currently exploited to derive sys- tematic insights from very rich experimental datasets and for solving complex problems in various disciplines [30]. Progress in ML has lead to decision rules that can in some cases be automatically derived by specific algorithms that are capable of learning, whilst exploiting the speed and the robustness of the available advanced computer infrastructure. 5 Machine-learning methods can be grouped into two major categories de- pending on the approach to a given problem viz. supervised and unsupervised learning [31]. Supervised learning algorithms try to identify the relationship between input and output. This dependency is learned as a function f (x) by using a set of labeled data {X = [ai, bi], i = 1, ..., N} consisting of N pairs (a1, b1), (a2, b2), ... , (aN , bN ), where the input variables ai are D-dimensional vectors ai ∈ RD and the output variables (or system responses) bi are dis- crete values (e.g., Boolean) for classification problems and continuous values (b ∈ R) for regression tasks. Support Vector Machines (SVMs) and Artificial Neural Networks (ANN) are widely used techniques that fall in this category. Typical tasks that can currently routinely be carried out by a supervised ma- chine learning algorithm are image segmentation and classification. In the computer vision community, semantic segmentation, which is an extremely challenging task, aims to partition the image into semantically meaningful parts (such as differentiating a cat, a car or Einstein in the same image), and to classify each part into one of the predetermined classes making use of ML or the more advanced deep learning methods [32]. 3DFIM represents a very good example of a complex problem where rather large experimental datasets are available and which are to date widely under-exploited. For the case of FIM data analysis, this concept could help in determining whether a group of pixels represents, e.g. an atom, a defect such as a vacancy, a dislo- cation or a grain boundary. This idea can be also extended to 3DFIM, where the data can be subjected to systematic segmentation and classification into sub-volumes that may contain or represent these various features. The moti- vation for using such algorithms lies in enabling the automated reconstruction of a fully atomically-resolved three dimensional point cloud representing the imaged specimen, including its population of crystalline defects. However, to apply deep learning algorithms for semantic segmentation, the structure of 3DFIM data must be defined precisely, followed by labeling of a large enough dataset to be used as a training dataset for the algorithm. Unsupervised learning refers to any ML process that attempts to learn the structure in an unlabeled dataset {X = [ai], i = 1, ..., N}, where ai ∈ RD in the absence of the output variables bi [31]. Various clustering techniques and dimensionality reduction routines fall into this category. The fact that the unsupervised learning techniques do not need a labeled training dataset is a considerable advantage. However, since there is usually no sufficiently robust or large dataset in this field to train the algorithm reliably enough, accuracy that can be obtained by these methods is usually lower than that 6 Figure 1: FIM image of pure bcc W obtained using He as imaging gas. Regions with atomic resolution are highlighted. The region marked by 1, is used for identifying all the atoms (marked in red in the upper right corner). Machine learning is then used to assign correct labels to those atoms that assemble crystallographic planes. In this case, atoms labeled blue belong to the first, black to the second and purple to the third plane layer, respectively, assembling altogether a pyramid type motif (lower right corner). achieved by the help of supervised learning algorithms. In this paper, we focus on unsupervised methods to identify the underlying structure of the FIM data, to visualize them and to label them for further application of supervised learning techniques. 4. Advanced data mining routines for 3D field ion microscopy In this section we describe the advanced data mining routines which in- clude unsupervised learning routines (Sec. 4.1) and a routine purely based on image processing (Sec. 4.2). To demonstrate the utility of these methods we apply them to a 3DFIM dataset obtained for bcc tungsten. The speci- men for 3DFIM was obtained from a tungsten wire oriented along the [011] direction. A very sharp needle like shape was achieved by using electrochem- ical polishing at 5-8 V using AC in a 5% molar NaOH solution. The local radius across the main [011] pole was estimated to be approximately 20 nm by the ring counting method [33]. 3DFIM experiments were performed on a 3DAP-LAR FIM [34]. FIM images were obtained on a phosphor screen, 7 and recorded using a CCD camera (AVT Stingray) at 15 fps rate. The res- olution of each recorded image was set to 1280 pixels × 960 pixels. From this complete image, a region of interest (ROI) was selected around the (222) crystallographic plane. Each image of the ROI is 150 × 150 pixels in size. A set of (222) planes was chosen for the data analysis as the atomic density of these planes in a body centered cubic structure is sufficiently low to produce atomic resolution in FIM. In the following, "FIM image" generally refers to the selected ROI image of 150 × 150 pixels. 4.1. Unsupervised machine learning applied to 3DFIM Here we use unsupervised machine learning to understand the underlying structure of 3DFIM data and use this information to test supervised learning algorithms for data extraction. It will be seen that the extracted structure of 3DFIM data is related to various field evaporation phenomena. Although these phenomena of field evaporation have been long established, the ability of a machine to decipher this structure from 3DFIM data serves as a proof of concept. 4.1.1. Dimensionality reduction In problems with high levels of complexity, one of the challenges lies in ex- tracting information from a dataset that is made of a large number of samples and where each sample has high-dimensionality. The 3DFIM dataset used herein contains 10,000 FIM images each with an image size of 150 × 150 pixels, i.e. 22,500 dimensions (considering only the ROI). Relevant extrac- tion of information from such data requires identifying pertinent variables, determining the interaction of these variables with each other, and then refor- mulating the data using exclusively these specific variables. This procedure is referred to as "dimensionality reduction" and simplifies any further pro- cessing. Mathematical hurdles of working with high-dimensional data are often called the curse of dimensionality [35]. When the dimensionality of the data increases, a good representation of the data in 2D or 3D Euclidean spaces can be very helpful to reveal various phenomena present in the data but such a representation might not always be possible [36]. For data with high dimensionality, two complementary strategies are em- ployed to either avoid or at least reduce issues related to dimensionality. First, there are often a number of input variables that do not or only weakly affect the output and can hence be considered as irrelevant and ignored. Second, the dependencies amongst the pertinent variables are established. 8 Once the relevant variables have been established, the dimensionality of the observed data tends to be still larger than necessary. Let us consider two highly correlated variables where information about one can be derived from the other. Correlations between such variables can be sometimes very com- plex and retaining only one of them might not be sufficient to encompass the whole information. Hence, instead of arbitrarily removing a variable from the pair, the best is to reduce the dimensionality of the data by trying to find a new set of transformed variables that keeps as much of the available information as possible. The new set of variables must possess the following features: It must 1) contain fewer variables than the original set and 2) preserve the most im- portant information contained in the initial dataset. We hence seek a trans- formation that contains and shows the same information from a different perspective, i.e. a projection which preserves the geometry whilst represent- ing the relevant objects [37]. Linear and nonlinear transformations of the input used herein are referred to as projections, mainly because all these transformations aspire to preserve the characteristics that are geometrical or have a geometrical interpretation. In other words, when there is a depen- dency amongst two or more variables, their joint distribution does not spread over the whole space. The dependencies in the data create a structure in the distribution, that can be seen in the form of a geometrical locus. Dimension- ality reduction attempts to eliminate any redundancies in the initial set of variables. Linear multivariate analyses such as principal component analysis (PCA) or multidimensional scaling (MDS) have long been used to find such projections, but cannot always reveal low-dimensionality structures when the manifold formed by the geometrical loci of data is curved [38]. Hence, there has been recent interest in algorithms identifying non-linear manifolds in data [39, 36, 37]. There are many, somewhat heuristic, methods to discover non-linear manifolds based on the preservation of the geometric properties of local neighborhoods within the data, while embedding, unfolding or oth- erwise projecting the data to occupy fewer dimensions. Algorithms such as Isometric feature mapping (Isomap) or maximum variance unfolding (MVU) attempt to preserve local distances. They estimate global manifold proper- ties by continuation across neighborhoods and then project the manifold to lower dimensions by classical methods such as PCA or MDS. In the follow- ing, we discuss the application of a linear method (PCA) and a non-linear method (Isomap) to the 3DFIM dataset. 9 Principal Component Analysis PCA is a popular technique for dimensionality reduction that aims to identify the most consequential basis to re-express a given dataset. The new basis should reveal the underlying structure in the dataset and filter out the noise [36]. Mathematically, we can define PCA as follows: For N high di- mensional input variables {X = [ai], i = 1, ..., N}, where ai ∈ RD, PCA is employed to find a linear subspace with lower dimensionality d (d ≤ D), such that the maximum variance is maintained in the subspace. The linear subspace can be defined by d orthogonal vectors also called principal com- ponents, say u1, u2, ..., ud, forming a new coordinate system. Ideally d (cid:28) D In other words, PCA aims to reduce the (worst case would be d = D). dimensionality of the data, while preserving its information content which comes from the variation (or, equivalently, minimizing the loss of informa- tion). The principal components are given by the top eigenvectors of the D × D covariance matrix C = 1 i , with i running over the rows of the matrix X [36]. i Xi · X T (cid:80) N Figure 2 shows the result of projecting our 3DFIM dataset onto its first two principal components. In this plot, each point represents one micrograph (with a dimensionality of 150 × 150), and is color-coded according to the number of atoms in the first terrace of the corresponding picture (see Fig. 1). The numbers on the outer circle represent the average number of atoms on the first terrace of images in the corresponding slice of the circle. The PCA result reveals a cyclic pattern in the 3DFIM dataset and a specific grouping of the colors. Note that starting with the gray color cloud to the left representing a terrace of 27 atoms and going counter-clockwise the number of atoms in the terrace gradually decreases until 3, before a jump to the next terrace occurs. Isomap Isomap is a nonlinear dimensionality reduction method that is built on a linear dimensionality reduction approach, more precisely it is built upon a classical multidimensional scaling (MDS) (see reference [40] for details). In MDS, the input X is projected into a smaller subspace with dimension d by preserving the pairwise Euclidean distances Xi − Xj2, or the dot products Xi · Xj, i.e. the L2 norm. The orthogonal vectors, w1, w2, ..., wd, which form the subspace are the top d eigenvectors of the N × N Gram matrix with elements Gij = Xi · Xj [37]. Isomap also tries to preserve the local geometry of the data. To this end, one assumes that the high-dimensional data in X lie on a low-dimensional manifold M , and one replaces the Euclidean dis- 10 Figure 2: PCA applied to the 3DFIM dataset. Each point represents one FIM picture in this dataset, and its color indicates the number of atoms in the first terrace of the corresponding picture. The numbers on the outer circle represent the averages over the number of atoms in the first terrace for that particular arc of the circle. tances between the points in the high-dimensional space with the distances on this manifold. Since the manifold is unknown, to a first approximation, a k-nearest neighbor graph between points in the dataset is considered, where k needs to be chosen such that any two data points in the dataset are con- nected by at least one path but it should not become too large to have a computationally efficient approach. The shortest path between the points is then used to approximate the distance on the manifold. The complete Isomap algorithm has three main steps [37]: 1. Constructing a k-nearest neighbor graph based on the distances dX(i, j) between pairs of points i, j in the input space X. Either all the points within some fixed radius are connected to each point, or each point is connected with each of its k-nearest neighbors. These neighborhood relations are expressed as a weighted graph G over all the data points, where edges are represented as weights dX(i, j) between the neighboring points. 11 2. The geodesic distances dM (i,j) between all pairs of points on the man- ifold M are estimated by Isomap from computing their shortest path distances dG(i,j) in the graph G. 3. Finally a MDS is applied to a matrix of graph distances DG = {dG(i, j)}, which embeds the data in a d-dimensional Euclidean space Y which preserves the manifold's estimated intrinsic geometry. The coordinate vectors yi for points in Y are chosen to minimize the cost function: E = τ (DG) − τ (DY )L2 i,j A2 (cid:113)(cid:80) (1) where DY denotes the matrix of Euclidean distances dY (i, j) = yi − yj and AL2 the L2 matrix norm ij . Distances are converted into the inner products by the τ operator, these characterize the geometry of the data in a form that supports efficient optimization. The global minimum of Eq. (1) is achieved by setting the coordinates yi to the top d eigenvectors of the matrix DG. Figure 3 shows the result of applying Isomap to the 3DFIM dataset. Each point represents one micrograph (with dimensionality of 150 × 150) in a reduced 2-dimensional space. Each image is color-coded according to the number of atoms in the first terrace. Again, the numbers on the outer circle represent the averages over the number of atoms in the first terrace of pictures whose Isomap representations are in a slice of the circle with a corresponding angle. The Isomap result reveals a similar cyclic behavior in the dataset as found with PCA. This cyclic behavior detected by both linear and non linear dimensionality reduction algorithms is a consequence of the field evaporation behavior in crystalline materials. Field evaporation in many pure and crystalline materials occurs layer-by-layer and proceeds from atoms sitting at the ledge of a terrace and towards the center. When a terrace field evaporates the atoms sitting on the ledge disappear decreasing the number of atoms on the terrace as the evaporation proceeds. As shown in Fig. 4(b), by evaporating atoms from the surface of the sample during FIM, the first terrace area decreases until it evaporates completely and the subsequent terrace is exposed. This field evaporation behavior is seen for all terraces. The number of cycles is a measure of the number of atomic terraces evaporated/analyzed during the 3DFIM. Figure 4 (a) shows the angles of the first 500 3DFIM images after projec- tion onto the Isomap space with horizontal axes (see Fig. 3). Each period 12 Figure 3: Isomap manifold learning on the 3DFIM dataset. Each point represents one picture in the reduced dimensional space, and its color indicates the number of atoms in the first terrace of the corresponding picture. The numbers on the outer circle represent the averages over the number of atoms in the first terrace of pictures. of this plot represents the field evaporation of one terrace from the moment of its exposure to the surface until its complete evaporation. Figure 4 (b) depicts the evolution of the FIM pictures and the number of atoms in the first terrace during one of these periods (highlighted in Fig. 4(a) by the red dashed rectangle). Although Isomap is built on an MDS approach, the MDS algorithm ap- plied to the same dataset is not able to reveal the cyclic pattern observed within the data (see Fig. 5). The reason is that the objective function in MDS aims to preserve the pairwise Euclidean distances in the low-dimensional space and therefore it fails to appropriately represent the 3DFIM dataset. A comparison between the projections obtained from Isomap and PCA shows that the Isomap projection reveals finer details and enables a better separa- tion between micrographs based on the number of atoms in the first terraces (see the color of points in Fig. 3). However, there is still an overlap of colors 13 Figure 4: (a) Angles of the first 500 3DFIM images after projection onto the Isomap space with horizontal axes (see Fig. 3). (b) Evolution of a terrace seen from FIM pictures and the number of atoms in the first terrace of the highlighted period in (a) (red dashed rectangle). especially in the images with a higher number of atoms in the first terrace. The reason is that for high numbers of atoms in the first terrace, the number of available configurations that these atoms can adopt is significant. Hence an ideal projection would be one that could distinguish the data based on the configuration of atoms in the corresponding image. To this end, the lo- cal information within the images should also be taken into account whilst calculating the similarity (e.g. pair distances) between images. This could be achieved by using autoencoders in combination with convolutional neural networks, which is beyond the scope of the current article. (See reference [41] for additional details.) We showed here that a complex 3DFIM dataset can be used as input for dimensionality reduction efficiently unraveling the underlying structure 14 Figure 5: MDS applied to the 3DFIM dataset. Each point represents one picture in the reduced dimensional space, and its color indicates the number of atoms in the first terrace of the corresponding FIM picture. of the data. Further we discussed that the structure revealed by the linear and non-linear dimensionality reduction is in fact a consequence of the field evaporation behavior of a pure and crystalline material. The output of these algorithms was able to classify the data according to the number of atoms on the first terrace and also, to some extent, the configurations formed by those atoms. Further analyses can be performed in the direction of analyzing the stability of certain configurations for a given crystallographic plane. In the following, we describe some advanced image processing methods and clustering algorithms which can be used to maximize the data extracted from 3DFIM. 4.2. Image processing based data extraction routines applied to 3D field ion microscopy In a recent paper [19], we showed an improved method for reconstructing 3DFIM data. It was also shown recently [Katnagallu et al., 2017 communi- 15 Figure 6: Intensity at the center of the FIM image versus the image number. Peaks indicate the number of planes evaporated. Compare with Fig. 4 which also shows the same periodic behavior, a consequence of layer-by-layer evaporation. cated] that the higher the intensity of an imaged position, the less represen- tative is the imaged position of the true atomic position, which implies that the positions of atoms imaged in the center of a terrace are more accurate. Keeping this in mind, we developed another new routine for 3DFIM data reconstruction. In every 3DFIM data extraction routine, the first step is to determine the overall depth of the analyzed volume. In an atom-by-atom approach, this problem boils down to calculating the number of planes of in- terest that have evaporated, which can be directly derived from monitoring the intensity in the center of the image. As the evaporation proceeds from the edge of the terrace inwards, the intensity in the center keeps increasing. As the terrace collapses, the atoms left on the terrace are imaged brightly. The reason is the high local curvature leading to an increase in the local field, thus increasing the current of gas ions originating from above of these specific atoms. Such an analysis is shown in Fig. 6 revealing a clear periodic behav- ior where each peak indicates the final image of a terrace before complete evaporation. The same information can be obtained from an ML approach as was shown by the dimensionality reduction using PCA and Isomap (see Fig. 4(a)). The number of peaks in this intensity variation thus gives the number of planes evaporated. 16 After this number is computed, the image at which a new plane is exposed can also be calculated based on the peak positions. The determination of the depth of the data enables the next steps of the data extraction procedure. These steps can be classified into three categories a) identification of atoms (atom detection; Sec. 4.2.1) b) assigning the right crystallographic plane to the identified atoms (plane detection; Sec. 4.2.2) and c) depth assignment based on the number of planes evaporated. Depth assignment is easier to implement and as described above capitalizes on layer-by-layer evaporation, while the first two steps employ a plethora of algorithms encompassing vari- ous methods to build a 3D point cloud. The aim in the following is to look at these algorithms in terms of applicability and efficiency. 4.2.1. Atom detection Atom detection is the extraction of x and y coordinates of each atom from the digital FIM images. A closer inspection of a FIM image reveals that the images are made by a superposition of Gaussian-like distributions of intensities centered around the atomic positions. These atomic positions are also distorted by the image projection from the specimen onto the screen and by the atomic arrangement surrounding the imaged atom. Certain high index planes posses enough corrugation in the electric field so as to enable the ionization of gas atoms almost at every atomic site. The probability of field ionization and hence the intensity of the imaged spot depends on the electric field strength at that atomic site. This explains the intensity differences between the brightly imaged ledge atoms and dimly imaged inte- rior terrace atoms. Dagan's routine [18] was based on identifying the atoms which were imaged above a certain threshold intensity. As every atom gets brighter and brighter until it field evaporates, each atom can be detected over the course of the evaporation of a complete plane. This protocol also requires tracking the identified atoms until evaporation. Dagan's routine is well suited for reconstructing the data to get a statistical distribution of de- fects and/or ad-atoms. In order to understand the physics of field ionization and field evaporation, as well as to characterize the atomic strains in the im- aged lattice, it becomes critical to identify and track every atom right from the moment of its exposure to the surface. In this effort, we developed an alternative method based on edge detection to identify almost all atoms in each image. Two methods for identifying atoms based on edge detection are explained below. 17 Gaussian edge detection Since the Gaussian-shaped intensity distribution of each atom overlaps with that of the neighboring atoms, detection of individual atoms in a FIM image is not a trivial task. This is particularly the case for atoms from the center of the terrace, where the overlap is much higher as the intensity dis- tribution spreads more widely. To make these intensity distributions more distinguishable from each other, we apply a Laplace operator on the images. As shown in Fig. 7 for a one-dimensional Gaussian distribution, the applica- tion of a Laplacian operator localizes the intensity distribution for individual Gaussian distributions. In Fig. 7 the solid green line is formed by the combi- nation of two individual overlapping Gaussian distributions (the dashed green lines). The red line shows that, by applying the Laplacian operator, the two maxima corresponding to the individual Gaussian intensities (dashed green lines) can now be discriminated as a maximum and a shoulder. These prin- ciples are extended to the two dimensional FIM images. Figure 8(b) shows the result of applying the Laplacian operator to an original image shown in Fig. 8(a). For detecting the atoms, as shown in Fig. 7, the second derivative (blue line) of the localized Laplacian-modified signal can be used as it is more negative in the regions close to the original positions of the two atoms. In a similar way we use the second derivative of the Laplacian-modified image to identify the regions occupied by the atoms and then, the atom's positions are assigned according to the identified regions centroid. C spline fitting and peak detection Recently, we showed that the local electric field on the atomic sites re- arranges over the course of field evaporation giving an impression of atomic movement in FIM images [Katnagallu et al., 2017 communicated]. Another conclusion from this work was that atom positions extracted from less intense parts of the images are least erroneous. Hence for an accurate reconstruc- tion, the information from the center of the terrace needs to be maximized as the corresponding intensity is relatively low. We therefore describe next a method to extract atom positions from the center of a terrace. The key idea is to improve the edge detection at the center of the terrace by improving the signal to noise ratio (SNR). This is achieved by averaging a certain number of similar images of a given terrace. Provided that the evaporation rate is slow, more images can be recorded for a fixed analysis volume. The images required for this routine are the images of a terrace when it first appears. The image index corresponding to a terrace's first 18 Figure 7: A schematic showing the resulting intensity (solid green) of two overlapping Gaussians (dashed green) and the application of a Laplacian operator (red) localizing the two individual contributions. The second derivative (blue) of the Laplacian operator becomes negative in the relevant regions. Figure 8: Gaussian edge based detection of atoms in FIM image's ROI. Identified atoms shown in (b) from the original image (a) using this procedure are marked in green. 19 exposure to the surface can be inferred from the analysis shown in Fig. 6. The blue circles indicate the images where a new terrace has been exposed to the surface, right after evaporation of the previous terrace. To average a certain number of images with the identified image index, a comparison of similarity for the next few images is performed. The comparison is based on a structure similarity index measure (SSIM) [42]. The SSIM for two images i and j is given by SSIM(i, j) = [l(i, j)]α · [c(i, j)]β · [s(i, j)]γ, (2) which is a function of the intensity in both images l(i, j) given by the mean of the pixel values, the contrast of both images c(i, j) given by the variance of the pixel values, and the texture of both images s(i, j) given by the covariance of the pixel values in images i and j. α, β and γ are weights given to each term. In this case all the weights are set to 1. SSIM is 1 if the images are identical and is 0 when the images are completely different. Thus, the number of images to be averaged is automatically identified based on the computed SSIM. The threshold value was set so that the images to be averaged over should not differ by more than 0.90 of the SSIM. This value was set based on visual inspection of images where a SSIM of 0.90 corresponds to one to two atoms being evaporated from the first terrace in the image. Once the number of images to be averaged over is identified, a resultant averaged image is computed. Now the atoms are identified in this image based on a different edge based detection method. First the averaged image is fitted with a cubic spline function in two-dimensions. Using C-spline basis functions to represent a set of samples is advantageous. Operations such as derivation, integration etc., which assume that the data samples are drawn from an underlying continuous function can be computed from the spline coefficients1. A Laplacian operator is then convolved with the fitted C-spline function to identify the edges of the given image. The output of the previous operation is used to compute the maxima inside these edges. These local maxima are then recorded as atomic positions. For the case of the tungsten FIM images around the (222) plane where 67 (222) planes were evaporated, the above explained procedure was able to identify 98 % of atoms. The various steps of the explained method are shown in the Fig. 9 for one of the (222) planes. 1https://docs.scipy.org/doc/scipy/reference/tutorial/signal.html 20 Figure 9: C spline fitting and local maxima based detection of atoms in an averaged image. The average image is computed based on the difference image and the spline edge filter and peak based detection are applied to identify the atoms. All atoms are identified due to improved SNR and are marked in blue. 21 4.2.2. Plane determination Once the atoms are identified, the detected positions have to be correctly assigned to their corresponding crystallographic planes. As can be seen in Fig. 9 ("Original image"), at least two (222) terraces are imaged in the chosen ROI. The outermost terrace is bounded by the brightly imaged atoms and the atoms lying outside this boundary are not of the first terrace. In order to correctly assign the depth coordinate, these atoms need to be classified into those belonging to the top terrace and those not. The methods used to achieve this goal are based on image processing algorithms or on spectral clustering. Image processing based techniques Classifying the atoms according to their crystallographic planes is not trivial, especially using image processing based techniques. Image processing based algorithms rely on identifying the boundaries of the outermost terrace. Since atoms at the ledge of the terrace have a smaller number of nearest neighbors and are the sites of higher electric field, they appear more brightly in FIM. Thus the atoms imaged from the ledge are more similar to themselves than the images of atoms from the terrace's center. This information can be exploited by virtue of the "entropy" of the image [43]. The entropy, H, of the image is given as H = −(cid:88) pk log2(pk), (3) k where k is the number of gray levels and pk is the probability associated with gray level k. Since the atoms at the ledge sites are imaged similarly the pixels in the imaged atoms are associated with a similar pk. An implementation of this entropy as a filter can be used to identify the ledge atoms. Which gives us the boundary for the outermost terrace. The entropy can thus be used as a way to classify atoms to their corresponding terraces. The result of this entropy filter on one of the FIM images is shown in Fig. 10. Other image processing based techniques for this task are watershed segmentation and Fourier filtering based methods. A detailed explanation of these methods is beyond the scope of the current article. Spectral clustering The task of identifying elements in a given dataset based on similarity or dissimilarity amongst them is defined as clustering [44]. Clustering is a type of unsupervised learning in which the goal is to partition a set of elements into 22 Figure 10: A successful application of the entropy based filter to identify the outermost terrace atoms. groups called clusters. Differences between various clustering algorithms are mainly about the ways similarity between the data points is measured [45], typically through various distortion or distance measures. We now discuss the so-called spectral clustering applied to plane determination. Eigenvector techniques are frequently employed in multidimensional data in order to extract the underlying correlations of the data (see Sec. 4.1.1). Similarly one can apply such techniques for clustering. Spectral clustering is an Eigenvector based technique. In spectral clustering, we benefit from the node-node adjacency matrix of the graph. For a given graph containing N nodes (each node corresponds to one point in the dataset), we create a N × N adjacency matrix, in which each entry (i, j) corresponds to the weight of the edge between the nodes i and j. This essentially corresponds to the similarity between samples i and j in the dataset. Such weights wij are recorded in a matrix W . As we are working with undirected graphs, the matrix is assumed to be symmetric. It implies that wij = wji ∀ (i, j). Now any clustering algorithm will try to minimize the weights across the clusters. In spectral clustering, the minimization function is constructed based on the adjacency weight matrix and another diagonal matrix called a degree matrix D. Each element in D is the degree of every vertex in the similarity graph, such that dii is equal to the sum of the weights of the incident edges, so dii = n(cid:80) wij. j=1 23 In addition, we formally define the Laplacian matrix L as follows: L is defined by subtracting the weighted adjacency matrix from the degree matrix. Hence we have L = D − W . The graph Laplacian matrix defined in this way conceals both its structural and its eigenvector behavior. Such a graph Laplacian can now be adopted to identify the pertinent clusters in the data. The number of connected components in the underlying graph can be related to the number of eigenvectors with zero eigenvalues for the Laplacian matrix L as they are equal. The result of spectral clustering applied to our 3DFIM dataset is shown in Fig. 11. As a notion of similarity a combination of two metrics is used. The first metric is the distance between each of the identified atoms, the second is the intensity of the identified atom. Spectral clustering was able to identify two distinct planes in each FIM image as can be seen in Fig. 11. Some points near the top right corner were associated to the wrong cluster. This solution to plane determination is still not completely efficient and, here again, the ML could provide significant improvement. The ML algorithms need labeled training data, which will be first labeled by employing the image processing and clustering algorithms outlined herein. The accuracy of these algorithms on each image then needs to be manually verified and that data can be used as a training dataset. 4.2.3. Atom tracking Since atoms are imaged multiple times before their evaporation, each atom can be tracked through multiple images. This allows us to identify the position of each imaged atom as a function of its surrounding configuration and also of the operating conditions. In a recent work we used this approach to track the atoms' image positions. We concluded that after field evapora- tion events there is a local rearrangement of the local electric field, leading to a more precise understanding of how field ion images are formed. Depending on the task and the reconstruction routine used, it might be necessary to track atoms. In the following we describe another clustering algorithm to easily track atoms through multiple images if required. Hierarchal clustering Tracking the same atom imaged multiple times can be considered as a clustering problem. The identified (x, y) coordinates need to be clustered along z such that the final clustering construction is comprised of a set of 24 Figure 11: Two clusters resulting from spectral clustering indicating two planes present in the given FIM images. 25 n(cid:83) i=1 Ki and Ka ∩ Kb = ∅, ∀ subsets K = K1, K2, ..., Kn in M , where M = a (cid:54)= b [46]. The method used here to cluster these data points is hierarchical In an agglomerative hierarchical clustering, initially every clustering [47]. data point is considered as a cluster. These clusters are then merged suc- cessively until the sought after cluster construction is found. The merging of clusters is based on a criterion which is a measure of similarity of these clusters. Here, we use a single linkage criterion to connect the clusters. In this criterion, the distance between any two clusters is defined as the shortest distance between any data point of one cluster to any member of the other. The clustering decision is made based on which of the cluster distances is the minimum and those two clusters are merged. The result of using a single linkage hierarchical algorithm on (x, y) coordinates in images corresponding to the evaporation of one (222) terrace is shown in the Fig. 12. The result is the clustering of theses (x, y) coordinates to form individual atoms tracked through successive images. Each cluster is assigned a number as indicated by the color bar in Fig. 12. Visual inspection reveals that coordinates corre- sponding to the same atom through the successive images do not displace by more than 10 pixels. This value was used as a decision metric for the single linkage criterion. The physical grounds for these displacements are discussed in a recent article [Katnagallu et al., 2017 communicated]. 5. A word about 3D field ion microscopy reconstruction After the data are extracted and labeled correctly, in accordance with pre- vious sections, the data need to be converted to absolute length scales, that are meaningful to materials physics. Previous atom-by-atom approaches, used to calibrate the measured distances between atoms on the image with theoretical nearest neighbor spacing for the crystallographic plane analyzed. Such conversions to an absolute length scale inherently assume that there are no surface relaxations for atoms imaged and atoms are at their equilibrium positions. Figure 13 shows a typical 3D reconstruction of extracted 3DFIM data. The x and y coordinates are in pixels and the z axis is the plane num- ber. The periodicity of the reconstruction is evident as planes in the figure. The noticeable noise in the reconstruction still needs to be understood. One of the sources could be attributed to errors in data extraction. But also, as mentioned before, the image of the local electric field is not directly cor- related with the true atomic positions. Nevertheless the distortion in the 26 Figure 12: Atoms tracked through a sequence of 69 images using hierarchal clustering. Each color indicates the cluster of imaged positions of one atom through the sequence. Images on the left show the start (left bottom) and the end (left top) images of the 69 images sequence. reconstructed volume can be expunged by invoking a molecular static relax- ation protocol, in which the reconstructed volume is encased in a similarly oriented perfect lattice. The encasing is done after removing the equivalent atomic sites from the perfect volume. Then using a potential for the material analyzed, a molecular relaxation is done on the encased data in LAMMPS [48]. Employing such a routine also preserves the defects recorded in the analyzed volume. The application of this routine on an analyzed 3DFIM volume of W with vacancies can be found in reference [19]. These conversions and relaxations do not affect the statistical distribu- tions of point defects [24], and hence the precision of reconstruction is not a key requirement. On the other hand, if 3DFIM is employed for characteri- zation and quantification of lattice strains, or strains surrounding crystalline defects the precision of the reconstruction needs to be greatly improved. It is important to make a distinction between the resolution of a FIM image and its intrinsic precision. The resolution within a FIM image can be seen as the size of the atom in the image which is determined by the size of the ionization zone, the lateral velocity of the gas atom at the time of ionization and the inherent Heisenberg's uncertainty [49, 50]. Whereas the precision of FIM is 27 Figure 13: Final 3D reconstruction of data extracted from 3DFIM, the dimensions are arbitrary image units. how well the imaged atom positions correspond to the true atomic sites. Im- proving the resolution of FIM images can only originate from a reduction in the base temperature of the specimen. However, extracting precise informa- tion on the atomic locations requires a thorough comprehension of the image formation in FIM, which directly relates to the details of the complexity of the electric field distribution and its rearrangement when an atom has field evaporated. The presence and the type of defects in the material affects the local bonding and the local arrangement of the atoms on the surface, and hence their effect on the local electric field needs to be precisely quantified. This may be achieved, for instance, by applying ML to the vast amounts of labeled data accessible by 3DFIM. 6. Outlook Field ion microscopy has provided a unique vantage point for atomic scale characterization of materials. The inherent resolution of FIM and the ability to also gather three dimensional information from materials through 3DFIM, puts the technique far ahead of any other contemporary microscopy technique. But the lack of advanced data processing and extraction routines are one of the reasons that likely hindered FIM from becoming a mainstream characterization technique. This article has detailed a burgeoning frame- work that will allow full data extraction, which we hope will support the 28 renewed interest in FIM and its extension to 3DFIM. The pure image pro- cessing techniques have helped us to semi-automate the data extraction from FIM images. The application of machine learning algorithms to our data not only showed the behavior of field evaporation but also helped us to improve the accuracy of the reconstructed data. The layer by layer field evaporation behavior was evident when using PCA or Isomap algorithms on the 3DFIM data. Also using Isomap clustered the images better with respect to num- ber of atoms an image had on its first terrace. The semi automated data extraction routine based on image processing has been helpful not only to improve the data extraction but also helps us to get much more accurately labeled data which can be used as a training dataset for supervised machine learning. With the use of such advanced algorithms for data extraction, we hope not only to completely automate the data extraction from 3DFIM but also identify and characterize various material defects. Apart from data extraction, machine learning has been gaining popularity in identifying fun- damental physics phenomena [51]. The physics of image formation in FIM is still not completely understood, and it is our firm belief that machine learning can be a powerful tool in this direction. The discussed methods have been implemented in a set of Python rou- tines. These routines are available upon request to the authors. 7. Acknowledgments SK and BG are grateful to Dr. Michal Dagan, Department of materials, University of Oxford, for providing us with 3DFIM data on tungsten. SK acknowledges IMPRS SurMat for financial support. The funding by the European Research Council (ERC) under the EUs Horizon 2020 Research and Innovation Programme (Grant No. 639211) is gratefully acknowledged. 8. References References [1] E. W. Muller, "Ergebn. d. exakten naturwissenschaften. 27 (1953) 330," Z. Elektrochem, vol. 59, p. 372, 1955. [2] E. W. Muller, "Das Feldionenmikroskop," Zeitschrift fur Physik, vol. 131, no. 1, pp. 136–142, 1951. 29 [3] E. W. Muller, "Resolution of the atomic structure of a metal surface by the field ion microscope," Journal of Applied Physics, vol. 27, no. 5, pp. 474–476, 1956. [4] M. K. Miller and R. G. Forbes, Atom-probe tomography: the local elec- trode atom probe. Springer, 2014. [5] T. J. Prosa and D. J. Larson, "Modern focused-ion-beam-based site- specific specimen preparation for atom probe tomography," Microscopy and Microanalysis, vol. 23, no. 2, pp. 194–209, 2017. [6] C. de Castilho and D. Kingham, "Best image conditions in field ion microscopy," Le Journal de Physique Colloques, vol. 47, no. C2, pp. C2– 23, 1986. [7] Y. C. Chen and D. N. Seidman, "The field ionization characteristics of individual atomic planes," Surface Science, vol. 27, no. 2, pp. 231–255, 1971. [8] W. A. Schmidt, N. Ernst, and Y. Suchorski, "Local electric fields at indi- vidual atomic surface sites: field ion appearance energy measurements," Applied Surface Science, vol. 67, no. 1-4, pp. 101–110, 1993. [9] T. T. Tsong, Atom-probe field ion microscopy: field ion emission, and surfaces and interfaces at atomic resolution. Cambridge University Press, 2005. [10] D. J. Larson, T. J. Prosa, R. M. Ulfig, B. P. Geiser, and T. F. Kelly, "Local electrode atom probe tomography," NY: Springer, New York, 2013. [11] D. G. Brandon, B. Ralph, S. Ranganathan, and M. S. Wald, "A field ion microscope study of atomic configuration at grain boundaries," Acta Metallurgica, vol. 12, no. 7, pp. 813–821, 1964. [12] S. Ranganathan and J. B. Thompson, "Field Ion Microscopic Observa- tions of Dislocation Structures at Grain Boundaries," Journal of Applied Physics J. Sci. lnstr. J. Appl. Phys. Journal of Applied Physics, vol. 37, no. 12, 1966. 30 [13] B. Deconihout, A. Bostel, P. Bas, S. Chambreland, L. Letellier, F. Danoix, and D. Blavette, "Investigation of some selected metallurgical problems with the tomographic atom probe," Applied Surface Science, vol. 76-77, pp. 145–154, 1994. [14] G. D. W. Smith, D. Hudson, P. D. Styman, and C. A. Williams, "Studies of dislocations by field ion microscopy and atom probe tomography," Philosophical Magazine, vol. 93, no. 28-30, pp. 3726–3740, 2013. [15] D. A. Smith, M. A. Fortes, A. Kelly, and B. Ralph, "Contrast from stacking faults and partial dislocations in field-ion microscope," Philo- sophical Magazine, vol. 17, no. 149, pp. 1065–1077, 1968. [16] R. Taunt and B. Ralph, "Observations of the fine structure of superdis- locations in Ni3Al by field-ion microscopy," Philosophical Magazine, vol. 30, no. 6, pp. 1379–1394, 1974. [17] F. Vurpillot, M. Gilbert, and B. Deconihout, "Towards the three- dimensional field ion microscope," Surface and Interface Analysis, vol. 39, pp. 273–277, feb 2007. [18] M. Dagan, B. Gault, G. D. W. Smith, P. A. J. Bagot, and M. P. Moody, "Automated Atom-By-Atom Three-Dimensional (3D) Reconstruction of Field Ion Microscopy Data," Microsc. Microanal, vol. 23, pp. 255–268, 2017. [19] S. Katnagallu, A. Nematollahi, M. Dagan, M. Moody, B. Grabowski, B. Gault, D. Raabe, and J. Neugebauer, "High fidelity reconstruction of experimental field ion microscopy data by atomic relaxation simu- lations," Microscopy and Microanalysis, vol. 23, no. S1, pp. 642–643, 2017. [20] E. Jones, T. Oliphant, and P. Peterson, "{SciPy}: open source scientific tools for {Python}," 2014. [21] F. Vurpillot, F. Danoix, M. Gilbert, S. Koelling, M. Dagan, and D. N. Seidman, "True atomic-scale imaging in three dimensions: A review of the rebirth of field-ion microscopy," Microscopy and Microanalysis, vol. 23, no. 2, pp. 210–220, 2017. 31 [22] R. M. Scanlan, D. L. Styris, and D. N. Seidman, "An in-situ field ion microscope study of irradiated tungsten: Part II Analysis and Interpre- tation," Philosophical Magazine, vol. 23, no. 186, pp. 1459–1478, 1971. [23] L. A. Beavan, R. Scanlan, and D. Seidman, "The defect structure of depleted zones in irradiated tungsten," Acta Metallurgica, vol. 19, no. 12, pp. 1339–1350, 1971. [24] M. Dagan, L. R. Hanna, A. Xu, S. G. Roberts, G. D. W. Smith, B. Gault, P. D. Edmondson, P. A. J. Bagot, and M. P. Moody, "Imaging of radia- tion damage using complementary field ion microscopy and atom probe tomography," Ultramicroscopy, vol. 159, pp. 387–394, 2015. [25] A. Cerezo, M. G. Hetherington, J. M. Hyde, M. K. Miller, G. D. W. Smith ", and J. S. Underkoffier, "Visualisation of three-dimensional mi- crostructures," Surface Science, vol. 266, pp. 471–480, 1992. [26] S. Cazottes, F. Vurpillot, A. Fnidiki, D. Lemarchand, M. Baricco, and F. Danoix, "Nanometer scale tomographic investigation of fine scale precipitates in a cufeni granular system by three-dimensional field ion microscopy," Microscopy and Microanalysis, vol. 18, no. 5, pp. 1129– 1134, 2012. [27] P. Jessner, R. Danoix, B. Hannoyer, and F. Danoix, "Investigations of the nitrided subsurface layers of an fe–cr-model alloy," Ultramicroscopy, vol. 109, no. 5, pp. 530–534, 2009. [28] F. Danoix, T. Epicier, F. Vurpillot, and D. Blavette, "Atomic-scale imaging and analysis of single layer gp zones in a model steel," Journal of materials science, vol. 47, no. 3, pp. 1567–1571, 2012. [29] J. Akr´e, F. Danoix, H. Leitner, and P. Auger, "The morphology of secondary-hardening carbides in a martensitic steel at the peak hardness by 3DFIM," Ultramicroscopy, vol. 109, no. 5, pp. 518–523, 2009. [30] E. Mjolsness and D. DeCoste, "Machine learning for science: state of the art and future prospects," Science, vol. 293, no. 5537, pp. 2051–2055, 2001. [31] T.-M. Huang, V. Kecman, and I. Kopriva, Kernel based algorithms for mining huge data sets, vol. 1. Springer, 2006. 32 [32] A. Garcia-Garcia, S. Orts-Escolano, S. Oprea, V. Villena-Martinez, and J. Garcia-Rodriguez, "A review on deep learning techniques applied to semantic segmentation," arXiv preprint arXiv:1704.06857, 2017. [33] M. Drechsler and P. Wolf, "Zur analyse von feldionenmikroskop- aufnahmen mit atomarer auflosung," in Physikalisch-Technischer Teil, pp. 835–848, Springer, 1960. [34] P. Panayi, "A reflectron for use in a three-dimensional atom probe," 11 2006. [35] J. Friedman, T. Hastie, and R. Tibshirani, The elements of statistical learning, vol. 1. Springer series in statistics New York, 2001. [36] J. A. Lee and M. Verleysen, Nonlinear dimensionality reduction. Springer Science & Business Media, 2007. [37] K. Q. Weinberger and L. K. Saul, "Unsupervised learning of image man- ifolds by semidefinite programming," International journal of computer vision, vol. 70, no. 1, pp. 77–90, 2006. [38] M. Vlachos, Dimensionality Reduction. Boston, MA: Springer US, 2017. [39] J. B. Tenenbaum, V. De Silva, and J. C. Langford, "A global geometric framework for nonlinear dimensionality reduction," science, vol. 290, no. 5500, pp. 2319–2323, 2000. [40] M. A. Cox and T. F. Cox, "Multidimensional scaling," Handbook of data visualization, pp. 315–347, 2008. [41] I. Goodfellow, Y. Bengio, and A. Courville, "Autoencoders," in Deep learning, pp. 499–523, MIT Press, 2016. [42] L. Zhang, L. Zhang, X. Mou, and D. Zhang, "Fsim: A feature similar- ity index for image quality assessment," IEEE transactions on Image Processing, vol. 20, no. 8, pp. 2378–2386, 2011. [43] R. Lewis, Practical digital image processing. Prentice Hall PTR, 1990. [44] C. Sammut and G. I. Webb, eds., Clustering, pp. 226–226. Boston, MA: Springer US, 2017. 33 [45] A. K. Jain, M. N. Murty, and P. J. Flynn, "Data clustering: a review," ACM computing surveys (CSUR), vol. 31, no. 3, pp. 264–323, 1999. [46] L. Rokach and O. Maimon, Clustering Methods, pp. 321–352. Boston, MA: Springer US, 2005. [47] C. Fraley and A. E. Raftery, "How many clusters? which clustering method? answers via model-based cluster analysis," The computer jour- nal, vol. 41, no. 8, pp. 578–588, 1998. [48] S. Plimpton, P. Crozier, and A. Thompson, "Lammps-large-scale atomic/molecular massively parallel simulator," Sandia National Lab- oratories, vol. 18, 2007. [49] R. Gomer, "Velocity Distribution of Electrons in Field Emission. Reso- lution in the Projection Microscope," The Journal of Chemical Physics, vol. 1772, no. 1952, 1952. [50] Y. C. Chen and D. N. Seidman, "On the atomic resolution of a field ion microscope," Surface Science, vol. 26, no. 1, pp. 61–84, 1971. [51] L. Zdeborov´a, "Machine learning: New tool in the box," Nature Physics, vol. 13, no. 5, pp. 420–421, 2017. 34
1808.09418	1	1808	2018-07-11T16:11:19	The Diffusion-Hardening Effect on the Technological Properties of High-Temperature Steel	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The results of research pipes diffusion hardening as an effective method for increasing the durability and reliability of power equipment are presented. The experience of commercial operation of pipes manufactured from diffusion chromized and heat-hardened steel 14 MoV6-3; on the heating surfaces of the supercritical pressure boilers are generalized. The possibility and effectiveness of this method on the example of capacities of Trypil'ska CHP are shown.	physics.app-ph	physics	The Diffusion-Hardening Effect on the Technological Properties of High- Temperature Steel A.V. Hruzevyich1,2, *D.O. Derecha2,3 1 Trypil'ska CHP, UA-08720 Ukrainka, Kyiv Region, Obukhov District, Ukraine 2 Institute of Magnetism Nat. Acad. of Sci. of Ukraine and Min. of Edu. And Sci. of Ukraine, 36b Academician Vernadsky Blvd., UA-03142 Kyiv, Ukraine 3 National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 37, Prosp. Peremohy, Kyiv, UA-03056, Ukraine e-mail: [email protected] PACS numbers: 06.20.−f, 44.90.+c, 61.05.−a, 61.50.−f, 61.72.−y, 61.82.Bg Abstract The results of research pipes diffusion hardening as an effective method for increasing the durability and reliability of power equipment are presented. The experience of commercial operation of pipes manufactured from diffusion chromized and heat-hardened steel DIN: 1.7715; WNr: 14 MoV6-3; BS: 1503-660- 440 on the heating surfaces of the supercritical pressure boilers are generalized. The possibility and effectiveness of this method on the example of capacities of Trypil'ska CHP are shown. Keywords: diffusion-hardening, steel, durability, structure, heat resistance, weldability, corrosion resistance, heat resistance. 1 1. Introduction Modern metal materials which are used in power engineering, machine building, ship-building and construction industry should provide the necessary resource and reliability of the structures. The most important characteristics that determine the reliability of the work of components and constructions are the wear indexes, long-time strength and corrosion characteristics. The corrosive characteristics and characteristics of mechanical degradation are largely dependent on the condition of the components surface. The material surface condition significantly effects on the cracking rate in the material, therefore, for the improvement of the metal components characteristics the matter of technological processing of their surface is extremely important. In order to improve the wear resistance of the components the different methods of the surface treatment are used, such as mechanical-thermal treatment, metalwork and mechanical processing, surface impregnation etc. The most effective treatment techniques are those that lead to the surface strengthening and simultaneously create in the surface layers residual compressive stresses [1 -- 4]. In this case, the resistance to the origination and propagation of crack at the surface area increase simultaneously. The strengthening also complicates the lateral movement of the surface layers, and compressive stresses prevent the crack opening and decrease the effect of tensile stresses during the component cyclic loading. The mechanical-thermal treatment combined with the doping of the surface layer is the best method that allows us to obtain such a combination of properties in the surface layers of alloys [4]. It is known [5 -- 9] that good results for surface condition improving are achieved by forming the amorphous layer on the surface during doping and heat treatment. However, for massive components with a complex profile of the surface 2 it is a complicated technological problem to obtain an amorphous state on entire surface. In the present paper we describe the technology that has significantly improved the durability and corrosion characteristics of industrial pipes (such as corrosion resistance, welding, heat resistantance). This technology fits well with the existing industrial pipe manufacturing process and allows, with a small cost, to significantly increase the operational and resource characteristics of pipes and other structures made of ferrous metals based on iron alloy. The service life of heating surfaces of the lower radiant section (LRS) of boilers with the power of 300MW and higher manufactured of steel pipes manufactured by DIN: 1.7715 (WNr: 14MoV6-3, BS: 1503-660-440) is limited to 20 thousand hours, in the area of burners and on the upper slit of the clipping tooth -- 8-12 thousand hours (composition of the steel are specified in table 1). The main reason for the pipes cracking is their low corrosion resistance in combustion products of organic fuel. Table 1. -- Composition of the steel C, % Si, % Mn, Ni, S, % P, % Cr, Mo, V, % Cu, Fe, % % % % % % 0.08 -- 0.17 -- 0.4 -- <0.3 <0.025 <0.03 0.9 -- 0.25 -- 0.15 -- <0.2 ~96 0.15 0.37 0.7 1.2 0.35 0.3 One of the effective ways to increase the corrosion resistance of pipes is the application of protective metal coatings on their surface. The most significant increase in the corrosion resistance of pipes manufactured of this steel is achieved by diffusion chromizing. High corrosion resistance of chromium coatings in synthetic slags, whose composition simulates the combustion products of organic fuels, is also confirmed [10]. 3 However, there is no information in the literature on the use in the pipes heat power engineering for the manufacturing the pipes for heating surfaces with protective chromium coatings. Probably, this is due to the insufficient technological properties of electrodeposited chromium coatings, and in the case of diffusion chromium plating -- due to the negative effect of the chromium process on the structure and properties of the parent steel. In this connection, the aim of current work is the technology's comparison of pipes manufacturing with a protective diffusion coating of chromium, the structure and properties of the parent steel of which would meet the requirements of the heating engineering. Three technological schemes of pipes manufacturing with a protective diffusion chromium coating have been studied. The first one is from chrome billets by hot moulding to the final size or with the production of tube stocks and subsequent cold rolling. The second one is the high-speed diffusion chrome plating of pipes with the finite size and their subsequent heat treatment. The third one is the ion-plasma diffusion coating at the atmospheric pressure condition. In the first case, a significant deformation of the metal at pipes production from chromium billets allows complete eliminating the effect of the high- temperature heating during chrome plating. For diffusion chrome plating of billets, the vacuum resistance furnaces of the types OKV 554AM, SLV16-128-16/14 and others with similar design can be used. The chromed pipes from billets can be obtained by pressing, for example, on the presses of Zorya-Mashproekt Co, Interpipe Nico Tube, LLC "NZEST". In the second case, when the chromium processing does not exceed 3-5 minutes, the effect of high-temperature heating is completely eliminated by 4 subsequent heat treatment. However, this requires the vacuum furnaces of a special design. In the third case, for the production of diffusion chrome plating the bulky expensive equipment is not required. Such equipment gives the mobility and possibility of coating under repair conditions. The conclusion about the corrosion resistance of chrome-plated pipes can be obtained only as a result of their testing on the thermal power boilers during the burning of various types of organic fuel. The installation of pipes in the areas with the highest boilers heating surfaces damaging has been made at the Trypil'ska CHP. The Trypil'ska CHP has six power units with the power of 300MW operating on pulverized coal and gas-oil fuels. 2. Experiment and Discussion 2.1 Hot-Pressed Chrome-Plated Pipes Manufacturing To produce the hot-pressed pipes the temporary technical specifications were developed by UkrNIIspetsstal and VNITI. According to them, the pipes manufacturing from chrome-plated welded billets of 150x270 mm diameter had to be done by pressing on the size of 32x6 mm. The steel billets of 150x270 mm and 160x280 mm diameters, were chromium plated in the vacuum resistance furnace SLV 16-128-16/14.5x64 in the FERRO ALLOY PLANT PJSC (Zaporozhye). The temperature of the chromium plating was 1400oC, the time of isothermal aging was 20 hours. The depth of the chromium plating was 1.7-2.8 mm depending on the level of carbon in the billets. The macrotemplate of the chrome plated billet is shown in Fig. 1. 5 Fig. 1. The macrotemplate of the chrome plated billet. As a depth of chromium plating the surface layer which is not etched in the 4% alcohol solution of HN03 was taken. The concentration of chromium on the surface was 60-70 %. During the hot-pressed pipes manufacturing at the VNITI experimental plant, the dimensions of the billets and the heating technique before pressing were specified. Also, the methods for chemical treatment of pipes in order to remove lubricant after pressing and boiler scale after heat treatment (including sandblasting of the inner surface) were developed. 6 In order to determine the rational method of metal heating before pressing, three heating assemblies were used: horizontal and vertical inductors of high- frequency furnace OKV-665 and the chamber silicon carbide furnace OKV-210A. The heating of the metal was carried out to the previously established temperature of 1210-1215 oC. The analysis showed that the worst quality of the outer surface of pipes is obtained after pressing the billets heated in a high-frequency furnace with a horizontal inductor. This is due to scoring on the surface formed when the billet moves during heating on the guide plates. The formed scuffs in the process of molding grows into scabs and cavities. For pipes pressed from billets heated in a vertical inductor or chamber furnace, the surface quality was better than that one in pressed billets heated in a horizontal inductor. The difference in the metal heating time in a chamber furnace or in the inductor did not affect on the pipes quality. In terms of productivity, the induction heating should be preferred. Due to the fact that the steam superheating pipes were manufactured by the cold deformation technique, it was advisable to produce chrome-plated pipes in the same way. The cold deformation was carried out by single-pass and two-pass rolling with 1 hour intermediate heating at the temperature of 740oC. For the intermediate heat treatment, it was chosen the temperature which, according to VNITI, provides a minimum strain resistance of the steel and the best resistance to fragile breakings. It was steel with 25% of chromium. The pipes of a final size were heat treated. This treatment included the normalization at the temperature of 950-980oC and tempering in the air at 720-750o C for 1 hour. 7 Under the cold strain, the pipes become hardened. In addition, the degree of hardening is greater, the greater the degree of deformation. The grains of the basis metal are stretched along the direction of deformation. Normalization, followed by tempering, results in a uniaxial grains structure. In the case of the chrome pipes, after normalization and subsequent tempering, bainite was observed in the structure in more quantities than in the hot- pressed tubes of the same size (32×6). Apparently, this is connected with a cold strain that reduces the critical points during subsequent heating. However, this did not affect the mechanical properties of the pipes. After heat treatment, the hardening of the pipes is completely removed, the initial properties of the pressed pipe are restored, and the yield point slightly increases. The latter is related with the age-hardening of steel accompanied by the carbides precipitation and fragmentation of the solid solution. The mechanical properties of the pipes after the heat treatment are above the specifications and correspond to the level of hot-pressed pipes. The analysis of the macro- and microstructure of the chrome-plated pipes after pressing, cold deformation, subsequent heat treatment shows that the chromium layer decreases during the production of pipes but remains in all cases, there is a transition zone behind it and the basis metal behind the transition zone. On the pipes obtained by pressing and cold deformation, the microhardness of the protective layer decreases by the depth from 200 to 134 kgf/mm2 and from 200 to 114 kgf/mm2, respectively. The Vickers hardness of the basis metal is 200-230 kg/mm2. Thus, the method of pipes producing does not affect their hardness after heat treatment. 2.2 Production of the developed batch of chromated pipes with diameter of 32 and 42 mm at the UMT-4 installation 8 At present, the pipes with chromium protective diffusion coating are not produced in our country. This is explained by the lack of high-performance technology of diffusion chrome plating, as well as by significant material softening due to prolonged exposure at high temperature. Thus, for example, when chromizing in a backfill in a vacuum, the total duration of the process for obtaining a layer with the thickness of 0.2 mm is 25-50 hours, at the temperature of 1150oС, the strength properties are reduced in 1.5-2 times. Under contact gas chrome plating, the duration of the process is 35-56 hours, exposure is 25 hours, at the temperature of 950-970oС. The process of pipes chrome plating can be reduced to 2-2.5 hours with non-contact gas chrome plating by the container with pipes purging with chrome chloride. However, the authors note that in this case there is considerable unevenness of the coating along the tube length. The above circumstances did not allow the use of diffusion chromium processes to protect the pipes if high strength requirements are imposed on them. Taking into account the exposed, in UkrNIIspetsstal the pilot plant (UMT-4) was designed and manufactured to study the continuous chrome plating technology for ready pipes, which allows to apply chromium diffusion and condensate coatings to pipes with a diameter of 25-50mm and a length of up to 6 m. The general view of the plant is shown in Fig. 2. The technique developed by UkrNIIspetsstal together with other organizations for the production of pipes from chrome-plated billets makes it possible to obtain pipes that are not inferior in mechanical properties to pipes without coating. The developed batch of hot-pressed chrome-plated pipes after a detailed inspection was mounted for testing at 300MW blocks of the Trypil'ska CHP. 9 Fig. 2. -- UMT-4 installation 2.3 Pipes chromium plating The developed batch of tubes for diffusion chrome plating was produced at the Nikopol Yuzhnotrubnyi Plant according to technical requirments. The dimensions of pipes are 42 and 32 mm, the wall thickness is 6 mm, length is 5-9 m. According to the agreement with UTP, the pipes for chrome plating were supplied without surface lubrication. Inspection of the pipes showed that the external surface contains a significant number of defects in the form of concavities and burrs with traces of oxide scale, the nature and size of which are allowed by the technical requirements. The presence of such defects hinders the quality of chromium plating. Therefore, the surface of the pipe before chrome plating was 10 chipped manually or by means of grinding machine of the calibration operating department of the Dneprospetsstal plant. The metallographic examination of the pipes showed that there is a decarbonizated metal layer from the outer and inner surfaces, the value of which varies from 0 to 200-800 μm. The pipes were chrome plated in the through-type vacuum unit UMT-4 for 3 pcs. simultaneously. The ferrochrome with composition specified in Table 2 was used as a metallizer. Cr, % 65 Table 2. -- Composition of ferrochrome C, % 0.1 Si, % 0.5 P, % 0.02 S, % 0.02 Al, % 0.2 3. Metallographic studies and mechanical tests of the chromium-plated pipes After chrome plating, a sample was cut from both ends of each third pipe to control the depth of the diffusion layer and the chromium content on the surface. Selectively, on some pipes, the character of the distribution of chromium and the microhardness along the depth of the diffusion layer were analyzed. The depth of the chromium plating was controlled on the cross-sections. For the depth of the diffusion layer of chromium, a white metal layer on the surface that is not etched in the 4% solution of nitric acid was adopted. The measurements have shown that the depth of the diffusion layer of chromium on the pipes of the tested batches varies within the limits of 0.1-0.2 mm and is practically independent on the fluctuations in the carbon content within the steel grade. This, apparently, explained by the presence of a decarbonizated layer on the surface of the original pipes. 11 The structure of the chromium diffusion layer, γ, is solid solution of chromium in iron with a smoothly varying concentration of chromium and microhardness over the depth of the layer (Fig. 3). Fig. 3. -- Change in the concentration of chromium (curve 1) and microhardness (curve 2) in the depth of the protective layer on the pipes. Two characteristic curves for the distribution of chromium and microhardness in the depth of the layer with an inflection near and without the surface are noted. In the first case, there is a sputtered layer of ferrochrome on the surface, in the second one -- the layer is absent. The sputtering part of the layer appears after etching in the Marbley reagent and has a fine-grained columnar structure. Usually the deposited layer is present 12 approximately in 0.6-0.7 of the pipe perimeter. It appears in the initial period of chromium on the part of the pipe that faces to the metallizer. The formation of the sputtered layer is explained by the prevalence of the rate of evaporation of the metal over the diffusion rate due to weak heating of the pipe at the moment of its appearance above the metallizer. On the part of the pipe that has time to warm up, only the diffusion layer is formed (Fig. 4a). The concentration of chromium on the surface of the pipes is 35-50%. Smaller values refer to the diffusion part of the layer, large ones -- to sputtered. The fact that the chromium concentration on the sputtered part is less than in ferrochrome, and also that the concentration gradient in the deposited layer is observed, indicates its formation in the initial chromium period. The boundary of the transition from the deposited part of the layer and the diffusion layer depends on the quality of the preparation of the surface for chromium plating. With careful preparation, it is practically not detected during etching of thin sections in nitric acid. In other cases, the interface is visible in the form of a chain of pores along the boundary (Fig. 4b). Investigations of diffusion coatings with a sputtered part of the layer on the surface show that just in this part a considerable number of pores, carbonitride precipitates and sigma phases are observed. The presence of brittle structural components and pores in the sprayed part of the layer makes it undesirable. 13 a b Fig. 4. -- Microstructure of chrome plated layer on pipes, UMT-4, ×200 magnification: a) without a sputtered layer; b) with a sputtered layer. Only the diffusion layer can be formed on the surface of the pipes either by raising the surface temperature of the pipes in the initial chromium period or by reducing the intensity of evaporation of ferrochromium. Diffusion chrome plating of pipes at the industrial plant UMTP-11 excluded the formation of sputtered layers and allowed to regulate the concentration of chromium on the surface in any limits. As studies have shown, the transition part of the diffusion layer (α → γ transition in the chromium process), as a rule, does not contain characteristic for the chromium-plating steels interlayer of a metal with the pearlite structure. The microhardness curve in this region has a smooth character. This feature is explained by the presence of a significant decarburized layer on the surface, as 14 well as by the short duration (3-5 min) of diffusion chrome plating. As a result, carbon redistribution processes do not occur in the diffusion layer. The investigations of the structure of the basis metal indicate that, as a result of heating during the chromium plating, a coarse-grained structure consisting of 90% of bainite with a ferrite layer along the boundaries of the former austenitic grains is formed (Fig. 5a). a b Fig. 5. -- Microstructure of the basis metal on the pipes (magnification ×100): a) after chrome plating; b) after chrome plating and heat treatment. The chromium plated pipes were heat treated according to the standard mode: normalization at 950-980oС, 30 minutes exposure; drawing-back at 720- 750oC, 1 hour exposure. The structure of the basic metal becomes a fine-grained ferrite-pearlite with a small amount of drawed-back bainite (Fig.5b). 15 The comparison of the microstructure of chrome and non-chrome pipes of the same melting after normalization with tempering under identical conditions shows that the chromium-plated pipe has a smaller structure and the amount of tempered bainite in it is smaller. This affected on the level of mechanical properties of pipes -- the strength characteristics are lower, and the plastic ones are higher in chrome pipes than in non-chromated pipes, although they meet the technical requirements (Table 3.). Table 3. -- Mechanical characteristics of pipes Material Melting lot Dimensions Mechanical characteristics number of pipes, σB σT δ5 mm resistance to flow limit elongation rupture at failure Source steel 5830 1250 Ø42х6 5849 1218 ÷ Chrome- 5830 1250 Ø42х6 platted steel Technical - - requirements - 57 54 52 45 45 41 41 28 28 26 26 21 Note: Chromium-plated pipes had satisfactory flattening with the allowed technical requirements -- for this steel the distance between the flattened surfaces is 24 mm and less. 16 4. Investigation of the technological properties of chromium-plated pipes As it was noted above, the structure of the basis metal of the chrome pipes after heat treatment, irrespectively to the method of production (hot pressed or chrome plated on the UMT-4 plant), meets the requirements. The diffusion chrome plating does not also have a significant effect on the mechanical properties of pipes, slightly reducing the strength and increasing the plastic properties of the metal. Despite the high concentration of chromium in the surface diffusion layer, the latter has a sufficiently high plasticity and does not break down when testing the pipes for flattening and bending. Other requirements for chrome pipes are high corrosion resistance in combustion products of organic fuel and good weldability. 4.1. Corrosion resistance The main purpose of the diffusion layer of chromium on the boiler pipes is to protect it from corrosion in the combustion products of organic fuel. The available information on the composition of the gas phase and fuel ash deposits on the surface of the pipes, obtained as a result of the analysis at temperatures different from that one of the pipes operation, makes it difficult to assess the responsibility of various elements or compounds for the course of corrosion processes. Based on the analysis of corrosion products, there are two main types of corrosion: sulphide and the so-called "vanadium". In sulphurous corrosion, the main cause of the destruction of the pipes surfaces is the formation of complex sulfates of K3Fe(SO4)3; Na3Fe(SO4)3 with a 17 melting point about 600oC which is close to the operating temperature of the boiler pipes. For complex alloy steels, sulfates can also be formed with lower melting points. In the case of vanadium corrosion, low-melting compounds are also formed with a melting point close to the melting temperature of sulfates. The presence of the diffusion layer of chromium on the pipe surface excludes the possibility low-melting compounds formation at the temperature of the boiler pipes operation. Considering that the existing methods of corrosion resistance testing in synthetic slags cannot fully characterize the serviceability of pipes in combustion products of organic fuel, the research of pipes was carried out directly on boilers of thermal power units. The developed batch of pipes was installed in the lower radiation part of the TGMP-314 boiler. The fuel of the boiler was gas with fuel oil. After 21 thousand hours of the boiler operation, a partial cutting of pipes and a study of their condition were carried out. It has been established that the appearance and diameter of the chrome pipes have not changed. In ash deposits on the adjacent side and pipes, chromium was not detected by spectral analysis. These facts indicate that there is no general corrosion of chrome plated pipes. Metallographic examination of pipes that have worked 21 thousand hours showed that the continuous diffusion layer of chromium was preserved, but the thickness of the layer slightly increased in comparison with the initial state, that is explained by the processes of secondary diffusion of chromium into the base of the pipes. Analyzing the results of corrosion tests of chromium-plated pipes on a TGMP-314 boiler, it can be assumed that the service life of chrome-plated pipes 18 will be determined mainly by the kinetics of the secondary chromium diffusion process in the operation of pipes. An approximate estimation of the surface of chrome-plated pipes destruction beginning can be made using the following equation: (1) where Co and Cm are the initial and final chromium concentrations on the surface of the pipes, correspondingly; δ is the depth of the diffusion layer, D is the diffusion coefficient of chromium, and τ is the time of decrease in chromium concentration on the surface of pipes from Co to Cm. It follows from Eq. (1) that as the initial concentration of chromium on the surface and the depth of the layer increase, the time of its operation increases. Figure 6 shows curves of the change in chromium concentration on time for different thicknesses of the diffusion layer. If in the capacity of the time of active oxidation beginning we take the time when the chromium concentration on the surface drops to 12%, then for pipes with the initial chromium concentration of 25% and the diffusion layer thickness of 0.05, 0.10 and 0.15 mm, respectively, it will be equal to 15, 60 and 120 hours. 19 ДСCtgот41222 Fig. 6. Chromium concentration change in the layer during operation. Taking into account the cyclic nature of the pipe loading, wide range of operating temperatures in the zone of lower radiant section that include the brittle transition temperature of chromium steels, probability of short-term overheating of pipes during operation, presence of welds that differs in structure and composition from the diffusion layer, other types of protective layer destruction can be expected along with general corrosion. The investigations of the developed batch of chrome-plated pipes at Trypil'ska CHP allow to answer these and other questions related to the resistance of chrome pipes. 20 4.2 Heat resistant properties, welding Two variants of welding of chrome plated pipes were investigated. The first one is the welding of the entire weld by ZL-39 (ESAB: OK 76.18) electrodes at butt-welding of chromium-plated pipes with non-chromated ones. The second one is welding with ZL -39 electrodes with overlapping of the top layer by ZT-15 (ESAB: ОК 61.85; OK 61.80; OK 61.86) electrodes at docking of the chrome plated pipes with each other. The welding was performed in accordance with guidance document "Welding, heat treatment and control of pipe systems of boilers and pipelines in the installation and repair of power plant equipment". Before welding, the edges of pipes were cut at the angles of 40-45 degrees. The diameters of the wire of electrodes are 2.5 mm for ZL-39 and 3 mm for ZT-15. At visual control, the welds are in satisfactory condition. The cracks, pores, shells, undercuts and other external defects were not observed. The internal defects in butt welds by ultrasonic inspection were not detected. The presence of a diffusion layer of chromium on the pipes surface did not prevent the ultrasonic inspection of welded joints. The microstructure of the metal in the weld zone and the joints is satisfactory in both cases. The damage of the diffusion layer in the near-weld zone was not observed. Conclusions The diffusion chrome plating of pipes allows to increase their corrosion resistance in combustion products of organic fuel (pulverized coal and gas-oil fuel) by several times. The durable strength of 100'000 hours for chromium-plated pipes is on the same level as for non-chrome plated pipes. 21 References [1] G.H. Farrahi, H. Ghadbeigi, An investigation into the effect of various surface treatments on fatigue life of a tool steel, Journal of Materials Processing Technology 174 (2006) 318 -- 324. [2] C. Swett, Outpatient phenothiazine use and bone marrow depression. A report from the drug epidemiology unit and the Boston collaborative drug surveillance program, Archives of general psychiatry 32 (1975) 1416 -- 1418. [3] Z.L. Zhang, T. Bell, Structure and Corrosion Resistance of Plasma Nitrided Stainless Steel, Surface Engineering 1 (2013) 131 -- 136. [4] G.E. Totten, Steel heat treatment: Metallurgy and technologies / George E. Totten, editor, Taylor & Francis, Boca Raton, FL, 2007. [5] C. Borchers, T. Al-Kassab, S. Goto, R. Kirchheim, Partially amorphous nanocomposite obtained from heavily deformed pearlitic steel, Materials Science and Engineering: A 502 (2009) 131 -- 138. [6] A.V. Byeli, O.V. Lobodaeva, S.K. Shykh, V.A. Kukareko, Solid-state amorphization of a tool steel by high-current-density, low-energy nitrogen ion implantation, Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 103 (1995) 533 -- 536. [7] Daniel J. Branagan, Joseph V. Burch, Methods of forming steel, US6258185B1 (1999), US. [8] K. Hashimoto, N. Kumagai, H. Yoshioka, J.H. Kim, E. Akiyama, H. Habazaki, S. Mrowec, A. Kawashima, K. Asami, Corrosion-resistant amorphous surface alloys, Corrosion Science 35 (1993) 363 -- 370. [9] Y.N. Petrov, V.G. Gavriljuk, H. Berns, F. Schmalt, Surface structure of stainless and Hadfield steel after impact wear, Wear 260 (2006) 687 -- 691. 22 [10] Shang‐Hsiu Lee, High-Temperature Corrosion Phenomena in Waste-to-Energy Boilers. Submitted in partial fulfilment of the , USA, Columbia, 2009. 23
1809.05248	4	1809	2019-01-29T04:03:51	Manipulate elastic waves with conventional isotropic materials	[ "physics.app-ph" ]	Transformation methods have stimulated many interesting applications of manipulating electromagnetic and acoustic waves by using metamaterials, such as super-lens imaging and cloaking. These successes are mainly due to the form-invariant property of the Maxwell equations and acoustic equations. However, the similar progress in manipulating elastic waves is very slow, because the elastodynamic equations are not form-invariant. Here we show that the expression of the elastodynamic potential energy can almost retain its form after conformal mapping, if the longitudinal wave velocity is much larger than the transverse wave velocity, or if the wavelength can be shortened by converting the waves into surface modes. Based on these findings, it is possible to design and fabricate novel devices with ease to manipulate elastic waves at will. One example presented in this paper is an efficient vibration isolator, which contain a 180-degree wave bender made of conventional rubbers. Compared with a conventional isolator of the same shape, similar static support stiffness and smaller damping ratio, this isolator can further reduce wave transmissions by up to 39.9dB in the range of 483 to 1800 Hz in the experiment.	physics.app-ph	physics	Manipulate elastic waves with conventional isotropic materials Hexuan Gao, Zhihai Xiang* Department of Engineering Mechanics, Tsinghua University, Beijing 100084, China Abstract: Transformation methods have stimulated many interesting applications of manipulating electromagnetic and acoustic waves by using metamaterials, such as super-lens imaging and cloaking. These successes are mainly due to the form-invariant property of the Maxwell equations and acoustic equations. However, the similar progress in manipulating elastic waves is very slow, because the elastodynamic equations are not form-invariant. Here we show that the expression of the elastodynamic potential energy can almost retain its form after conformal mapping, if the longitudinal wave velocity is much larger than the transverse wave velocity, or if the wavelength can be shortened by converting the waves into surface modes. Based on these findings, it is possible to design and fabricate novel devices with ease to manipulate elastic waves at will. One example presented in this paper is an efficient vibration isolator, which contain a 180-degree wave bender made of conventional rubbers. Compared with a conventional isolator of the same shape, similar static support stiffness and smaller damping ratio, this isolator can further reduce wave transmissions by up to 39.9dB in the range of 483 to 1800 Hz in the experiment. I. INTRODUCTION A popular paradigm to manipulate waves is using transformation method, which was originated in manipulating electromagnetic wave [1,2], and has been extended to other types of waves, such as acoustic wave [3,4], surface water wave [5], and even matter wave [6]. However, * Correspondence to: [email protected]. 1 in contrast to the great success of transformation optics [7-9] and transformation acoustics [10- 12], the progress of transformation elastics is rather slow. This is because the classical elastodynamic equations are generally not form-invariant [13], which is the essential prerequisite of transformation methods [1]. To circumvent this difficulty, people have tried to design elastic metamaterials in some special cases, such as the Cosserat type material [14,15], high-frequency approximations [16,17], bending waves in thin plates [18-21], and utilizing the pre-stresses [22- 24]. Unfortunately, it is still very difficult to use either of these special solutions to manipulate broadband low-frequency elastic waves of complex modes [25]. Because of this thorny issue of lack of invariance for elastodynamic equations [9], it is difficult to expect perfect metamaterials to manipulate elastic waves of complex modes. Instead, we can design almost perfect elastic metamaterials based on an idea of minimizing the extra term in the expression of elastic potential energy after transformation. This can be generally achieved by using conformal mapping, requiring the longitudinal wave velocity being much larger than the transverse wave velocity, and converting the wave into surface modes of slow velocities. Since the conformal mapping [2, 17, 26] is adopted in the design, the metamaterial can be easily fabricated with conventional isotropic materials free of local resonance, and has good performance in broad frequency bandwidth. A particular example of the elastic metamaterial designed in this way will be reported here is an elastic wave isolator, which is composed of an aluminum alloy shell and a 180-degree elastic wave bender made of two kinds of conventional rubbers. This paper is organized as follows. In Section II, we present the theoretical derivation of the effective properties of the elastic metamaterial and an index η, which measures the extent of form- invariance of potential energy expression after conformal mapping. Based on this theory, we design a vibration isolator in Section III. Numerical simulations are also presented to demonstrate 2 its performances. Then, experimental verifications are detailed in Section IV. Finally, summaries are given in Section V. II. THEORETICAL BASIS Transformation methods used to design metamaterials [1, 2] usually start from a homogeneous and isotropic material in the virtual space. The geometry of this material in the virtual space is transformed into the desired geometry of a metamaterial in the physical space by using certain coordinate mapping. In this way, a point x in the virtual space is transformed into a point x  x x in the physical space. At the same time, field variables in the virtual space is also   transformed into the physical space with a given gauge. Consequently, by using the change of variables, the governing equations in the virtual space are transformed into the physical space. If the transformed governing equations have the same forms as those in the virtual space, we can easily obtain the effective inhomogeneous material properties of the metamaterial in the physical space by comparing the corresponding terms in these two sets of equations. The selection of the gauge connecting the field variables before and after transformation could be arbitrary [15], and has nothing to do with the form-invariance of the governing equations [24]. However, this gauge is usually chosen as the deformation gradient x / x [1, 2, 13], so that the deformed grid in physical space shows the wave trajectories [27] and the transformed elasticity tensor is symmetric [15]. Milton et al. has proved that the form-invariance of classic elastodynamic equations are not form-invariant [13]. This is also true if using conformal mappings [28]. All existing work on discussing of form-invariance of elastodynamic equations involve the complex analysis of the gradient of elasticity tensor, because the metamaterial is inhomogeneous. To avoid this 3 complexity, we focus on the form-invariance of the elastic potential energy (integral invariant) instead of the elastodynamic equations (differential invariants) [29] in this paper. Neglecting the body force, the elastic potential energy in the virtual space of volume V is    V 1 2   ij kl D ijkl where D ijkl   kl ij  G    jk  ik jl il d V   V  u u V d  i i   u u V d  i i   V  S T T u S d i i , (1)  is the elasticity tensor with Lamé parameters  and G; ij is the Kronecker delta; ij is the strain tensor; iu is the displacement vector; T i n ij j is the given traction vector on boundary TS with normal vector jn ; ij is the stress tensor on boundary TS ; is mass density; is the damping factor; and the overhead dot denotes the derivative with respect to time. Here we use the Einstein's notation of index summation. Since conformal mapping will be used in the following, all indices are taken the value of 1 or 2. We also use the deformation gradient as the gauge to connect the field variables before and after transformation: u i u  j x  j x  i and    kl  J ij x  k x   i x  l x   j , (2) where J  det  x  /j  x  i  , and all values with superscript prime are defined in the physical space. In addition, it is well known that the conformal mapping used for the geometrical transformation has the following properties: x   k x  i x   l x  i  J  kl and x 2  k x x   i i  0 . (3) Applying the transformation specified in Eqs. (2) and (3) to Eq. (1) and noticing d V  J V d 1  , n S d j  1  J  x   i /   x n S d  j i  , T i n   ij  j and     ij u i / x u    j j /  x i  / 2 , we can obtain the transformed elastic potential energy in the physical space: 4     V  1 2    J   jj  ii + JG    ij ij  Ex V d       V   u u V d    i i    u u V d    i i    V   S  T T u S d  i i   , (4) where the extra term is Ex  G J    u u   i j x 2   i x x   k l 2 x   j x x   k l  2 u    i rs x 2   i x x   k l x  x   r k x   s x  l    . (5) So, the expression of elastic potential energy is not form-invariant, either. This coincides with existing findings [13, 28]. We cannot obtain effective material properties of the metamaterial in the physical space by directly comparing Eq. (4) with Eq. (1), unless the contribution of extra term Ex to the strain energy can be neglected. The conditions of ignoring Ex can be found by conducting the following dimensional analysis. Under harmonic excitation of magnitude a and wavelength L at frequency f, the    can be approximated as displacement at location Re exp i   κ z    f t 2 ,x x  1 z  iu  a   ,    2     where κ is the wave number vector of magnitude κ 2 / L  . Since     ij u i / x u    j j /  x i  / 2 , the magnitude of the strain in the physical space is about  ij  κ  iu  . With all these results and noticing that the velocities of longitudinal wave and transverse wave are pC    G  2    /   and sC   G   /  , respectively, we can calculate the ratio of the magnitude of traditional strain energy terms over the magnitude of the extra term: 1 2 J   JG +    ij ij  ii   jj Ex  x 2   i x x   k l     J C   p L C   s 2    x 2  j x x   k l  2    4  L  . (6) x   s x  l     x  x   r k If 1 , the potential energy is almost form-invariant: 5     V  1 2    J   jj  ii + JG    ij ij    d V    u u V d    i i    V   u u V d    i i    V   S  T T u S d  i i   . (7) In this case, the material properties of the metamaterial in the physical space can be obtained by comparing Eq. (7) with Eq. (1) as: ,    J  G JG   , .       ,   (8) III. REALIZING A VIBRATION ISOLATOR A. Effective material properties of the wave bender With the theory presented in Section II, it is possible to design many interesting elastic metamaterials. Here just gives a simple example, a wave bender, for demonstration. As Fig. 1 shows, the design starts from a rectangular virtual space filled with homogeneous and isotropic material. This virtual space (described by the point Z  x 1  x 2i of the complex plane) is transformed into a semi-circular ring shaped physical space (described by the point z  x  1  i x  2 ) with the conformal mapping z Z e . In this case, we have J  x 2 2 x 2 1  2   2 2 r  , (9)  r x  x  k x   s x  l  J , 2 x  i X X   k  r 2 2  l , (10) where 1  m to make J dimensionless. Consequently, Eq. (6) becomes   2 L  r 2 2  2 2   rL  2    C   p  C  s . (11) To ensure a sufficiently large  , we set  1000 kg/m3,  673.91MPa and G  0.07MPa . According to Eq. (8), material properties of this semi-circular ring in the physical space are: 6    2 2 r  ,  G   2 2 r  G ,         , . (12) Since C  p      2 G   /    C r p /  and C  s  G  /    C r s /  , it is easy to verify that a large  can be easily achieved. In addition, since pC and sC are proportional to r, elastic waves will bend inside this metamaterial. This wave bending effect only depends on the distribution pattern of pC and sC regardless of their specific values, i.e., independent of pC and sC . FIG. 1. Snap shots of the total displacement (normalized to the amplitude of incident wave) of elastic waves simulated by COMSOL Multiphysics, where u n   0 and u n     0 denote the normal constraints on the boundaries in the virtual space and the physical space, respectively. (a) The virtual space with normal constraints on the upper and the lower boundaries (f = 10000 Hz). (b) The virtual space with normal constraint only on the upper boundary (f = 1000 Hz). (c) The physical space transformed from (a). (d) The physical space transformed from (b). (e) The modified total displacement field J u of (c). (f) The modified displacement field J u of (d). 7 With these material properties, we conduct numerical simulations with pure sinusoidal incident longitudinal waves to check the wave bending effect. In Fig. 1(a), since the upper and lower boundaries are normally constrained, only pure longitudinal waves propagate in the virtual space. As Fig. 1(c) shows, the wave does bend in the corresponding physical space with the displacement magnitude inversely proportional to r. This coincides with the relation u i u  j  x j /  X i . Since the incident waves pass along the tangent direction of the outer boundary rather than transmitting through its normal direction, the supported object (denoted by a little square on the top) will not be disturbed. In this way, this wave bender can be used as a novel vibration isolator. In addition, as implied by Eq. (7), the distribution of the modified total displacement J u can represent the distribution of strain energy density. This means Fig. 1(d) gives the distribution of strain energy inside the wave bender. The η in Eq. (11) is a crucial index that measures the performance of this wave bender. We can increase η by using a material with large ratio of C C /  p  s  G /    2 . Since E       / 1       1 2      and  G E     / 2 1        ( E is the Young's modulus and  is the  Poisson's ratio), the ratio of C C /  p  s  2 / 1 2         can be maximized when 2   . 0.5 Therefore, it is a natural choice to use rubbers to fabricate this isolator. Eq. (11) also implies that we should decrease the wavelength L. For this purpose, it is preferable to convert the elastic waves into surface modes, e.g. the wavelength of Rayleigh wave is slightly smaller than the transverse wave [30] and much smaller than the longitudinal wave at the same frequency. The wave mode conversion can be achieved by releasing the normal constraint on the boundary, where most waves concentrate near the free surface in Rayleigh mode (see Fig. 1(b), (d) and (f)). 8 FIG. 2. Snap shots of the modified total displacement J u (normalized to the amplitude of incident wave) for different  simulated by COMSOL Multiphysics. To check the impact of  on the performance of this wave bender, we compare the wave bending effects for different  in Fig. 2. In the simulation, C C /p  is in the range of 9.8 to 31.0, s while the wave frequency f is adjusted to keep a constant wavelength of L r  / 2 in order to ensure clear comparisons. In this way,  is in the range of 10 to 100 according to Eq. (11). The simulation results indicate approximately that good bending effects can be achieved when 80 . This means when the extra term is two orders smaller than other terms, the elastic wave equations are nearly form-invariant. B. Fabricating the vibration isolator The effective material properties of this wave bender given in Eq. (12) can be easily realized, because they are isotropic. A simple way is mixing two kinds of conventional rubbers inside a representative cell (see Fig. 3). The material properties of two rubbers adopted here are as follows: 9 Natural Rubber (NR) NRE  2.61 MPa, NR  1003 kg/m3, p NRC    1592 m/s; Silicone Rubber SRE  0.11 MPa, SR  1092 kg/m3, p SRC    821 m/s. Since the small difference of densities can be neglected compared with the huge difference of stiffness for these two kinds of rubbers, we can focus on the realizing of  and G at different r. This can be achieved by adjusting the volume fraction ratio of NR inside a cell at different r (see Fig. 3(c)). FIG. 3. (a) Effective wave velocities obtained from finite element models with different volume fractions of NR, which is solved by COMSOL Multiphysics. (b) Model of the rectangular domain containing 100 cells with the incident wave from left side, and the view of one cell composed of two kinds of rubbers. (c) The fan cell and the distribution of effective wave velocities pC and sC . (d) The vibration isolator with inside wave bender and outside aluminum shell. 10 The relationships between effective wave velocities and the volume fraction of NR and are obtained by numerical simulations solved by COMSOL Multiphysics. A rectangular domain containing 100 cells is used in the simulation in Fig. 3(b). Each cell is 10 5 mm and is consisted of two different kinds of rubbers mentioned above. The volume fraction of NR is defined as /a b , and it varies from 0.1 to 0.95. For the simulation using longitudinal incident waves, the upper and lower boundaries are normally constrained. While for the simulation using shear waves, tangential displacements are constrained on the upper and lower boundaries. The obtained relationships between effective wave velocities and the volume fraction of NR are plotted in Fig. 3 (a), which indicates that wave velocities (especially the shear velocity) are almost proportional to the volume fraction of NR in the range of 0.0 to 0.8. Based on Fig. 3 (a), we can design the internal structures of the wave bender with an inner radius of 0.04 m and an outer radius of 0.1 m (see Fig. 3(c)). To facilitate load bearing, the thickness of this bender is set to be 0.04 m. 18 evenly distributed fan cells are in charge of wave bending. Inside each fan cell, there are 13 pair of points on the two interfaces between NR and SR and are evenly distributed along the radius. The circumferential distance of each pair of points determines the volume fraction of NR at position r. This distance is determined by the shear velocity curve in Fig. 3 (a) to guarantee that sC is strictly proportional to r. Applying the curve fitting through these 13 pair of points, obtains the two interfaces between NR and SR. As Fig. 3(c) shows, pC  1390m/s and sC  15.5m/s on the outer boundary. Referring to Fig. 2, if we require 80 to ensure the bending performance, L on the outer boundary should be less than 2.74 m according to Eq. (11). This implies that the effective bending of longitudinal waves on the outer boundary approximately starts from the frequency of 507 Hz. 11 To ensure the integrity of this wave bender, there remains two margins of SR in 5 mm thickness on the inner and outer boundary (see Fig. 3(d)). This does not affect its vibration isolation performance because waves have already been bent inside the wave bender. The final assembly of the vibration isolator has a 6061 aluminum alloy shell in 5mm thickness on the outer boundary of this wave bender. Since the acoustic impedance of the aluminum shell is much larger than that of the rubber, the shell can be used as normal constraint required on the outer boundary of the wave bender (see Fig. 1(d)). In addition, the silicone oil with a viscosity of 350 cs is used to lubricate the interface between the shell and the rubber, so that the shear force on the interface can be reduced. IV. EXPERIMENTS A. Basic properties of specimens To test the performance of the isolator fabricated in Section III (denoted as isolator A), we compare it with another isolator of the same shape (denoted as isolator B). Both isolators have the same aluminum shell, but in contrast to isolator A, isolator B replaces the wave bender (see Fig. 4(a)) with a pure NR of E  1.12 MPa and  1072 kg/m 3 (see Fig. 4(b)). A layer of SR is attached at the bottom of the NR in isolator B, so that the measured static support stiffnesses of these two isolators are almost the same (71428 N/m for isolator A and 76923 N/m for isolator B). Since the damping has great impact on vibration isolation performance, we also measure the damping ratio by using the free vibration test. As Fig. 4(a) and (b) show, low frequency pulses indicated by P are applied on the top of the wave bender and the pure NR with a rubber hammer. The damping ratio  and the resonance frequency nf can be extracted from the signals in time domain (see Fig. 4(d)) as n Af    76 Hz , n Bf    85 Hz , A  0.1039 and B  0.1013 . The 12 conclusion of B   A can also be confirmed in Fig. 4(e), because the peak at n Af   is lower than that at n Bf   . The result of free vibration test for the aluminum shell is also presented in Fig. 4 (e), in which we can find two dominant nf at 162 Hz and 934 Hz, which coincide with the first and the third nf simulated by COMSOL Multiphysics (Fig. 4 (f) and Fig. 4 (h)). FIG. 4. (a) Free vibration test of the wave bender. (b) Free vibration test of the pure rubber. (c) Free vibration test of the aluminum shell. (d) Acceleration signals on the top of the wave bender (A) and the pure rubber (B), normalized to the maximum amplitude of the pulse P. (e) Power spectrum density of the acceleration signals. (f), (g), (h) First three vibration modes of the aluminum shell solved by the COMSOL Multiphysics. 13 B. Simulation results Before the real testing, the performances of these two isolators are simulated by using the COMSOL Multiphysics. Since it is very difficult to accurately determine the damping inside the complex assembly of these isolators, we use the same damping ratio of 0.1 in the simulations. "The Thin Elastic Layer boundary condition" in COMSOL Multiphysics with normal stiffness of nk   1 10 N/m 13 2 and tangent stiffness of tk  0 N/m 2 are used on the interface between the shell and rubber. FIG. 5. (a) Snap shot of the total displacement field (normalized to the amplitude of incident wave) of isolator A at 1000Hz. (b) Snap shot of the total displacement field (normalized to the amplitude of incident wave) of isolator B at 1000Hz. (c) Transfer functions of isolator A and isolator B. As Fig. 5(a) shows, most of the incident waves are converted into the Rayleigh mode and bent along the inner boundary of isolator A, so that there is just little disturbance on the aluminum shell. While as Fig. 5(b) shows, large fraction of waves directly transmit through the aluminum shell of isolator B, so that it could have poorer isolation performance than that of isolator A. These observations can be further confirmed in Fig. 5(c), where the transfer function of isolator A has 14 the maximum reduction of 25.9 dB at 853 Hz. The transfer function is defined as  T f   20log 10 u top    f  / u input  f   , where  u input  f  and topu  f  are the amplitudes of incident waves and the top of isolators at frequency f, respectively. C. Experimental results FIG. 6. The setup of shaking table test. (a) Front view. (b) Top view. As Fig. 6 shows, shaking table test with sweep frequency input signals is firstly conducted to test the real performances of these two isolators. These isolators are placed closely on an aluminum base mounted on the shaking table. An accelerator is fixed in the center of the aluminum base to measure the input signal generated by the shaking table. Other two identical accelerators are fixed on the top of isolator A and isolator B, respectively. The recorded accelerations are plotted in Fig. 7(a), which shows the obvious reduction of amplitude by using isolator A. More detailed comparisons are given in Fig. 7(c). The transmissibility is defined as  H f   10log 10 X   T  f  / X I  f   , where  TX denotes the power spectrum density of the acceleration measured on the top of isolator A or B, and IX denotes the power spectrum density of the input acceleration measured on the aluminum plate. Since isolator A has little bit larger damping ratio than that of isolator B as we measured in Fig. 4, its transmissibility around resonance is smaller than that isolator B. According to the traditional passive isolation theory [31], isolator A should be less effective than isolator B at frequencies above the resonance. However, as Fig. 15 7(c) shows, after f  483Hz isolator A surpasses isolator B and achieves its best performance at f  833Hz with 39.9dB further reduction of transmissibility. These observations consistent with the estimation of start frequency of 507Hz given in Section II.B and the best performance frequency of 853 Hz given in Fig. 5 (c). In addition, the bending effect disappears at high frequencies. This is because compared with the circumferential size of a cell, the wavelengths are not large enough to ensure the effective wave velocities. FIG. 7. Results of sweep frequency test. (a) Accelerations measured on the aluminum base (input), the top of isolator B and the top of isolator A. (b) The power spectrum density of the accelerations (c) The transmissibility of isolator A and isolator B. Besides the sweep frequency test, the tap test is also conducted for comparison. In the tap test, the positions of two isolators and three accelerators are the same as those of the sweep frequency test (see Fig. 6). While input signal is generated by tapping on the aluminum base using a steel hammer. The recorded signals are plotted in Fig. 8(a), which confirms that isolator A super performs isolator B. Detailed comparison in frequency domain is presented in Fig. 8(c). Similar to the sweep frequency test, after f  387Hz isolator A surpasses isolator B and achieves its best performance at f  910Hz with 44.1dB further reduction of transmissibility. However, we cannot find peaks at resonant frequencies, because the input signal due to tapping is mainly at high 16 frequencies, so that the power around nf is not high enough to excite the resonance of the isolators. FIG. 8. Results of tap test. (a) Accelerations measured on the aluminum base (input), the top of isolator B and the top of isolator A. (b) The power spectrum density of the accelerations (c) The transmissibility of isolator A and isolator B. V. SUMMARY In this paper, we propose a general method to design elastic metamaterials based on the conformal transformation. The great challenge of form-invariance is solved by using the material whose longitudinal wave velocity is much larger than the transverse wave velocity. In addition, the performance of the resultant elastic metamaterial can be further improved by introducing interfaces to convert elastic waves into the modes with short wavelengths. The validity of this design method is demonstrated by an elastic wave bender acting as a vibration isolator. Although this is a simple example, it shows superior performance at broad frequency bandwidth and even breaks the limit on damping ratios required by the classic passive vibration theory. Thus, we can envisage that other interesting implementations of elastic metamaterials can be achieved in future by using the method proposed in this paper. 17 Acknowledgments: This work was supported by the National Natural Science Foundation of China with the grant number 11672144. The assistance of B. Wang and Q. H. Lu for the shaking table test is greatly appreciated. References: [1] J. B. Pendry, D. Schurig, D. R. Smith, Controlling electromagnetic fields. Science 312, 1780- 1782 (2006). [2] U. Leonhardt, Optical conformal mapping. Science 312, 1777-1780 (2006). [3] S. A. Cummer, D. Schurig, One path to acoustic cloaking. New Journal of Physics 9, 45 (2007). [4] H. Chen, C. T. Chan, Acoustic cloaking in three dimensions using acoustic metamaterials. Applied Physics Letters 91, 183518 (2007). [5] M. Farhat, et al., Broadband cylindrical acoustic cloak for linear surface waves in a fluid. Physical Review Letters 101, 134501 (2008). [6] S. Zhang, D. A. Genov, C. Sun, X. Zhang, Cloaking of matter waves. Physical Review Letters 100, 123002 (2008). [7] D. Schurig et al., Metamaterial electromagnetic cloak at microwave frequencies. Science 314, 977-980 (2006). [8] R. Liu, C. Ji, J. J. Mock, J. Y. Chin, T. J. Cui, D. R. Smith, Broadband Ground-Plane Cloak. Science 323, 366-369 (2009). [9] P. Ball, Bending the laws of optics with metamaterials: an interview with John Pendry. National Science Review 5, 200 -- 202 (2018). [10] S. Zhang, C. Xia, N. Fang, Broadband acoustic cloak for ultrasound Waves. Physical Review Letters 106, 024301 (2011). 18 [11] G. Ma, P. Sheng, Acoustic metamaterials: From local resonances to broad horizons. Science Advances 2, e1501595 (2016). [12] W. S. Gan, New Acoustics Based on Metamaterials (Springer, Singapore, 2018). [13] G.W. Milton, M. Briane, J. R. Willis, On cloaking for elasticity and physical equations with a transformation invariant form. New Journal of Physics 8, 248 (2006). [14] M. Brun, S. Guenneau, A. B. Movchan, Achieving control of in-plane elastic waves. Applied Physics Letters 94, 061903 (2009). [15] A. N. Norris, A. L. Shuvalov, Elastic cloaking theory. Wave Motion 48, 525-538 (2011). [16] J. Hu, Z. Chang, G. Hu, Approximate method for controlling solid elastic waves by transformation media. Physical Review B 84, 201101(R) (2011). [17] Z. Chang et al., Controlling elastic waves with isotropic materials. Applied Physics Letters 98, 121904 (2011). [18] M. Farhat, S. Guenneau, S. Enoch, Ultra broadband elastic cloaking in thin plates. Physical Review Letters 103, 024301 (2009). [19] N. Stenger, M. Wilhelm, M. Wegener, Experiments on elastic cloaking in thin plates. Physical Review Letters 108, 014301 (2012). [20] M. Dubois, et al., Flat lens for pulse focusing of elastic waves in thin plates. Applied Physics Letters 103, 071915 (2013). [21] J. Zhu, et al., Elastic waves in curved space: mimicking a wormhole. Physical Review Letters 121, 234301 (2018). [22] W. J. Parnell, A. N. Norris, T. Shearer, Employing pre-stress to generate finite cloaks for antiplane elastic waves. Applied Physics Letters 100, 171907 (2012). 19 [23] D. J. Colquitt et al., Transformation elastodynamics and cloaking for flexural waves. Journal of Mechanics Physics and Solids 72, 131-143 (2014). [24] Z. H. Xiang, The form-invariance of wave equations without requiring a priori relations between field variables. SCIENCE CHINA Physics, Mechanics & Astronomy 57, 2285-2296 (2014). [25] S. Brûlé, et al., Experiments on seismic metamaterials: molding surface waves. Physical Review Letters 112, 133901 (2014). [26] M. K. Lee, Y. Y. Kim, Add-on unidirectional elastic metamaterial plate cloak. Scientific Reports 6, 20731 (2016). [27] U. Leonhardt, T. Philbin, Geometry and light (Dover, New York, 2010), pp. 166-170, 173- 175. [28] G. D. Manolis, T. V. Rangelov, R. P. Shaw, Conformal mapping methods for variable parameter elastodynamics. Wave Motion 36, 185-202 (2002). [29] P. J. Olver, "Chapter 7 Symmetries of Variational Problems" in Equivalence, Invariants, and Symmetry (Cambridge University Press, Cambridge, 1995), pp. 221-251. [30] J. L. Rose, "Chapter 7 Surface and Subsurface Waves" in Utrasonic Guided Waves in Solid Media (Cambridge University Press, Cambridge, 2014), pp. 108-119. [31] H. LeKuch, "Shock and Vibration Isolation Systems" in Harris Shock and Vibration Handbook, A. G. Piersol, T. L. Paez, Eds. (McGraw-Hill, ed. 6, 2009), pp. 39.5-39.6. 20
1906.07879	1	1906	2019-06-19T02:02:20	Scalable, green fabrication of single-crystal noble metal films and nanostructures for low-loss nanotechnology applications	[ "physics.app-ph", "cond-mat.mtrl-sci", "physics.optics" ]	High quality metal thin films and nanostructures are critical building blocks for next generation nanotechnologies. They comprise low-loss circuit elements in nanodevices, provide new catalytic pathways for water splitting and $CO_2$ reduction technologies, and enable the confinement of spatially extended electromagnetic waves to be harnessed for application in information processing, energy harvesting, engineered metamaterials, and new technologies that will operate in the quantum plasmonics limit. However, the controlled fabrication of high-definition single-crystal subwavelength metal nanostructures remains a significant hurdle, due to the tendency for polycrystalline metal growth using conventional physical vapor deposition methods, and the challenges associated with placing solution-grown nanocrystals in desired orientations and locations on a surface to fabricate functional devices. Here, we introduce a new scalable, green, wet chemical approach to monocrystalline noble metals that enables the fabrication of ultrasmooth, epitaxial, single-crystal films of controllable thickness. They are ideal for the subtractive manufacture of nanostructure through ion beam milling, and additive crystalline nanostructure via lithographic patterning to enable large area, single-crystal metamaterials and high aspect ratio nanowires. Our single-crystal nanostructures demonstrate improved feature quality and pattern transfer yield, reduced optical and resistive losses, tailored local fields, and greatly improved stability compared to polycrystalline structures, supporting greater local field enhancements and enabling new practical advances at the nanoscale.	physics.app-ph	physics	Scalable, green fabrication of single-crystal noble metal films and nanostructures for low-loss nanotechnology applications Sasan V. Grayli, Xin Zhang, Finlay C. MacNab, Saeid Kamal, Gary W. Leach* Department of Chemistry, Laboratory for Advanced Spectroscopy and Imaging Research, and 4D LABS, Simon Fraser University, 8888 University Dr., Burnaby, BC V5A 1S6 Canada High quality metal thin films and nanostructures are critical building blocks for next generation nanotechnologies1-3. They comprise low-loss circuit elements in nanodevices4,5, provide new catalytic pathways for water splitting and CO2 reduction technologies6,7, and enable the confinement of spatially extended electromagnetic waves to be harnessed for application in information processing8, energy harvesting9,10, engineered metamaterials,11 and new technologies that will operate in the quantum plasmonics limit12. However, the controlled fabrication of high-definition single-crystal subwavelength metal nanostructures remains a significant hurdle, due to the tendency for polycrystalline metal growth using conventional physical vapor deposition methods, and the challenges associated with placing solution-grown nanocrystals in desired orientations and locations on a surface to fabricate functional devices. Here, we introduce a new scalable, green, wet chemical approach to monocrystalline noble metals that enables the fabrication of ultrasmooth, epitaxial, single-crystal films of controllable thickness. They are ideal for the subtractive manufacture of nanostructure through ion beam milling, and additive crystalline nanostructure via lithographic patterning to enable large area, single-crystal metamaterials and high aspect ratio nanowires. Our single-crystal nanostructures demonstrate improved feature quality and pattern transfer yield, reduced optical and resistive losses, tailored local fields, and greatly improved stability compared to polycrystalline structures, supporting greater local field enhancements and enabling new practical advances at the nanoscale. The immense and growing interest in metal thin films and nanostructures13 results from their ability to support surface plasmons (SPs) that concentrate light below the diffraction limit providing a bridge between high bandwidth photonic fiber-based technology and the nanometer- scale structures that comprise current integrated circuitry.8 SPs are characterized by ultrafast response and can mediate rapid photon-to-hot electron conversion which can be exploited for new solar energy, photosensor, and photocatalyst applications.10,14-16 Engineered metamaterials can provide negative refractive index17,18, subwavelength resolution19,20, and field manipulation,21-23 enabling diffraction-free imaging and pattern transfer. Improvements in nanoscale fabrication methods offer design flexibility and structure generation with the ability to manipulate the local photonic density of states and to control light -- matter interactions at the quantum level12,24,25. Local near fields can significantly enhance light−matter interactions enabling new nano-atto-physics26 and the opportunity to engineer the radiative rates of quantum emitters confined to nanocavities, with the prospects of single-molecule sensing, nanoscale light sources, single-photon emitters, and all-optical transistors.27,28 These applications place stringent requirements on surface quality in defining local fields and field enhancements, as well as the nanometer-level positional and orientational control of emitters with respect to surface features. In practice, plasmonic metals deposited by conventional methods (e.g. physical vapour deposition) are characterized by polycrystalline morphologies comprised of grain boundaries, defects, and other material imperfections that act as local scattering sites, sources of increased optical absorption loss, dissipative damping, and positional uncertainty. 29,30 They compromise pattern transfer fidelity and device yield, and limit functional performance. Likewise, strategies that employ the synthesis of solution-grown nanocrystals suffer from the major challenge of placing them in desired locations onto substrates with high fidelity, and the additional barrier associated with surfactants and nanocrystal capping agents necessary to prevent particle aggregation and agglomeration in solution, but that prevent direct electrical contact. In order to improve device yield, decrease optical and resistive losses, and to exploit the local electromagnetic fields of noble metal nanostructures fully, improved control over surface quality and chemistry is imperative. While this has remained a significant challenge in the field and has led to growing efforts to identify alternative low-loss materials for plasmonic and metamaterial applications31, their high carrier concentrations with visible and near infrared optical responses remain extremely attractive and continue to foster new strategies to exploit noble metal-based plasmonics. Here we describe a new, green approach to monocrystalline noble metal plasmonic structures that is based on the deposition of noble metals from solutions of their commonly available salts (Fig. 1). Figure 1. Epitaxial electrochemical deposition of single-crystal noble metal thin films and nanostructures. Center: Solution-phase reduction of Au3+-containing complex ions to Au atoms at the Ag(100)/aqueous alkaline electrolyte interface. Center-Right: Deposition of a uniform, ultrasmooth, epitaxial, single-crystal Au(100) film of controlled thickness (upper), with wafer-scale demonstration (lower). Right: Excitation of a bowtie nanoantenna fabricated via FIB milling of the single-crystal Au film (upper) with a top-view SEM image of a bowtie with 20 nm nanocavity gap (lower). Center-Left: Electron beam lithography is used to create a patterned resist layer on a solution-deposited single-crystal gold film (upper). Epitaxial electrochemical deposition into the pores yields a Au nanopillar array showing angle-dependent light scattering (lower). Left: The resulting nanopillars are monocrystalline and oriented, displaying (100) top facets and dominant (111) angled sidewalls (upper), with a top-view SEM demonstrating a large area metasurface comprised of oriented crystalline pillars with uniform shape, orientation and location (lower). Aqueous solutions of chloroauric acid (HAuCl4) contain hydrated Au(III)-based complex ions (AuCl4¯) whose standard reduction potentials (AuCl4¯ + 3e¯ → Au + 4Cl¯ : Eᵒ = 1.00 V) are greater than that of silver (Ag+ + e¯ → Ag : Eᵒ = 0.80 V). Reduction of Au3+ to Au in the presence of silver substrates proceeds spontaneously by galvanic replacement, in which Au3+-containing ions are reduced at the expense of oxidation of the Ag atoms of the substrate, resulting in porous, polycrystalline gold and gold/silver alloy materials. This chemistry has been exploited to yield hollow colloidal nanostructures in solution with tunable and controlled properties for application in plasmonics, photocatalysis, and nano-medicine.32 It has also been demonstrated that control over the relative rates of galvanic replacement and Au3+-complex ion reduction in the presence of organic acid reducing agents can provide core-shell colloidal nanocrystals containing thin epitaxial layers of gold.33,34 However, the ability to affect noble metal ion reduction without galvanic replacement, over large surface areas, with thickness control, and with nanometer-scale patterning capability, would provide a new level of control over surface nanostructure and open new opportunities for practical implementation of novel nanometer-scale technologies. Here we describe the reduction of Au3+-complex ions in highly alkaline environments in the absence of other reducing agents to yield the controlled epitaxial deposition of Au onto large area single-crystal Ag(100) substrates. Under strongly alkaline conditions, two important effects supress galvanic replacement. At high pH, OH¯ ions displace the Cl¯ ligands of the AuCl4¯ complexes leading to the formation of Au(OH)4¯ ions35, whose standard reduction potentials are lowered to 0.57 V (supplementary information). This is below the silver reduction potential under non-alkaline conditions. However, surface hydroxide residing at the Ag/electrolyte interface under highly alkaline conditions presents a significant additional barrier to surface oxidation, and also helps to arrest galvanic replacement. The available surface oxidation processes under these alkaline conditions have been attributed36 to the electroformation of soluble [Ag(OH)2]¯ and the growth of Ag2O which appears at standard reduction potentials of 1.40 V (supplementary information). In the absence of silver substrate oxidation, gold ion reduction can then proceed spontaneously through readily available hydroxide ions in the absence of other reducing agents: Reduction: Oxidation: Au(OH)4¯ + 3e¯ → Au + 4OH¯ 4OH¯ → O2 + 2H2O + 4e¯ (Eᵒ = 0.57 V) (Eᵒ = -0.40 V) Spontaneous Red-Ox: 4Au(OH)4¯ → 4Au + 3O2 + 6H2O + 4OH¯ (Eᵒ = 0.17 V) The highly alkaline conditions provide a high concentration and uniform distribution of hydroxide ions that leads to uniform noble metal ion reduction, affording large area metal deposition. Note that electrochemical deposition of noble metals typically involves electrolyte baths that contain highly toxic complexing agents and bath additives designed to improve bath stability and metal deposition characteristics, but that have a negative environmental footprint.37 In contrast, our chemistry affords large area uniform gold deposition without the use of toxic additives, employing only alkaline conditions which can later be removed through bath neutralization to yield water. Metal deposition rates and film thickness can be tuned by control over reduction kinetic parameters including metal salt concentration, deposition temperature, and deposition time. Further, the chemistry can be carried out at the wafer level (Fig. 1), and therefore represents a scalable pathway to single-crystal noble metal thin films and nanostructure. Solution phase Au deposition from uncontrolled pH HAuCl4 solutions (pH 6) onto single-crystal Ag(100)/Si(100) substrates leads to the deposition of polycrystalline gold and concomitant silver film oxidation, consistent with the AuCl4¯ -induced galvanic replacement mechanism, as described. Two-dimensional X-ray diffraction (2D-XRD) patterns display (111), (200), and (220) Au diffraction arcs characteristic of polycrystalline metal deposition (Fig. 2a). In contrast, electroless Au deposition from high alkalinity (pH 14) HAuCl4 solutions onto the same Ag(100)/Si(100) substrates display well-defined Au(200) diffraction spots and an absence of diffraction arcs, characteristic of oriented, substrate-aligned crystalline metal deposition (Fig 2b). Solution-deposition onto (Ag(100)/Si(100)) single-crystal silver substrates under high alkalinity conditions results in uniform, large area, ultra-smooth Au surfaces (Fig 2c). Physical vapor deposition (PVD) of gold onto Si(100) substrates with a 5nm Cr adhesion layer (a typical PVD- based deposition method) results in polycrystalline gold island growth and coalescence into thin gold films that are far less uniform by comparison (Fig. 2d). Transmission electron microscopy (TEM) provides evidence of the nature of the gold deposition from solution. Elemental mapping (Fig. 2(e)-(h)) reveals the deposition of a well-defined, dense, uniform gold layer atop the Ag(100)/Si(100) single-crystal substrate rather than a porous Au/Ag alloy film, confirming that under high alkalinity conditions, gold ion reduction does not occur through Ag substrate oxidation and galvanic replacement. High resolution transmission electron microscopy (HRTEM) and selected area electron diffraction (SAED) images of the Ag/Au interface region (Fig 2(i)-(l)) demonstrate that under these high alkalinity conditions, gold deposition occurs epitaxially, resulting in a well-defined interface region with alignment of the deposited Au film atoms with those of the underlying single-crystal silver substrate. Figure 2. (a) 2D-XRD of gold deposited from an uncontrolled pH HAuCl4 solution onto a Ag(100)/Si(100) single-crystal substrate. (b) 2D-XRD of gold deposited from a pH 14 HAuCl4 solution onto a Ag(100)/Si(100) single-crystal substrate. (c) Top view SEM of a 100 nm thick gold film deposited from pH 14 HAuCl4 solution onto a Ag(100)/Si(100) single-crystal substrate. (d) Top view SEM of a 100 nm thick Au film evaporated onto an atomically flat Si(100) substrate with a 5nm Cr adhesion layer. High resolution transmission electron microscopy of pH 14 solution-deposited, 70 nm thick Au film onto a Ag(100)/Si(100) single-crystal substrate: (e) TEM cross section image of protective Pt-overlayer/Au(100)/Ag(100) /Si(100) with Pt appearing in the lower left and silicon wafer appearing dark in the upper right hand region of the image. (f)-(h) Elemental mapping of the Au(100)/Ag(100)/Si(100) structure (silicon upper right). (i) Cross-sectional TEM image of the Pt /Au(100)/Ag(100) interface region. (j) Expanded view of the Au(100)/Ag(100) interface. (k) The Au(100)/Ag(100) interface showing alignment of atomic planes across the interface. (l) Selected area electron diffraction from the region highlighted in (k) viewed along the [011] zone axis. The utility of this chemistry and some of its advantages over conventional physical vapor deposition-based methods are demonstrated in Figure 3. Focused ion beam (FIB) milling has been used to fabricate Au nanostructures from solution-deposited single-crystal epitaxial films and from the polycrystalline PVD-deposited Au films described above. Without exception, the pattern transfer fidelity and structure definition of our solution-deposited single-crystal films are far superior to conventional polycrystalline PVD-deposited films. Anisotropic, crystal direction- dependent ion milling rates in polycrystalline films yield non-uniform structures that reduce pattern transfer quality and that act as local scattering centers for electronic, photonic and plasmonic excitations. Four point probe transport measurements of these 100 nm-thick gold films show that single-crystal solution-deposited films yield sheet resistances more than 20 times below those of PVD-deposited polycrystalline films of the same thickness (supplementary information). Spectroscopic ellipsometry performed on 100 nm thick Au films show that optical absorption losses in the single-crystal films are significantly reduced compared to those of the polycrystalline PVD-deposited films (supplementary information). Our observations regarding the solution-phase growth of these noble metal films suggest a radically different growth mechanism to the island formation and coalescence observed through PVD growth. Film thickness can be controlled through kinetic parameters including deposition time (Fig. 3(d)) to yield continuous films even at sub-10 nm film thickness (supporting information), suggesting rapid 2-dimensional growth along the in-plane <110> directions, and the potential utility of this deposition strategy for ultrathin metal applications. Further, this chemistry provides metals of electronic device quality as demonstrated by the well-formed Schottky diode behavior of the planar Au/ZnO interface (Fig. 3(e)). Figure 3. Focused ion beam milling of 100 nm thick, polycrystalline, PVD-deposited Au nanostructures and monocrystalline, solution-deposited Au nanostructures. SEM images of (a) ring resonator structures from polycrystalline, PVD-deposited Au (left) and solution-deposited Au (right), (b) patterned windows in PVD-deposited Au (left) and solution-deposited Au (right), (c) 90 nm diameter holes patterned in PVD-deposited Au (left) and solution-deposited Au (right). (d) Au(100) film thickness measured by cross-sectional SEM versus deposition time demonstrating an approximately linear relationship. Film thickness measurements in triplicate (inset) demonstrate thickness reproducibility. (e) Current-Voltage (I-V) diode curve of the high asymmetry solution- deposited Au(100)/ZnO Schottky interface. We compare directly bowtie nanoantenna devices manufactured through FIB milling of monocrystalline and polycrystalline films (Fig. 4). These structures have stringent deposition and patterning requirements to yield precision structures that display uniform and reproducible local gap fields at the antenna's feedpoints. The bowtie nanoantenna features were patterned with sequential FIB milling steps of rectangular and square features to yield bowtie nanocavity gaps of 20 nm. This method of fabrication also highlights regions of the bowtie structures where there are metal step edges that result from this pattern generation scheme. SEM images of the structures show significantly higher quality pattern transfer and structure definition of the single- crystal bowtie nanoantennas compared to polycrystalline devices fabricated identically (Figs 4(a)- (b)). Two-photon photoluminescence (2PPL) imaging has been used extensively to characterize the resonant behaviour of plasmonic nanostructures38,39 and is used here (Fig. 4(c)-(g)) to provide insight into the nanoantenna plasmonic response and local field generation from the bowtie structures. The 2PPL maps of 3 x 3 bowtie arrays demonstrate that the fabrication yield of functional devices is greatly impacted by the material quality and associated pattern transfer characteristics. The yield of monocrystalline antennas is close to 100% as measured by the appearance of an enhanced local near-field resulting in 2PPL intensity at the antenna feed points and the uniformity of this 2PPL intensity for all nanoantennas (Fig 4c). Structures fabricated identically but with polycrystalline-deposited gold, show poor fabrication yield with fewer than 50% of the devices showing near-field intensity enhancements at the antenna feed points, and of these, no uniformity in 2PPL intensity (Fig 4d). Figure 4. Single-crystal versus polycrystalline bowtie nanoantenna fabrication and performance. SEM image of bowtie nanoantenna patterned by FIB milling of (a) solution-deposited Au(100) and (b) PVD-deposited polycrystalline Au films. Scanning laser microscope image of 2PPL (horizontally-polarized, 780 nm excitation, 120 fs pulse duration) of 3 x 3 bowtie nanoantenna arrays fabricated from (c) solution-deposited Au(100) and (d) PVD-deposited polycrystalline Au films. 2PPL image of (e) individual solution-deposited Au(100) nanoantenna and (f)-(g) individual PVD-deposited polycrystalline Au nanoantennas. Our single-crystal structures also afford superior ability to control and tailor local fields. The single-crystal bowties show relatively uniform 2PPL intensity across all nanoantennas at the antenna feed points and in regions of the fabricated structure (Fig 4e) where sharp gold edges and discontinuities are formed due to the FIB pattern generation scheme. Polycrystalline bowties (Fig. 4(d),(f)-(g)), in contrast, show that two photon photoexcitation results in non-uniform plasmonic excitation over the entire milled area of the bowties due to structural inhomogeneity and grain boundary-induced plasmon excitation and dissipation. In few cases do the polycrystalline structures yield enhanced near-fields at the antenna's feed points. Importantly, our single-crystal solution-deposited bowtie antennas demonstrate superior thermal and mechanical stabilities compared to their polycrystalline counterparts. Illumination of the bowtie antennas with increasing incident illumination intensity results in higher intensity 2PPL emission (2PPL intensity is proportional to I2, where I is the local near-field intensity enhancement40,41) until they are catastrophically damaged through photothermally induced structural modification and rupture. Intensity dependent studies of the 2PPL from the bowtie structures indicate that the single-crystal bowties can support more than one order of magnitude more incident illumination intensity (and therefore 104 local field enhancement) than the polycrystalline bowties before irreversible and catastrophic loss. We assert that this is a direct result of less local heat dissipation through grain boundary loss and increased thermal and mechanical stability of the single-crystal structures, suggesting that single-crystal nanocavity structures such as these will find beneficial application in strong field coupling and atto-nano-physics applications, where enhanced local fields and high damage thresholds are key functional requirements. Solution-deposited Au(100) bowtie devices fabricated through FIB milling demonstrate multiple advantages over their polycrystalline counterparts. Nevertheless, the broader integration of nanostructured elements into useful device structures requires cost effective, manufacturable strategies that provide large area patterning capability. Here we demonstrate the utility of our green chemistry with the use of electron beam lithography (EBL) to deposit large area arrays of single-crystal noble metal nanostructures through additive patterning. Figure 5(a) shows a top view SEM image of a gold nanopillar array solution-deposited onto an e-beam patterned, solution-deposited Au(100) substrate: A 200 nm thick layer of PMMA A4 electron-beam resist is spin cast onto a solution-deposited Au(100) top surface. Following electron beam patterning and resist development, Au is deposited from solution into the 350 nm diameter, 700 nm period, cylindrical pores of the patterned resist layer by immersion into the noble metal salt-containing electrolyte used to obtain the underlying ultrasmooth Au(100) films. Following metal deposition, subsequent resist removal yields the patterned nanopillar array, demonstrating high quality pattern transfer. Single pillars (Fig. 5b) display (111)-faceted side walls and top (100) facets consistent with monocrystalline pillar deposition. 2PPL from the single-crystal plasmonic Au metamaterial array (Fig. 5(c)) shows pillar-resolved emission and demonstrates near-field plasmonic enhancement associated with each of the faceted gold nanopillars. Fig. 5(d) demonstrates the compatibility of this chemistry with silver deposition. The top-view SEM image shows a faceted single silver nanopillar from a Ag nanopillar array deposited onto a Au(100) substrate from a 1.0 M OH- ion containing electrolyte bath prepared from AgNO3, in a manner similar to that described for gold nanopillar deposition. Figure 5. Additive patterning of single-crystal metals through solution-deposition on EBL-patterned substrates. (a) SEM top view image of a large area crystalline Au nanopillar array with pillar diameter of 350 nm and period 700 nm, solution-deposited on an EBL-patterned, solution-deposited Au(100) substrate. (b) SEM top view image of an individual gold nanopillar exhibiting crystalline facets. (c) Pillar-resolved 2PPL from the Au plasmonic metamaterial array. (d) SEM top view image of a crystalline 120 nm diameter silver nanopillar, solution-deposited onto a solution-deposited Au(100) substrate, exhibiting well defined top facets. (e) SEM top view image of a faceted gold-capped silver nanopillar obtained by solution-deposition of 10 nm of Au onto a Ag(100) nanopillar array. (f) SEM top view image of high aspect ratio concentric square Au nanowire structures EBL-deposited from solution onto a Ag(100) substrate. (g) The high aspect ratio nanowires are continuous and characterized by 40 nm widths and 2 mm lengths without dose optimization (h) 2PPL image of the concentric square nanowire structure described in (f) excited by 800 nm light polarized horizontally, perpendicular to the vertical nanowire axes. The high definition faceted structure implies successful silver-on-gold heteroepitaxial solution phase deposition. While silver structures are known to possess superior plasmonic properties to those comprised of gold, they suffer from chemical instability and ready oxidation under ambient conditions. Deposition of a thin, oxidation-resistant, gold overlayer can provide chemical resistance without significant perturbation to the plasmonic properties of the underlying silver structures.33,34 Figure 5(e) shows a top-view SEM image of a silver nanopillar with a thin ( 10 nm) overlayer of gold. The image shows that the resulting core-shell nanopillar displays octagonal faceted structure suggesting epitaxial deposition and conformal gold coating of the silver pillar. We have also investigated the utility of this chemistry for the deposition of high aspect ratio gold nanowires. Shrinking feature size and increasing density of nanoscale circuit elements will benefit from low resistance monocrystalline structure to help manage thermal budgets. Figure 5(f) shows the top-view SEM image of a portion of a concentric square Au nanowire array deposited onto a Ag(100) substrate by EBL patterning and solution phase deposition of Au, as described. Figure 5(g) shows the pattern transfer of these continuous nanowire structures with nominal widths of 40 nm. Together with typical lengths of 2 mm, these features yield an aspect ratio > 104, with further improvements anticipated by electron beam dose optimization. The concentric square nanowire array also displays broadband plasmonic response (Fig 5(h)) when illuminated with horizontally polarized light, perpendicular to the vertically oriented nanowire long axes. The image shows preferential emission from vertically oriented nanowires, consistent with short-axis polarized plasmonic excitation and two-photon photoluminescence. Additive patterning of single-crystal nanowire structures and plasmonic elements on the same silicon support offers an interesting opportunity for nanometer scale technology integration and the broader implementation of nanophotonic devices into existing silicon platforms. Conclusion In summary, we have developed a new scalable, green fabrication platform that enables the deposition of epitaxial, single-crystal noble metal thin films and nanostructures from solution. The chemistry is compatible with both subtractive and additive patterning methods and shows high fidelity pattern transfer to generate single-crystal structures over extended geometries. We demonstrate that single-crystal bowtie nanoantennas fabricated with this chemistry and focused ion beam milling show improved fabrication yield, greater control over local fields, and improved thermal and mechanical stability compared with polycrystalline structures patterned identically. The utility of this chemistry with additive lithographic patterning methods provide large area single-crystal metamaterial arrays and high aspect ratio nanowire structures. We anticipate that this accessible and cost-effective approach will be broadly exploited to fabricate new robust single-crystal structures with limited optical and resistive losses and unrivaled homogeneity, enabling efficiency improvements and new practical nanometer-scale technologies. REFERENCES 1076 (2017). 1) Samarth, N. Quantum materials discovery from a synthesis perspective, Nat. Mat. 16, 1068- 2) Huang, J-S. et al. Atomically flat single-crystalline gold nanostructures for plasmonic nanocircuitry, Nat. Comm., DOI: 10.1038/ncomms1143 (2010). 3) Mahenderkar, N. K. et al. Epitaxial lift-off of electrodeposited single-crystal gold foils for flexible electronics, Science, 355, 1203 -- 1206 (2017). 4) Nanowires: Building Blocks for Nanoscience and Nanotechnology, Zhang, A., Zheng, G. & Lieber, C. M. Springer (2016) ISBN 978-3-319-41979-4. 5) Shi, J., Monticone, F., Elias, S., Wu, Y., Ratchford, D., Li, X., Alù, A. Modular assembly of optical nanocircuits, Nat. Comm., DOI: 10.1038/ncomms4896 (2014). 6) Voiry, D., Shin, H. S., Loh, K. P., Chhowalla, M. Low-dimensional catalysts for hydrogen evolution and CO2 reduction, Nat. Rev. Chem., 2, Article number: 0105 (2018). 7) Zhu, W. et al. Monodisperse Au nanoparticles for selective electrocatalytic reduction of CO2 to 8) CO. J. Am. Chem. Soc. 135, 16833 -- 16836 (2013). Ozbay, E. Plasmonics: Merging photonics and electronics at nanoscale dimensions, Science 311, 189-193 (2006). 9) Atwater, H. A., & Polman, A. Plasmonics for improved photovoltaic devices, Nat. Mat., 9, 10) Moskovits, M. The case for plasmon-derived hot carrier devices, Nat. Nanotechnol. 10, 6-8, 865 (2010). (2010). 11) Smith, D. R., Pendry, J. B. & Wiltshire, M. C. K. Metamaterials and Negative Refractive Index, 12) Tame, M.S. et. al. Quantum Plasmonics. Nature Physics 9, 329-340 (2013). 13) Fernández-Domínguez, A. I., García-Vidal, F. J. & Martín-Moreno, L. Unrelenting plasmons, Science 305, 788-792 (2004). Nat. Photon. 11, 8 -- 10 (2017). 14) Linic, S., Christopher, P. & Ingram, D. B. Plasmonic-metal nanostructures for efficient conversion of solar to chemical energy. Nat. Mater. 10, 911 -- 921 (2011). 15) Clavero, C. Plasmon-induced hot-electron generation at nanoparticle/metal-oxide interfaces for photovoltaic and photocatalytic devices. Nat. Photon. 8, 95 -- 103 (2014). 16) Brongersma, M. L., Halas, N. J. & Nordlander, P. Plasmon-induced hot carrier science and technology, Nat. Nanotechnol. 10, 25 -- 34 (2015). 17) Shalaev, V. M., Cai, W., Chettiar, U. K., Yuan, H.-K., Sarychev, A. K., Drachev, V. P. & Kildishev, A. V. Negative index of refraction in optical metamaterials, Opt. Lett. 30, 3356-3358, (2005). 18) Liu, N., Guo, H., Fu, L., Kaiser, S., Schweizer, H. & Giessen, H. Three-dimensional photonic metamaterials at optical frequencies, Nat. Mat. 7, 31-37 (2008). 19) Zhang, X. & Liu, Z. Superlenses to overcome the diffraction limit, Nat. Mat. 7, 435-441 20) Lu, D. & Liu, Z. Hyperlenses and metalenses for far-field super-resolution imaging, Nat. Comm., DOI: 10.1038/ncomms2176 (2012). 21) Pendry, J.B., Schurig, D. & Smith, D.R. Controlling Electromagnetic Fields, Science 312, 1780- (2008). 1782, (2006). 22) Leonhardt, U. Optical conformal mapping. Science 312, 1777 -- 1780 (2006). 23) Shalaev, V. M. Transforming light. Science 322, 384 -- 386 (2008). 24) Altewischer, E., van Exter, M. P. & Woerdman, J. P. Plasmon-assisted transmission of entangled photons. Nature 418, 304 -- 306 (2002). 25) Jacob, Z. & Shalaev, V. M. Plasmonics goes quantum. Science 334, 463-464 (2011). 26) Ciappina, M. F. et al, Attosecond physics at the nanoscale, Rep. Prog. Phys. 80, 054401 (2017). 27) Chikkaraddy, R., de Nijs, B., Benz, F. , Barrow, S.J., Scherman, O.A., Rosta, E., Demetriadou, A., Fox, P., Hess, O. & Baumberg, J. J. Single-molecule strong coupling at room temperature in plasmonic nanocavities, Nature, 535, 127-130 (2016). 28) Vasa, P. & Lienau, C. Strong Light−Matter Interaction in Quantum Emitter/Metal Hybrid Nanostructures, ACS Photonics 5, 2−23 (2018). 29) Khurgin, J. B. How to deal with the loss in plasmonics and metamaterials. Nat. Nanotechnol. 30) Krenn, J. Perspective on Plasmonics, Nature Photonics 6, 714 -- 715 (2012). 31) Boltasseva, A. & Atwater, H. A. Low-loss plasmonic metamaterials. Science 331, 290 -- 291 10, 2 -- 6 (2015). (2011). 32) Xia, X., Wang, Y., Ruditskiy, A., & Xia, Y. 25th Anniversary Article: Galvanic Replacement: A Simple and Versatile Route to Hollow Nanostructures with Tunable and Well-Controlled Properties. Adv. Mater., 25 6313-6333 (2013). 33) Liu, H., Liu, T., Zhang, L., Han, L., Gao, C. & Yin, Y. Etching-Free Epitaxial Growth of Gold on Silver Nanostructures for High Chemical Stability and Plasmonic Activity. Adv. Func. Mater., 25, 5435-5443 (2015). 34) Yang, Y., Liu, J., Fu, Z.-W., & Qin, D. Galvanic Replacement-Free Deposition of Au on Ag for Core−Shell Nanocubes with Enhanced Chemical Stability and SERS Activity, J. Am. Chem. Soc., 136, 8153−8156 (2014). 35) Wang, S., Qian, K., Bi, X. & Huang, W. Influence of Speciation of Aqueous HAuCl4 on the Synthesis, Structure, and Property of Au Colloids, J. Phys. Chem. C., 113, 6505 -- 6510 (2009). 36) Abd El Rehim, S. S., Hassan, H. H., Ibrahim, M. A. M. & Amin, M. A. Electrochemical behaviour of a silver electrode in NaOH solutions, Monatshefte fuer Chemie, 129 1103-1117 (1998). 37) Green, T. A. Gold electrodeposition for microelectronic, optoelectronic and microsystem applications, Gold Bull. 40, 105 (2007). 38) Beversluis, M. R., Bouhelier, A. & Novotny, L. Continuum generation from single gold nanostructures through near-field mediated intraband transitions. Phys. Rev. B 68, 115433 (2003). 39) Imura, K., Nagahara, T. & Okamoto, H. Plasmon mode imaging of single gold nanorods. J. 40) Mühlschlegel, P., Eisler, H-J., Martin, O. J. F., Hecht, B. & Pohl, D. W. Resonant optical Am. Chem. Soc. 126, 12730 -- 12731 (2004). antennas. Science 308, 1607 -- 1609 (2005). 41) Ghenuche, P., Cherukulappurath, S., Taminiau, T. H., van Hulst, N. F. & Quidant, R. Spectroscopic mode mapping of resonant plasmon nanoantennas. Phys. Rev. Lett. 101, 116805 (2008). ACKNOWLEDGMENTS X. Yuan is thanked for technical assistance with ellipsometry data (supplementary information). Funding: This work is supported by the Natural Sciences and Engineering Research Council of Canada (Project number: RGPIN-2017-06882) and CMC Microsystems (MNT Financial Assistance Program). This work made use of 4D LABS and the Laboratory for Advanced Spectroscopy and Imaging Research (LASIR) shared facilities supported by the Canada Foundation for Innovation (CFI), British Columbia Knowledge Development Fund (BCKDF) and Simon Fraser University. Author contributions: S.V.G. and G.W.L conceived and designed the experiments, S.V.G. performed all film deposition, characterization, and nanofabrication experiments, F.C.M. developed the methodology and fabricated single-crystal silver substrates, X.Z. performed the TEM experiment and analysis, S.K. performed laser scanning 2PPL microscopy experiments and analyses, G.W.L. wrote the manuscript with input from all. Competing interests: The authors declare no competing interests. Data and Materials availability: All data are available in the manuscript and the supplementary information. Supplementary Information Scalable, green fabrication of single-crystal noble metal films and nanostructures for low-loss nanotechnology applications Sasan V. Grayli, Xin Zhang, Finlay C. MacNab, Saeid Kamal, Gary W. Leach* Single Crystal Ag(100)/Si(100) Substrates Single crystal Ag(100)/Si(100) substrates were prepared by thermal evaporation of silver onto H-terminated Si(100) substrates. Silver deposition was conducted using a Kurt J. Lesker Company PVD-75 thermal evaporation tool with a base pressure of < 2 × 10-7 Torr. Ag (99.99% Kurt J. Lesker Company) was evaporated from an alumina coated tungsten wire basket. The substrate was heated via a backside quartz lamp and the temperature was monitored with a K- type thermocouple attached to the backside of the sample chuck assembly. Deposition was carried out at a temperature of 340°C and a rate of 3 Å/s to a thickness of 500 nm. Prior to Ag deposition, substrates were immersed in commercial buffered oxide etch solutions (BOE, CMOS Grade, J.T. Baker Inc.), to remove the native oxide layer from the surface of the silicon wafer. All activities, prior to characterization of the films, were carried out under class 100 clean room conditions or better. A more complete description of the deposition characteristics and crystallite evolution of silver evaporated onto silicon substrates will appear in a forthcoming publication. Physical Vapour Deposition of Gold Films Thermal evaporation of gold onto Si(100) substrates was carried out to provide a source of thin film gold that would represent the typical polycrystalline film quality, characteristic of PVD deposition. Onto a native oxide covered Si(100) wafer was deposited 5 nm of chromium to act as an adhesion layer. Gold was thermally evaporated at 1 Å/s onto an unheated substrate under substrate rotation. This resulted in gold island growth and coalescence into thin polycrystalline gold films. A top view SEM of a typical film is displayed is Fig. 2d of the manuscript. Electroless Growth of Noble Metal Films Gold films were deposited spontaneously from solutions of chloroauric acid (HAuCl4) onto single crystal Ag(100) substrates prepared as described. Gold films deposited from aqueous HAuCl4 solutions without pH control resulted in galvanic replacement, in which the monocrystalline silver substrate was quickly oxidized and resulted in a poor quality, dark film which was later determined to be a porous polycrystalline film of silver and gold (Fig S1a). In contrast, the same deposition from pH 14 solutions led to the production of high optical quality gold films (Fig S1b). As discussed in the main text, galvanic replacement was avoided by maintaining a high concentration of hydroxide ions in solution. Single crystal Au(100) film deposition was carried out by immersing a 1 x 1 cm2 Ag(100)/Si(100) substrate into a deposition bath maintained at 60°C. The deposition bath was a mixture of 500 μL of 0.0025 M HAuCl4 in 10 mL of 1.0 M NaOH (all solutions prepared from Millipore purity water of 18.2 MΩ-cm resistivity). After 1 hour, the sample was removed from the deposition bath and rinsed with distilled water for 2 minutes and then air dried. Film thickness and deposition rate were found to be well controlled through control of kinetic parameters such as HAuCl4 concentration, deposition temperature, and deposition time. Optical images of Au deposited from solution onto single crystal Ag(100)/Si(100) substrates under conditions of galvanic replacement (uncontrolled pH) and highly alkaline conditions (pH 14) are shown in Figure S1. Figure S1. Photo of a Au film following Au deposition onto a single crystal Ag(100)/Si(100) substrate from (a) an electroless deposition bath containing HAuCl4 at uncontrolled pH and b) an electroless deposition bath containing HAuCl4 at pH 14 (1 cm x 1 cm substrate). Figure S2. Top view SEM images of ultrathin Au(100) films deposited from an alkaline solution of 0.000625 M HAuCl4 onto a single crystal Ag(100)/Si(100) substrate for 30 sec at 70°C. Film thickness is estimated to be 8 nm as determined by thickness measurements at defects, in general agreement with the calibration curve shown in Fig 3(d) of the main text. Cyclic Voltammetry A cyclic voltammetry study was carried out in order to determine the oxidation potential of Ag under the 1.0 M alkaline conditions. Standard three-electrode electrochemical cell conditions comprising a Ag/AgCl (3 M KCl) reference electrode and a platinum wire counter electrode were employed. Figure S3 shows the cyclic voltammagram of a Ag(100)/Si(100) single crystal working electrode immersed in a 1 M OH- electrolyte. The quasi-reversible CV shows the lowest energy redox process at 0.375 V versus Ag/AgCl, attributed to electroformation of soluble [Ag(OH)2]¯ and the growth of Ag2O. Relative to the standard hydrogen electrode (SHE) under standard (1 M [H+]) conditions, the measured redox potential (under pH=14 conditions) corresponds to a potential of E= 0.375 + 0.197(Ag/AgCl vs RHE) + 0.826(RHE at pH=14) = 1.398 V. Figure S3: Quasi-reversible cyclic voltammagram of a Ag(100)/Si(100) single crystal working electrode immersed in a 1.0 M OH¯ electrolyte. The redox potential appears at 0.375 V versus Ag/AgCl. Determination of Standard Reduction Potential of Gold Hydroxide Complex The standard reduction potential of the Au(OH)4 ¯ complexes in the deposition bath was measured by constructing a galvanic cell. The galvanic cell was formed by immersing a zinc (Zn) electrode into 10 mL of a 1.0 M ZnSO4 solution to form one half cell. The other half cell was comprised of a polished Pt wire immersed in a pH=14 electrolyte containing Au(OH)4 (0.025M) to 10 mL of a concentrated NaOH bath. The two half cells were connected with a salt bridge and a galvanic potential of 1.265 V was measured between the two electrodes with a high impedance digital volt meter. The cell potential was corrected for the concentration of Au(OH)4 ¯ ]= 625 μM) through the Nernst equation: ¯ , obtained by the addition of 250 μL of HAuCl4 ¯ ( [Au(OH)4 𝐸𝐸= 𝐸𝐸0−0.059𝑛𝑛 log𝑄𝑄 where E is the measured cell potential, E0 is the cell potential under standard conditions, n is the number electrons transferred, and Q is the redox reaction quotient. The corrected cell potential was determined to be 1.33 V. Employing the accepted standard reduction potential of Zn2+ (E0 = -0.76 V) allows one to determine the standard reduction potential of Au(OH)4 ¯: 3𝑍𝑍𝑛𝑛⇌3𝑍𝑍𝑛𝑛2++6𝑒𝑒− −𝐸𝐸0=0.76 𝑉𝑉 2𝐴𝐴𝐴𝐴(𝑂𝑂𝑂𝑂)4−+6𝑒𝑒−⇌2𝐴𝐴𝐴𝐴+8𝑂𝑂𝑂𝑂− 𝐸𝐸0=0.57 𝑉𝑉 3𝑍𝑍𝑛𝑛+ 2𝐴𝐴𝐴𝐴(𝑂𝑂𝑂𝑂)4−⇌3𝑍𝑍𝑛𝑛2++2𝐴𝐴𝐴𝐴+8𝑂𝑂𝑂𝑂− 𝐸𝐸0=1.33 𝑉𝑉 𝐴𝐴𝐴𝐴(𝑂𝑂𝑂𝑂)4−+3𝑒𝑒−⇌𝐴𝐴𝐴𝐴+4𝑂𝑂𝑂𝑂− 𝐸𝐸0=0.57 ± 0.010 𝑉𝑉 These results yield a standard reduction potential of: demonstrating that under high alkalinity conditions, the formation of Au(OH)4 dramatic decrease of the Au3+ complex ion reduction potential. ¯ complexes leads to a Nanopillar Array Fabrication Nanopillar arrays are formed by electroless deposition of Au and Ag from alkaline solutions of their commonly available salts onto electron-beam patterned thin film masks of poly(methyl methacrylate) (PMMA) spin cast onto single crystal Au(100)/Ag(100)/Si(100) substrates prepared as described above. Nanopillar arrays of small diameter pillars (< 200 nm diameter) (see Fig. 5(d)-(g) of the main text) were formed using 100 nm thick PMMA A2 electron beam resist layers. Nanopillar arrays with larger diameters (see Fig 5(a)-(b) of the main text) were prepared from 200 nm thick PMMA A4 resist layers. The fabrication procedures are described below. Arrays of nanoholes are formed on an electron-sensitive poly(methyl methacrylate) (PMMA) A2 film used as a mask to grow Au nanopillars on a single crystal Au film which was grown on a Ag(100)/Si(100) single crystal substrate, as described. The PMMA A2 film was spin-coated at 1000 rpm to achieve 100nm thickness and was soft baked for 4 minutes on a hotplate at 180°C. Electron beam exposure under conditions of 0.178 nA beam current, 0.1 dose factor x 0.15 pC dot dose exposure were employed to irradiate the PMMA with a Raith e-LiNE lithography tool at 30 μm aperture and 10 kV Extra High Tension (EHT). The exposed regions were developed to remove the electron beam-modified resist and expose the Au(100) surface at the base of each exposed region with a solution of developer (MIBK-IPA 3:1) for 120 s, followed by dipping the sample in isopropyl alcohol (IPA) for 120 s (used as an etch stop) and 120 s hard bake at 100°C on a hotplate. Resist development provided a patterned surface of 125 nm diameter cylindrical pores formed on a 2x2 mm2 Au(100) substrate with a square lattice of period 550 nm. The fabricated arrays are then placed in an alkaline bath containing HAuCl4 (see bath composition employed for planar film deposition above) for 2 minutes at 60°C to yield Au pillars of 70nm height. The sample was then removed, washed for 2 minutes in distilled water, followed by 1 minute in IPA and then placed in acetone for 2 minutes with sonication to remove the PMMA mask. After the PMMA lift-off, the sample was rinsed with water and air dried prior to SEM imaging. Thicker electron beam resist layers were also employed for larger diameter nanohole array masks. Exposure of an electron-sensitive poly(methyl methacrylate) (PMMA) A4 film, deposited at 4000 rpm onto a 1 x 1 cm2 single crystal Au(100) substrate, were used to achieve nominal 200 nm thickness patterned films, prior to 4 minutes of soft bake at 180°C, and exposure using the Raith e-LiNE EBL system. The electron beam exposure was performed at 7 mm working distance, with 20 μm aperture, 20kV extra high tension (EHT) and with area dose of 1.0 x 200 μC/cm2. After the patterning, the PMMA was developed in MIBK-IPA 3:1 for 120 s followed by 120 s of IPA rinse. Nanostructure growth and resist removal were carried out as previously described. Shown below in Fig. S4 are gold nanopillars grown in a nanohole array of height 200 nm, period 700 nm, and nanohole diameter of 350 nm following Au electroless deposition for 5 mins. The SEM image shows a well-formed array of oriented crystalline nanopillars with flat Au(100) top facets. Figure S4. SEM of Au nanopillars (100 nm height, 700 nm period, 350 nm diameter) grown on Au(100) substrate through a nano-electrode array formed with PMMA A4 resist. Heteroepitaxial deposition of silver nanopillars onto Au(100)/Ag(100)/Si(100) substrates was carried out in a similar manner except that nanopillar deposition was carried out using a deposition bath containing an equivalent concentration of AgNO3 rather than HAuCl4 as employed for gold nanopillar deposition and smaller diameter (120 nm) pillars were formed. Thin layer Au capping of the resulting silver nanopillar arrays was carried out by immersing the substrate containing the silver nanopillar array into a HAuCl4-containing bath as described above for 1 min. This yielded a Au capping layer of approximately 10 nm nominal thickness, as determined by SEM pillar diameter measurements before and after the gold capping layer deposition. 2-Dimensional X-ray Diffraction of Au films Au film crystallinity was assessed with a Rapid Axis Rigaku X-ray diffractometer equipped with an area plate detector. The X-ray exposure was carried out at 46 kV voltage and 42 mA current using a Cu Kα source incident on the sample through a 500 μm collimator. The sample stage was fixed at 45° angle for the χ axis, 180° rotation of the φ axis, and oscillation from 205° to 215° of the Ω axis. Figure 2a and 2b show the indexed 2D X-ray diffraction pattern from solution- deposited Au onto single crystal Ag(100)/Si(100) samples from uncontrolled pH solutions of HAuCl4 (Fig 2a) and pH 14 HAuCl4 (Fig 2b) solutions. The diffraction patterns show contributions from the underlying single crystal Si(100) and 500 nm thick Ag(100) layers which appear as well localized diffraction spots, in addition to the nominal 120 nm thickness Au overlayers. Deposition from uncontrolled pH deposition baths result from galvanic replacement and are characterized by polycrystalline Au deposition that shows Au(111) and Au(200) diffraction arcs at constant 2θ diffraction angles (Fig 2a). In contrast, deposition from pH 14 deposition baths yields oriented and aligned Au deposition resulting in well-defined diffraction spots (Fig 2b). Since the lattice constants of Au and Ag are 4.07 Å and 4.08 Å respectively, their diffraction spots are difficult to resolve and appear as overlapping diffraction signals. Nevertheless, their appearance as diffraction spots as opposed to extended diffraction arcs as observed in the case of polycrystalline Au deposition is consistent with substrate-aligned single crystal deposition. Cross-sectional SEM and TEM Analysis Transmission electron microscopy (TEM) was performed using a 200 kV FEI Tecnai Osiris S/TEM to image the crystalline lattice of Au and Ag films. Prior to analysis, a 10 x 6 x 5 μm3 portion of the sample was lifted-out using a FEI Helios focused-ion beam (FIB) tool and secured on a copper-based TEM grid. The sample was thinned to approximately 30nm prior to TEM analysis. A cross-sectional scanning-electron micrograph of a nominal 160 nm thickness Au film, electrolessly deposited onto the Ag(100)/Si(100) substrate is shown in Fig. S5 below. Also evident from the SEM is a top layer of protective platinum deposited with the FIB instrumentation on top of the Au, in order to protect the gold surface during focussed ion beam milling. Figure S5. A cross-sectional SEM image of the solution-deposited Au(100) film on single crystal Ag(100). Surface Roughness Analysis Surface roughness of the solution-deposited, epitaxial gold film was assessed and compared with a thermally evaporated polycrystalline gold surface using a NanoSurface NaioAFM atomic force microscope (AFM). The analysis was carried out over arbitrary 700 x 700 nm2 areas at 10 nN force with 0.4 s time/line scanning speed in contact mode with an AFM tip of force constant 0.1 N/m . The results are shown in Figure S6. b b) ) a a) ) Figure S6. AFM surface topography image of a) solution-deposited, electroless single crystal Au film and b) thermally evaporated, polycrystalline Au film. The area of the scanned regions is approximately 700 x 700 nm2. The area averaged surface roughness (SA) was assessed by the difference in height of each point compared to the arithmetical mean of the surface (𝑆𝑆𝐴𝐴=1𝐴𝐴∬�𝑍𝑍𝑥𝑥,𝑦𝑦�𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) for the imaged regions. SA was found to be 122.2 pm for the solution-deposited, electroless single crystal Au film and 2.84 nm for the physical vapour deposited polycrystalline Au film. Using the tool software, three-dimensional topographic images of both the solution-deposited, and PVD-deposited Au films were also constructed and are shown in Fig. S7. Figure S7. The constructed 3D AFM image of the surface of a) solution-deposited, electroless single crystal Au film and b) thermally evaporated, polycrystalline Au film. Focused-Ion Beam Nano-Patterning The FEI Helios NanoLab 650 dual SEM/Focused-Ion beam (FIB) tool was used to fabricate the nanoscale structures and devices presented in Figures 2 and 3 of the manuscript. Subtractive patterning of mono- and polycrystalline gold films were carried out using the focussed gallium ion beam, employing the tool's pre-set conditions for Au to achieve a desired milling depth of 50 nm following a dose study. The ion beam current was set to 7.7 pA for the 30 kV source voltage. Under these conditions, 50 nm-depth etching was achieved with a dose of 33 pC/μm2 for the evaporated polycrystalline films. This dose had to be doubled to achieve 50 nm-depth milling of the single crystal Au films because of their more uniform and higher packing density of atoms. Milling of the evaporated gold films leads to anisotropic, crystal direction-dependent milling rates, resulting in non-uniform milled regions and poor quality pattern transfer. In contrast, FIB milling of single crystal Au deposited from solution leads to a high degree of uniformity in the milled regions and much improved pattern transfer characteristics. Figure S8 shows a fabricated bowtie antenna on both monocrystalline and polycrystalline Au under the FIB milling conditions described. Figure S8. Top view SEM image of a FIB-milled bowtie nanoantenna fabricated with a) epitaxially-grown solution-deposited monocrystalline Au, and b) thermally evaporated polycrystalline Au. Electron-Beam Lithographed Lines An e-LiNE Raith EBL system was used to pattern lines to fabricate high aspect ratio, single crystal Au nanowires. A PMMA A2 electron beam resist layer was spin-coated at 4000 rpm to achieve 50nm thickness on a thermally evaporated single crystal Ag(100)/Si(100) substrate prepared as described. The PMMA A2 layer was soft baked for 4 minutes at 180°C on a hotplate prior to electron beam exposure. The PMMA film was irradiated a 20kV EHT source, 20μm aperture with 1.6 x 300 pC/cm line exposure factor with 5nm step size at 0.162 nA write current. After the exposure, the substrate was immersed in MIBK:IPA (3:1) for 120 s, followed by 120s IPA rinse and then hard baked at 100°C for 120 s on a hotplate. The exposed Ag regions were then used to grow epitaxial Au nanowire lines by immersing the patterned substrate in the electroless deposition bath for 5 minutes at 60ᵒC. Figure S9 shows a large area SEM image of a portion of the Au lines which were patterned to form a large area concentric square structure capable of acting as a broadband plasmonic nanoantenna. A detailed discussion of the broadband plasmonic response of these structures is beyond the scope of the current manuscript, but will appear in a forthcoming publication. Figure S9. Top view SEM image of epitaxially grown Au lines on a single crystal Ag(100) substrate patterned by EBL and deposited from an alkaline Au(OH)4¯ deposition bath as described. Schottky Diode Fabrication Schottky diodes were fabricated according to the method of Benlamri, M. et al. High breakdown strength Schottky diodes made from electrodeposited ZnO for power electronics applications, ACS Appl. Electron. Mater. 1, 13−17, (2019). The deposition conditions comprised a 0.1 M zinc nitrate hexahydrate (ZnNO3·6H2O) bath prepared by dissolving 1.487 g of ZnNO3·6H2O (reagent grade 98% Sigma-Aldrich) in 50 mL of DI-water maintained at 70°C. A standard 3-electrode cell was employed, where a solution deposited Au(100) substrate was used as the working electrode. The cell also emloyed a Pt wire counter electrode and a Ag/AgCl reference electrode. The deposition of ZnO was carried out at -800 mV for 20 minutes under the described conditions. Following deposition, devices were rinsed in DI-water for 2 minutes and air dried. 250 μm aluminum contacts were deposited by thermal evaporation to provide ohmic contacts. I-V measurements were carried out using a Cascade Microtech probe station (Model: M150). Sheet Resistance The sheet resistance of electroless, solution-deposited epitaxial Au films were measured with a 4P Model 280 4-point probe electrical characterization system and compared with Au films deposited by evaporation, as described. The thickness of films was 100 nm as determined by SEM. At this thickness, the films are expected to display their limiting, bulk resistivity and not be affected by the markedly different electrical properties of the underlying substrates on which they are deposited (see for example, K. L. Chopra, L. C. Bobb, and M. H. Francombe "Electrical Resistivity of Thin Single-Crystal Gold Films", Journal of Applied Physics 34, 1699-1702 (1963)). The measured sheet resistance for the solution-deposited monocrystalline Au film was determined to be 0.023 ± 0.001 Ω/□ while that of the evaporated polycrystalline gold film was determined to be 0.457 ± 0.011 Ω/□ respectively, indicating a greater than 20 times lower resistivity of the single crystal Au film relative to the evaporated polycrystalline Au film. Spectroscopic Ellipsometry Ellipsometry was performed with a Horiba MM-16 Spectroscopic Ellipsometer. Ellipsometry was carried out on 100 nm thick polycrystalline Au films prepared by thermal evaporation, and on 100 nm thick solution-deposited monocrystalline Au films. This thickness is beyond the optical skin depth of gold (approximately 25 nm in the spectral region investigated -- see for example, R. L. Olmon, B. Slovick, T. W. Johnson, D. Shelton, S-H. Oh, G. D. Boreman, and M. B. Raschke, Optical dielectric function of gold, Phys. Rev B, 86, 235147 (2012)). Plotted in Fig. S9 are the real (n) and imaginary (k) parts of the refractive index measured from the mono- and polycrystalline films. Optical absorption, associated with the imaginary part of the refractive index, is observed to be measurably lower for the monocrystalline Au film compared to the polycrystalline Au film at energies below 2.5 eV, the onset of the well-known visible interband optical transition in gold. Figure S10. The real (n) and imaginary (k) parts of the refractive index as determined from spectroscopic ellipsometry of a 100 nm thick polycrystalline Au film deposited by thermal evaporation (blue) and a 100 nm thick, electroless, solution-deposited monocrystalline Au(100) film (red).
1908.06536	1	1908	2019-08-18T23:08:20	Influence of chemical stability on the fabrication of MnGa-based devices	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Ferromagnetic films of L10-ordered MnGa have shown promise not only in the applications in ultrahigh-density magnetic recording and spintronic memories, oscillators, and sensors, but also in controllable studies of novel electrical transport phenomena. However, the stability of MnGa in chemicals and oxygen plasma that are commonly used in the standard micro-/nano-fabrication process has unsettled. In this work, we report a systematic study on the chemical stability of the MnGa films in acids, acetone, ethanol, deionized water, tetramethylammonium hydroxide (TMAOH) and oxygen plasma. We find that MnGa is very stable in acetone and ethanol, while can be attacked substantially if soaked in TMAOH solution for sufficiently long time. Deionized water and acids (e.g., HCl, H3PO4 and H2SO4 solutions) attack MnGa violently and should be avoided whenever possible. In addition, oxygen plasma can passivate the MnGa surface by oxidizing the surface. These results provide important information for the fabrication and the integration of MnGa based spintronic devices.	physics.app-ph	physics	Influence of chemical stability on the fabrication of MnGa-based devices 1. State Key laboratory of Superlattice and microstructures, Institute of Semiconductors, Chinese Academy of Sciences Lijun Zhu1,2* and Jianhua Zhao1 2. Cornell University, Ithaca, New York 14850, USA *[email protected] Ferromagnetic films of L10-ordered MnGa have shown promise not only in the applications in ultrahigh-density magnetic recording and spintronic memories, oscillators, and sensors, but also in controllable studies of novel electrical transport phenomena. However, the stability of MnGa in chemicals and oxygen plasma that are commonly used in the standard micro-/nano-fabrication process has unsettled. In this work, we report a systematic study on the chemical stability of the MnGa films in acids, acetone, ethanol, deionized water, tetramethylammonium hydroxide (TMAOH) and oxygen plasma. We find that MnGa is very stable in acetone and ethanol, while can be attacked substantially if soaked in TMAOH solution for sufficiently long time. Deionized water and acids (e.g., HCl, H3PO4 and H2SO4 solutions) attack MnGa violently and should be avoided whenever possible. In addition, oxygen plasma can passivate the MnGa surface by oxidizing the surface. These results provide important information for the fabrication and the integration of MnGa based spintronic devices. Keyword: Chemical stability, Perpendicular magnetic anisotropy, Spintronics, Wet etching 1. Introduction Ferromagnetic MnGa films with L10 long-range crystalline ordering show promise for both the spintronic technology and correlated electrical transport phenomena [1-3]. It has been established that MnGa films can have giant perpendicular magnetic anisotropy [4,5], ultrahigh coercivity [4,5], low Gilbert damping constant [6], strong magneto- optical Kerr effect [7,8], and high Curie temperature[5]. These fascinating magnetic properties make MnGa films a very compelling candidate material for ultrahigh-density perpendicular magnetic recording, nonvolatile spin-torque magnetic random access memories [9-12], terahertz spin oscillators, and linear magnetic sensors [13]. Meanwhile, the tunable magnetism and structural disorders make MnGa films an excellent playground for novel electrical transport phenomena, e.g., the topological Hall effects due to the stabilization of magnetic skyrmions [14], the orbital two- channel Kondo effects due to the coherent tunneling of two-level systems [15], and the anomalous Hall effect [16]. For the practical application in electronics and transport research, it is crucial to understand the stability of MnGa films in the micro-/nano-fabrication processes. The two key types of structural stability are thermal stability and chemical stability. L10-MnGa films are thermally stable at least at up to 800 K [17] in bulk and at 350 oC in contact with GaAs [5], which make MnGa compatible with the CMOS technology. However, it has remained unsettled as to the chemical stability of MnGa in the commonly used chemicals and oxygen plasma during micro-/nano- fabrication. In this work, we report a systematic study of the chemical stability of MnGa films in the representative chemical solutions that are in micro-/nano-fabrication processes and oxygen plasma. Our results indicate that the chemical stability must be carefully considered during the fabrications of the MnGa-based functional devices. 2. Experiments and discussions 2.1 Growth of MnGa samples For this study, we prepared 30 nm MnGa epitaxial films on semi-insulating GaAs (001) substrate (see Figure 1(a)) by molecular-beam epitaxy [4,5]. The base pressure was below 1×10-9 mbar. A 1.5 nm MgO capping layer was used to protect the MnGa layer from oxidation in the atmosphere [5,18]. The composition was designed to be Mn50Ga50 by carefully controlling the Mn and Ga fluxes during growth and later verified by high-sensitivity x-ray photoelectron spectroscopy (XPS) measurements (Thermo Scientific ESCALAB 250Xi) with Al Kα source and relative atomic sensitivity factors of 13.91 (1.085) for Mn 2p (Ga 3d). As shown in Figure 1(b), MnGa films show asymmetric elemental XPS peaks for both Mn 2p and Ga 3d spectrums. The asymmetry of these peaks should be attributed to the significant shielding effect to the core levels by the high-density at the Fermi level due to the alloy nature of the film, which further reveals the good protection of a 1.5 nm MgO layer from oxidation. 2.2 Chemical stability in acid solutions We first exam the chemical stability of MnGa flowing the workflow in Figure 2(a) in acid solutions that are popular in the wet etching process. A 30 nm MnGa film protected by a 1.5 nm MgO layer was first annealed at 110 oC in vacuum for 20 minutes and then coated with an ultrathin HMDS layer and a 1 µm photoresist of AZ6130. The patterns were the photomask onto the sample surface after the ultraviolet exposure with a Suss MicroTec MA/BA6 Contact Mask Aligner, tetramethylammonium Hydroxide (TMAOH) solution (TMAOH:H2O=4:1), a 10 the development in reasonably transferred from 1 second rinsing in deionized water, drying with nitrogen gas flow, and the removal of the residue resist in oxygen plasma. As a reference, we first performed dry etching with argon ion milling. As indicated by the optical microscopy imaging in Figure 2(b), the dry etching yields reasonably defined patterns with sharp smooth edges after cleaning off the photoresist with acetone and ethanol. This indicates that the argon-ion milling is a highly selective etching process, with the etching rate much faster in the film normal direction than the in-plane direction. However, the wet etching using acid solutions, i.e., HCl, H3PO4 and H2SO4 solutions, result in strong sidewall etching, regardless of the combination of different volume ratio and etching time. As shown in Figure 2(c), all the edges are very rough, and some patterns which were supposed to stay was gone even before the parts which were intended to be etched away still stay on the sample. We also note that some of the connection wires between big pads, which are 10 or 20 µm wide and covered with the resist, were completely etched away even before some of the regions that were not covered with the resist. This observation clearly indicates that the wet etching of MnGa is rather non-uniform with very quick sidewall etching and that the resist seems to be unable to completely protect the MnGa film beneath it, especially in the edge regions. We conclude that acid solutions should be avoided in the fabrication processes of MnGa devices whenever possible. 2.3 Chemical stability in non-acid solutions and oxygen plasma the to clean resist and Now we consider the robustness of the MnGa films against the organic solutions, bases, deionized water, and oxygen plasma that are very popular in the standard semiconductor fabrication process. Acetone and ethanol are the most commonly used organic solutions for sample cleaning and lift-off procedures. A typical base for developing the ultraviolet exposed photoresists, such as positive-tone AZ6130 and negative-tone L300, is the TMAOH solution. Deionized water is widely used to rinse samples after photoresist development and to dilute acid or base solutions. Oxygen plasma is usually introduced to etch the resists after photolithography development. Here we use the resistance (R) of the MnGa layer as a measure of the degree of the attack of the MnGa samples in various types of chemical environment (see Figure 3(a) for a schematic of the measurement geometry). In Figure 3(b) we plot the resistance of the 3 mm-wide and 6 mm-long MnGa film pieces with a 1.5 nm MgO capping layer as a function of soaking time in different chemicals. R remains constant in the acetone and the ethanol solutions, indicating that MnGa is not attacked at all by the acetone and ethanol. In striking contrast, in deionized water R increases dramatically with increasing soaking time and reaches 20 MΩ in 5 minutes, revealing a rather violent reaction between MnGa and H2O. Therefore, the use of chemicals containing water should be avoided or reduced during the device fabrication of the residual the sample MnGa whenever possible. For example, cleaning of the developer off after photolithography development should be done via a quick rinsing rather than a long time soaking in water. In the 75% TMAH developer, the resistance change of the MnGa sample is negligible in the first 3 minutes, while the extended soaking can increase the resistance by a factor of 10 due to a relatively slow attack. This indicates a short development time, e.g., < 1 min, is not a concern for the MnGa devices. Oxygen plasma with the power of 200 W can oxidize the MnGa surface and passivate the surface in the first 1 min, which indicates that influence of the oxygen plasma can be substantial for an ultrathin MnGa film while negligible for relatively thick films. 3. Conclusion We have systematically studied the chemical stability of the MnGa films in acids (e.g., HCl, H3PO4 and H2SO4), acetone, ethanol, deionized water, TMAOH developer, and oxygen plasma. We find that MnGa is very stable in acetone and ethanol, while can be attacked substantially if soaked in TMAOH solution for sufficiently long time. Deionized water and acids attack MnGa violently and should be avoided whenever possible. In addition, oxygen plasma can passivate the MnGa surface by oxidizing the surface. These results provide important information for the fabrication and integration of MnGa based spintronics devices. Acknowledgments the Research The work was supported by the National Program on Key Basic [MOST, Grant Nos. 2018YFB0407601, 2015CB921500], Key Research Project of Frontier Science of Chinese Academy of Science [Grant Nos. QYZDY-SSW-JSC015, XDPB12]. Project References [1] L. J. Zhu and J. H. Zhao, "Perpendicularly magnetized MnxGa films: a kind of promising material for future spintronic devices, magnetic recording and permanent magnets", Appl. Phys. A, 111, 379 (2013). [2] L. J. Zhu, S. H. Nie, and J. H. Zhao, "Recent progress in perpendicularly magnetized Mn-based binary alloy films", Chin. Phys. B 22,118505 (2013). [3] L. J. Zhu, J. H. Zhao, "Electrical Transport of perpendicularly magnetized L10-MnGa and MnAl films", Spin 7, 1730001(2017). [4] Lijun Zhu, Shuaihua Nie, Kangkang Meng, Dong Pan, Jianhua Zhao, and Houzhi Zheng, "Multifunctional L10- Mn1.5Ga Films with giant perpendicular magnetocrystalline anisotropy and large magnetic energy product", Adv. Mater., 24, 4547 (2012). [5] L. J. Zhu, D. Pan, S. H. Nie, J. Lu, and J. H. Zhao, "Tailoring magnetism of multifunctional MnxGa films with coercivity, ultrahigh 2 Appl. Phys. anisotropy", perpendicular giant Lett.,102,132403 (2013). [6] S. Mizukami, F. Wu, A. Sakuma, J. Walowski, D. Watanabe, T. Kubota, X. Zhang, H. Naganuma, M. Oogane, Y. Ando, T. Miyazaki, "Long-Lived Ultrafast Spin Precession in Manganese Alloys Films with a Large Perpendicular Magnetic Anisotropy", Phys. Rev. Lett. 106, 117201 (2011). [7] Kannan M. Krishnan, "Ferromagnetic δ-Mn1−xGax thin films with perpendicular anisotropy", Appl. Phys. Lett. 61, 2365 (1992). [8] L. J. Zhu, L. Brandt, J. H. Zhao, G. Woltersdorf, "Composition-tuned magneto-optical Kerr effect in L10- MnxGa films with giant perpendicular anisotropy", J. Phys. D: Appl. Phys. 49, 245001 (2016). [9] Q. L. Ma, T. Kubota, S. Mizukami, X. M. Zhang, H. Naganuma, M. Oogane, Y. Ando, T. Miyazaki, in L10-MnGa/MgO/CoFeB "Magnetoresistance effect perpendicular magnetic junctions with Co interlayer", Applied Physics Letters 101, 032402 (2012). [10] Q. L. Ma, T. Kubota, S. Mizukami, X. M. Zhang, H. Naganuma, M. Oogane, Y. Ando, and T. Miyazaki," Interface tailoring effect on magnetic properties and their utilization in MnGa-based perpendicular magnetic tunnel junctions", Phys. Rev. B 87, 184426 (2013). [11] K. Z. Suzuki, R. Ranjbar, J. Okabayashi, Y. Miura, A. Sugihara, H. Tsuchiura, S. Mizukami, "Perpendicular junction with a strained Mn-based magnetic tunnel tunnel nanolayer", Sci. Rep. 6, 30249 (2016). [12] Siwei Mao, Jun Lu, Xupeng Zhao, Xiaolei Wang, Dahai Wei, Jian Liu, Jianbai Xia, Jianhua Zhao, "MnGa- based fully perpendicular magnetic tunnel junctions with ultrathin Co2MnSi interlayers", Sci. Rep. 7, 43064 (2017). [13] X. P. Zhao, J. Lu, S. W. Mao, Z. F. Yu, H. L. Wang, X. L. Wang, D. H. Wei, J. H. Zhao, "L10-MnGa based magnetic tunnel junction for high magnetic field sensor", J. Phys. D: Appl. Phys. 50 285002 (2017). [14] K. K. Meng, X. P. Zhao, P. F. Liu, Q. Liu, Y. Wu, Z. P. Li, J. K. Chen, J. Miao, X. G. Xu, J. H. Zhao, and Y. Jiang, "Robust emergence of a topological Hall effect in MnGa/heavy metal bilayers", Phys. Rev. B 97, 060407(R)(2018). [15] Lijun Zhu, Georg Woltersdorf, Jianhua Zhao, "Observation of orbital two-channel Kondo effect in a ferromagnetic L10-MnGa film", Sci. Rep. 6, 34549 (2016). [16] L. J. Zhu, D. Pan, and J. H. Zhao, "Anomalous Hall in epitaxial L10-Mn1.5Ga films with variable effect chemical ordering", Phys. Rev. B 89, 220406 (R) (2014). [17] T. A. Bither, W. H. Cloud, "Magnetic Tetragonal δ Phase in the Mn -- Ga Binary", J. Appl. Phys. 36, 1501 (1965). [18] W. V. Roy, H. Akinaga, S. Miyanishi, K. Tanaka, L. H. Kuo, "Epitaxial MnGa/(Mn,Ga,As)/MnGa trilayers: Growth and magnetic properties", Appl. Phys. Lett. 69, 711 (1996). Figure 1. (a) Schematic of the sample structure. (b) XPS patterns of the MnGa layer, indicating a Mn: Ga atomic ratio of 1:1. 3 Figure 2. (a) Schematic depiction of the etching test workflow: the thin film before photolithography (top), resist patterns after photolithography (middle) and patterns after etching (bottom). Optical microscopy images for the MnGa sample surfaces patterned by (b) dry etching and (c) wet etching. 4 Figure 3. (a) Schematic of the resistance measurement geometry. (b) Evolution of the resistance of a 30 nm MnGa sample in ethanol, acetone, TMAOH developer, deionized water, and oxygen plasma as a function of time. 5
1906.00152	1	1906	2019-06-01T04:41:14	Thermal Conductivity Enhancement by Surface Electromagnetic Waves Propagating along Multilayered Structures with Asymmetric Surrounding Media	[ "physics.app-ph" ]	Enhancement of thermal conductivity via surface electromagnetic waves (SEWs) supported in nanostructures has recently drawn attention as a remedy for issues raised due to the reduction of thermal conductivity in nanoscale confinement. Among them, multilayered structures on a substrate are prevalent in nano-sized systems, such as electronic nanodevices, meaning that analysis on those structures is indispensable. In this work, three basic multilayered structures are selected and the analytical expressions for SEWs supported in each structure are derived. This analytical approach enables us to figure out which factors are crucial for enhancing SEW thermal conductivity using multilayers. It is also found that the solution can be extended to various materials and provide the guidelines on which configurations are desirable for increasing the thermal conductivity. Furthermore, the analytical solutions reduce the calculation time significantly such that the optimal configuration, which can additionally yield SEW thermal conductivity of 1.27 W/m$\cdot$K corresponding to 90\% of the thermal conductivity of bulk glass, is found with the genetic algorithm. This study thus provides a new method for efficiently managing thermal issues in nano-sized devices.	physics.app-ph	physics	Thermal Conductivity Enhancement by Surface Electromagnetic Waves Propagating along Multilayered Structures with Asymmetric Surrounding Media Mikyung Lim1, Jose Ordonez-Miranda2, Seung S. Lee1, Bong Jae Lee1∗, Sebastian Volz3,4∗ 1Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, South Korea 2Institut Pprime, CNRS, Universit de Poitiers, ISAE-ENSMA, F-86962 Futuroscope Chasseneuil, France 3LIMMS/CNRS-IIS(UMI2820), Institute of Industrial Science, University of Tokyo, Tokyo 153-8505, Japan 4Laboratoire d'Energtique Molculaire et Macroscopique, Combustion, UPR CNRS 288 CentraleSuplec, Universit Paris-Saclay, France (Dated: June 4, 2019) Abstract Enhancement of thermal conductivity via surface electromagnetic waves (SEWs) supported in nanostructures has recently drawn attention as a remedy for issues raised due to the reduction of thermal conductivity in nanoscale confinement. Among them, multilayered structures on a sub- strate are prevalent in nano-sized systems, such as electronic nanodevices, meaning that analysis on those structures is indispensable. In this work, three basic multilayered structures are selected and the analytical expressions for SEWs supported in each structure are derived. This analytical approach enables us to figure out which factors are crucial for enhancing SEW thermal conduc- tivity using multilayers. It is also found that the solution can be extended to various materials and provide the guidelines on which configurations are desirable for increasing the thermal con- ductivity. Furthermore, the analytical solutions reduce the calculation time significantly such that the optimal configuration, which can additionally yield SEW thermal conductivity of 1.27 W/m·K corresponding to 90% of the thermal conductivity of bulk glass, is found with the genetic algo- rithm. This study thus provides a new method for efficiently managing thermal issues in nano-sized devices. ∗ [email protected]; [email protected] 1 I. INTRODUCTION Thermal energy transport via surface electromagnetic waves (SEWs) propagating along thin films has been intensively investigated over the last decade [1 -- 6] because of its poten- tial application to compensate the reduction of the thermal performance of systems when their sizes are scaled down to nanoscales [2]. In particular, the booming miniaturization of electronic devices has been leading to a decrease of the size of their components to a few nanometers, which is smaller than the mean free path of the electrons or phonons of the constituting materials and reduces their effective thermal conductivities [7]. Given that the problem of heat spreading can generate hot spots, which overheat the electronic nanodevices and deteriorate their performance [8], the additional energy transport mechanism driven by SEWs supported in nano-sized systems has been regarded as a promising remedy to tackle this issue of overheating [1 -- 6]. SEWs are bounded at the interface along which they propagate carrying energy [9 -- 12]. The propagation length of these SEWs can be drastically increased as the thickness of the suspended glass layer decreases [1, 3]; however, when this single glass layer is placed on a substrate (i.e., asymmetric surrounding media), the frequency interval where SEWs can be supported becomes severely restricted [3], and thus, an additional energy transport via SEWs is rather limited. Furthermore, besides the single glass layer, a thermal conductivity enhancement by SEW within mutilayered structures should also be considered for wide-range of applications. Although SEWs supported in periodic multilayered structures has been widely inves- tigated [13 -- 19], their contribution on thermal conductivity has been rarely reported. A previous work [4] explored the thermal conductivity enhancement of a suspended nano- layered system due to the propagation of SEWs by regarding nano-layers as a single layer with the effective permittivity. Because this effective permittivity only depends on a rel- ative thickness of the constituting layers, it cannot fully take account of the configuration of the nano-layers. Accordingly, this means cannot be safely extended when a-few-layer structure on a substrate, which is more practical configuration than the suspended periodic multilayered structure, is considered. In this work, various a-few-layer structures which are on a substrate is evaluated and the best structure which can have the largest SEW thermal conductivity is discussed. The exact 2 propagation length of the SEWs supported in each configuration is numerically obtained first, and the approximate analytical solution is derived for each configuration. Then, based on this analytical description, the effects of the constituting layers and the configurations on the propagation length of the SEWs are thoroughly explored to achieve a significant enhancement in SEW thermal conductivity. Finally, the configuration is optimized with the genetic algorithm to maximize the SEW thermal conductivity by employing the analytical solutions. Depending on their nature, the SEWs have been classified into the following modes: surface phonon (plasmon) polaritons, Zenneck modes, and transverse magnetic (TM) guided modes [11, 20]. Among these modes, focus usually lied on surface polaritons [9] because they can significantly enhance the tunneling of evanescent waves across a vacuum gap and their dispersion relation can be tailored by the structure, and thus, can be employed in controlling near-field thermal radiation [21 -- 23], near-field thermophotovoltaics [15, 24] and microscopy [25]. However, a recent experimental study [20] showed that the Zenneck modes can provide longer propagation lengths in wider frequency intervals than surface polaritons, which means that Zenneck modes are more beneficial in carrying energy than these latter polaritons. Throughout the manuscript, both Zenneck modes and surface polaritons will be analyzed, while the TM guided modes will be not considered, because they are not coupled to the thermal sources. II. THERMAL CONDUCTIVITY MODELING AND DISPERSION RELATION The effective thermal conductivity, k, of layered structures due to the propagation of SEWs is given by [1, 3]: k = 1 4πd Z ∞ 0 ωΛβR ∂f0 ∂T dω = Z ∞ 0 kωdω (1) where is the Planck constant divided by 2π, ω is the angular frequency, d is the total thickness of the multilayer with average temperature T , and f0 is the Bose-Einstein dis- tribution function. The values of the unknown parameters βR = ℜ(β) and propagation length Λ = 1/(2ℑ(β)) can be obtained respectively from the real and the imaginary parts of the in-plane wavevector β = βR + iβI , which is defined by the dispersion relation of the SEWs propagating in a given layered structure. Considering that Λ cannot be longer than 3 FIG. 1. Schematics of the layered structures that are considered in this work: (a) GD (glass- dielectric), (b) DG (dielectric-glass), and (C) GDG (glass-dielectric-glass) deposited on a semi- infinite substrate. Volume filling factor of glass is kept constant for the three structures. Subscripts g, d, and s refer to the glass, dielectric, and substrate, respectively where ǫn and dn are the dielectric function and thickness of layer n. the system size L, throughout the manuscript, the thermal conductivity in Eq. (1) is then calculated with Λeff which is defined as 1/Λeff = 1/Λ + 1/L. The frequency interval for integration is taken as that for Zenneck modes and surface phonon polations [20]. In Eq. (1), the spectral thermal conductivity, kω is proportional to a product of ω ∂f0 Given that the value of ω ∂f0 ∂T and ΛβR. ∂T is larger at lower frequencies, the structure should support SEWs with large propagation length at low frequency to have large thermal conductivities. A. Glass-Dielectric structures Let us first consider the Glass-Dielectric (GD) structure (see Fig. 1(a)) which supports the propagation of the SEWs with the following dispersion relation [4, 26]: tanh(pddd) + αsd 1 + αsdtanh(pddd) = −αgd tanh(pgdg) + α0g 1 + α0gtanh(pgdg) (2) where αij = ǫipj/ǫjpi. Here, ǫi refers to the dielectric function of each layer and the transverse wavevector pi is given by p2 i = β2−ǫik2 0 with k0 = ω/c0 being the wavevector in vacuum. The subscripts 0, g, d, and s stand for the vacuum, glass, dielectric, and substrate, respectively and c0 is the speed of light in vacuum. The solution of Eq. (2) for the complex wavevector β is obtained numerically by the secant method with a proper choice of the initial point and using the glass dielectric function reported in [27]. The dielectric function of the substrate 4 is assumed to be ǫs=1.24 throughout this work, as done in [3, 4]. The wavevector βR and propagation length of SEWs are shown in Figs. 2(a) and 2(b), for different values of the dielectric function, ǫd. Note that βR is close to the light line of substrate (i.e., βR ∼ √ǫsk0) except for the high frequencies (ω > 200 Trad/s). In this work, this photon-like mode is only considered, because it has much larger propagation lengths than phonon-like mode, which is found to have propagation length smaller than 1 µm, as in Ref. [1]. It can be seen that a frequency interval where solution exists, varies with ǫd. We set the lowest value of frequency, where solution can exist, as a cutoff frequency, ωc for each case. Estimating this ωc for different configurations is crucial because at ωc, SEWs with the longest propagation length can be supported in the structures as can be seen in Fig. 2(b). When the system size is taken to be 10 cm, the corresponding spectral thermal conductivity, kω, for each case is plotted in Fig. 2(c), which shows that kω reaches its highest (a) Zc (Hd = 4.0) Zc (Hd = 8.0) Zc (Hd = 12.0) Zc (Hd = 16.0) (c) Light line of substrate (b) (d) Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 FIG. 2. Dispersion relation and SEW thermal conductivity of the GD structure with dielectric layers of various permittivities. The thicknesses of the glass and dielectric layers are set to 20 nm and 680 nm, respectively (total thickness d = 700 nm) and average temperature T is set to 300 K. Frequency dependent (a) βR, (b) propagation length, and (c) spectral thermal conductivity when system size is set to be 10 cm. (d) Resulting thermal conductivity depending on the size of the system. 5 value at ωc for each case. Although the propagation length is smaller for higher values of ǫd at the same frequency, the wavevector can be extended to lower frequencies with higher values of ǫd [refer to Fig. 2(b)]. As a result, in Fig. 2(d), when ǫd = 16.0 and system size is set to 10 cm, the corresponding thermal conductivity is 0.52 W/m·K, which corresponds to 37% of the thermal conductivity of bulk glass. Although the solutions shown in Fig. 2 can be obtained numerically, it is hard to see which factors affect the propagation length or frequency interval. Alternatively, an analytical solution, which may provide a physical insight on which structures are beneficial for enhancing SEW heat transfer, is derived with several approximations. Given that the propagation length of SEWs decreases as the thickness of the glass layer increases [1, 3], the case of glass with very small thickness will be considered to derive an analytical solution of Eq. (2). Under this thin-film approximation (i.e., pgdg << 1) and up to a first-order on pgdg << 1, Eq. (2) can be expressed as follows: pgdg + (ǫ0pg/ǫgp0) 1 + (ǫ0pg/ǫgp0)pgdg A + (ǫspd/ǫdps) 1 + A(ǫspd/ǫdps) ǫgpd ǫdpg = − (3) where A = tanh(pddd). The solution of Eq. (3) for ps can be expressed as: ps = − ǫspd[Aǫg(ǫ0 + dgǫgp0)pd + ǫd(ǫgp0 + dgǫ0p2 g)] ǫd[ǫg(ǫ0 + dgǫgp0)pd + Aǫd(ǫgp0 + dgǫ0p2 g)] (4) approximate as p2 Considering that ps → 0 when βR → √ǫsk0 and βI → 0, p2 0 can be 0. Taking into account the condition ℜ(pi) > 0 to ensure the bounding of SEW to the interfaces, pi ≈ √ǫs − ǫik0 = Dik0. The application of pi ≈ Dik0 s + (ǫs − ǫi)k2 i ≈ (ǫs − ǫi)k2 i = p2 to Eq. (4) provides the approximate solution as ps = − ǫsDdk0[A′Ddǫg(ǫ0 + D0dgǫgk0) + ǫd(D0ǫg + dgD2 gǫ0k0)] ǫd[Ddǫg(ǫ0 + D0dgǫgk0) + A′ǫd(D0ǫg + dgD2 gǫ0k0)] (5) where A′ = tanh(Ddk0dd). Figures 3(a) and 3(b) show the analytical solution for βR and propagation length, obtained using the relation β2 = p2 s + ǫsk2 0. Although several approx- imations were made, it is readily seen that the approximate solution agrees reasonably with the numerical one. In order to analyze the propagation length and frequency interval where the SEWs can exist, the analytical solution for βI is also derived. From the rela- tion β2 = p2 s + ǫsk2 0, when βR → √ǫsk0, βI can be approximated as (ps,Rps,I)/√ǫsk0 [3]. Here, ps,R and ps,I are the real and imaginary parts of ps, respectively. Note that D0 is 6 purely real and Dd is purely imaginary, such that A′ = tanh(Ddk0dd) can be expressed as A′ = itan(Dd,Ik0dd) = iA′ I with Dd = iDd,I. Equation (5) can then be re-written as: ps = − ǫsDd,Ik0[−A′ IDd,Iǫg(ǫ0 + D0dgǫgk0) + ǫd(D0ǫg + dg(ǫs − ǫg)ǫ0k0)] ǫd[Dd,Iǫg(ǫ0 + D0dgǫgk0) + A′ Iǫd(D0ǫg + dg(ǫs − ǫg)ǫ0k0)] (6) where ǫg = ǫg,R + iǫg,I. After conducting the complex expansion of Eq. (6), ps,R and ps,I can be obtained. Using the relaton βI ≈ (ps,Rps,I)/√ǫsk0 and assuming that pgdg is small, the approximate explicit expression for βI can be obtained as follows: βI,GD ≈ D3 d,Idgǫg,Iǫ1.5 s (1 + A′2 I )(A′ IDd,Iǫ0 − D0ǫd)[D2 0(ǫ2 g,R + ǫ2 g,R + ǫ2 g,I) ǫd(Dd,Iǫ0 + A′ ID0ǫd)3(ǫ2 g,I) + ǫ2 0ǫs]k2 0 (7) The predictions of Eq. (7) for βI,GD are shown in Fig. 3(c) in comparison with the numeri- cal solution obtained from Eq. (2). Although Eq. (7) has been derived through different as- sumptions, negligible discrepancy between the analytical and numerical solutions is observed except for higher frequency regime where spectral thermal conductivity has smaller values. Taking into account that Dd,I > 0, dg > 0, ǫs > 0, ǫd > 0, and ǫg,I > 0, βI,GD is positive only when (A′ IDd,Iǫ0−D0ǫd)/(Dd,Iǫ0 +A′ ID0ǫd) > 0. In other words, SEWs propagating along the GD structure exist for frequencies satisfying this latter condition, which varies with ǫd, as can be seen in Fig. 3(c). Higher values of ǫd satisfy (A′ IDd,Iǫ0− D0ǫd)/(Dd,Iǫ0 + A′ ID0ǫd) > 0 from lower frequency, such that the GD structure with higher ǫd can have higher SEW thermal conductivity, as shown in Fig. 2(d). Let us see how the other factors affect the propagation length. If we change dg under the assumption that pgdg is small, as in Fig. 4(a), the propagation length increases as the dg decreases, and the frequency interval where solution exists does not change much by changing dg. This trend agrees well with the prediction by Eq. (7) as βI,GD is proportional to dg and (A′ IDd,Iǫ0 − D0ǫd)/(Dd,Iǫ0 + A′ ID0ǫd) is not a function of dg. Reminding that A′ I = tan(Dd,Ik0dd), the frequency interval where solution exists increases as the dd increases. It should be noted that when dd = 580 nm, the solution in Eq. (7) exists from relatively high frequencies where the spectral thermal conductivity has small values. It is also worthwhile to mention that for dd = 380 nm and ǫd = 12.0, the solution does not exist within the frequency range where SiO2 can emit. Thus, the proper choice of the dielectric layer thickness is required to exploit the SEWs for carrying energy. Given that the term (A′ IDd,Iǫ0 − D0ǫd)/(Dd,Iǫ0 + A′ ID0ǫd) does not involve the dielectric function of glass ǫg, the frequency interval where the solution exists remains almost invariant 7 (a) (b) (c) Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 Hd = 16.0 Hd = 16.0 Hd = 12.0 Hd = 12.0 Hd = 8.0 Hd = 8.0 Hd = 4.0 Hd = 4.0 Hd = 16.0 Hd = 12.0 Hd = 8.0 Hd = 4.0 FIG. 3. (a) Dispersion relation and (b)-(c) propagation length of SEWs in GD structure obtained via numerical method and analytical solution. Analytical solution is calculated using Eq. (5) for (a)-(b) and Eq. (7) for (c). The conditions are the same as in Fig. 2. under the changes of ǫg, as can be seen in Figs. 4(b) and 4(c). Although a slight mismatch between analytical and numerical solutions is found due to large absolute values of the dielectric function of SiC, ǫSiC, the frequency interval does not change much, even if the glass is replaced by SiC or MgF2 in contrast to the notable difference in frequency interval 8 (a) (b) (c) dg = 20 nm, dd = 680 nm dg = 10 nm, dd = 680 nm dg = 50 nm, dd = 680 nm dg = 20 nm, dd = 580 nm dg = 20 nm, dd = 780 nm Hg = HSiO2 Hg = HSiO2/2.0 Hg = HSiO2(cid:9)2.0 Hg = HSiC Hg = HMgF2 Hd = 12.0 Hg = HSiO2 Hg = HSiO2/2.0 Hg = HSiO2(cid:9)2.0 Hg = HSiC Hg = HMgF2 Hd = 16.0 FIG. 4. The propagation length of SEWs propagating along the GD structure with different configurations, (a) when the thickness of each layer varies while the dielectric function of the dielectric layer ǫd remains as 12.0, (b) and (c) when the dielectric function of the glass layer varies while the thicknesses of glass (SiC or MgF2) layer and dielectric layer are set to 20 nm and 680 nm, respectively. The dielectric function of the dielectric layer is (b) 12.0 and (c) 16.0. 9 FIG. 5. Propagation length of the SEWs propagating along the GD and DG structures. Calcu- lations were carried out with thicknesses of the glass and dielectric layers of 20 nm and 680 nm, respectively. The dielectric function of the dielectric layer is set to 12.0. when ǫd is varied from 12.0 to 16.0 (see Figs. 4(b) and 4(c)). B. Dielectric-Glass structure In this section, the Dielectric-Glass (DG) structure introduced in Fig. 1(b) is analyzed to determine its performance for carrying thermal energy with SEWs. As in Eq. (2), the dispersion relation for the structure DG can be written as follows [4, 26]: tanh(pgdg) + αsg 1 + αsgtanh(pgdg) = −αdg tanh(pddd) + α0d 1 + α0dtanh(pddd) (8) where all the parameters have been previously defined. As in Eqs. (3)-(7), the simplification of Eq. (8) under the assumption of pgdg << 1 yields the following βI for the structure DG: (9) Dd,Idgǫg,Iǫ1.5 s (A′ ǫ3 d(Dd,Iǫ0 + A′ IDd,Iǫ0 − D0ǫd)BDGk2 ID0ǫd)3(ǫ2 g,R + ǫ2 g,I) βI,DG ≈ 0 where BDG = ǫ2 d(ǫd − ǫs)[D2 0(ǫ2 g,R + ǫ2 g,I) + ǫ2 0ǫs] + A′2 I [ǫ2 0(ǫ2 g,R + ǫ2 +2A′ ID0Dd,Iǫ0ǫd[ǫ2 g,I)(ǫd − ǫs)2 + D2 0ǫ4 g,R + ǫ2 dǫs − (ǫd − ǫs)(ǫ2 dǫs] g,I)] The difference between βI,DG and βI,GD can then be expressed as: βI,DG − βI,GD = Dd,Idgǫg,Iǫ1.5 s (A′ ǫ3 d(Dd,Iǫ0 + A′ IDd,Iǫ0 − D0ǫd)Ck2 ID0ǫd)3(ǫ2 g,I) g,R + ǫ2 0 10 (10) (11) where C = A′ I{2D0Dd,Iǫ0ǫd + A′ I[ǫ2 d(ǫs − ǫ0) + ǫ2 0(ǫs − ǫd)]}[ǫ2 dǫs − (ǫd − ǫs)ǫg2] (12) I{2D0Dd,Iǫ0ǫd − A′ Note that the SEWs propagating along the GD structure exists only when (A′ IDd,Iǫ0 − D0ǫd)/(Dd,Iǫ0 + A′ ID0ǫd) > 0, as well as when Dd,I > 0, dg > 0, ǫs > 0, ǫd > 0, and ǫg,I > 0. Accordingly, the sign of βI,DG − βI,GD is the same as the sign of C described in Eq. dǫs − (ǫd − ǫs)ǫg2 can both (12). Thus, A′ act as indicators for determining which structure supports SEWs with longer propagation 0(ǫs − ǫd)]} is found to remain positive in the frequency interval where the solution exists. Therefore, when dǫs−(ǫd−ǫs)ǫg2 < 0, βI,DG has smaller value than βI,GD, meaning that ΛDG > ΛGD. Except ǫ2 this frequency interval, it can be readily noted in Fig. 5 that the GD structure supports SEWs length. In Fig. 5, for given configuration, A′ I{2D0Dd,Iǫ0ǫd − A′ 0(ǫs − ǫd)]} and ǫ2 d(ǫs − ǫ0) + ǫ2 d(ǫs − ǫ0) + ǫ2 I[ǫ2 I[ǫ2 with longer propagation lengths. For this configuration, because GD structure supports SEWs with longer propagation lengths at lower frequency interval, GD structure can have higher SEW thermal conductivity. However, when it is applied to other materials, it should dǫs − (ǫd − ǫs)ǫg2 > 0 hold, either GD or be noted that depending on the condition where ǫ2 DG structure can be beneficial in carrying thermal energy. C. Glass-Dielectric-Glass structures For the three-layer structure shown in Fig. 1(c), the following dispersion relation can be derived from the Maxwell equation with proper boundary conditions [26, 28]: tanh(pddd) = − αdg[1 + αsgtanh(pg α2 dg[1 + αsgtanh(pg dg 2 )][α0g + tanh(pg dg 2 )][1 + α0gtanh(pg dg 2 )] + αdg[αsg + tanh(pg dg 2 )] + [αsg + tanh(pg dg 2 )][1 + α0gtanh(pg dg dg 2 )][α0g + tanh(pg 2 )] dg 2 )] where all the parameters have been previously defined. As in previous sections, an ap- proximate analytical solution for βI of SEWs supported in Glass-Dielectric-Glass (GDG) (13) structure can be expressed as: βI,GDG ≈ Dd,Idgǫg,Iǫ1.5 s (A′ 2ǫ3 d(Dd,Iǫ0 + A′ IDd,Iǫ0 − D0ǫd)BGDGk2 ID0ǫd)3(ǫ2 g,R + ǫ2 g,I) 0 (14) 11 where BGDG = 2ǫ2 d(ǫd − ǫs)[ǫ2 0ǫs + (ǫ2 g,I)(ǫd − ǫs) + ǫ2 g,R + ǫ2 +A′2 I (ǫ0 − ǫd)[(ǫ2 g,R + ǫ2 g,I)(−ǫ0 + ǫs)] dǫs][ǫ0ǫd − (ǫ0 + ǫd)ǫs] g,I)] g,R + ǫ2 dǫs − (ǫd − ǫs)(ǫ2 +2A′ ID0Dd,Iǫ0ǫd[ǫ2 (15) In order to compare the SEW thermal conductivity of the GDG structure with those of the GD and DG structures, βI,DG − βI,GDG is calculated. Interestingly, the resulting value is found to be half of the value of βI,DG − βI,GD. βI,DG − βI,GDG = Dd,Idgǫg,Iǫ1.5 2ǫ3 s (A′ d(Dd,Iǫ0 + A′ IDd,Iǫ0 − D0ǫd)Ck2 0 I D0ǫd)3(ǫ2 g,I) βI,DG − βI,GD g,R + ǫ2 = 2 (16) FIG. 6. (a) Propagation length of the SEWs supported by the GD, DG, and GDG structures. The total thickness of glass layer is set to 20 nm and the thickness of dielectric layer is 680 nm. The dielectric function of the dielectric layer is set to 12.0. (b) Resulting thermal conductivities of each structure when T = 300 K. 12 FIG. 7. Propagation length of the SEWs supported in GD, DG, and GDG structures. The total thickness of the SiC/MgF2 layer is 20 nm and the thickness of the dielectric layer is 680 nm. The dielectric function of the dielectric layer is set to 12.0 (a) Glass is replaced by SiC. (b) Glass is replaced by MgF2. Accordingly, βI,DG − βI,GDG = βI,GDG − βI,GD. Furthermore, both βI,DG − βI,GDG and βI,GDG − βI,GD have positive values when ǫ2 dǫs − (ǫd − ǫs)ǫg2 > 0 as the βI,DG − βI,GD does. When ǫ2 dǫs−(ǫd−ǫs)ǫg2 > 0, it is readily observed in Fig. 6(b) that ΛGD > ΛGDG > ΛDG and vice versa. Among the three structures, the GD one shows the highest SEW conductivity when the maximum system size is set to be 10 cm, because it supports SEWs with the longest propagation length at the lowest frequency regime. 13 III. DISCUSSION As discussed in the previous section, ǫ2 dǫs − (ǫd − ǫs)ǫg2 can act as the indicator for determining the best structure for carrying thermal energy with SEWs. The expression when ǫd − ǫs > 0. Therefore, we can infer that if the glass layer is replaced by a material which has a dielectric function dǫs − (ǫd − ǫs)ǫg2 > 0 can be re-written as ǫg2 < ǫ2 ǫ2 dǫs ǫd−ǫs with small absolute value, the propagation length for the GD structure will be the longest among three structures; otherwise, the SEWs for DG structure will have the longest. In Fig. 7, the propagation lengths for GD, DG, and GDG structures are shown, when glass layers are replaced by SiC and MgF2. As predicted, for SiC having a higher dielectric function than glass at lower frequency regime, DG structure shows the longest propagation length at the lowest frequency where SEWs can exist. On the other hand, for MgF2, ǫ2 dǫs − (ǫd − ǫs)ǫMgF22 > 0 holds for the entire frequency interval, and thus, ΛGD > ΛGDG > ΛDG can be observed for the frequency interval where the SEWs exist. Although we have simplified the analytical solutions for various structures, those solutions can be applied to a wide range of materials and they enable us to predict the structure that holds SEWs with the longest propagation lengths. In addition, it is worthwhile to mention that calculation of SEW thermal conductivity with the analytical solution requires significantly reduced calculation time compared to that for numerical solution. Accordingly, the optimal configurations for maximizing SEW ther- mal conductivity can be found by employing genetic algorithm. As in Table 1, the maximum calculated SEW thermal conductivity of 0.974 W/m·K, which is 70% of the thermal conduc- tivity of bulk glass, is obtained with GD structure with ǫs of 1.24. For that configuration, the numerical exact solution results in SEW thermal conductivity of 1.00 W/m·K. In other words, the maximum thermal conductivity can be obtained with error less than 3% in much shorter calculation time by employing analytical solution. While varying the range for di- electric function of dielectric layer and substrate and maximum system size, the error is kept less than 5%. The analytical solution enables us to conduct multiple optimization processes under various boundary conditions with high accuracy. As a result, in Table 1, it is read- ily noted that thinner glass layer, larger dielectric function of dielectric layers, and smaller dielectric function of substrate are beneficial for enhancing SEW thermal conductivity. 14 TABLE I. Bounds for the variables and corresponding optimal configuration and SEW thermal conductivity for GD structure at T = 300 K Variable Lower bound Upper bound Optimal SEW thermal configuration conductivity [error] dg (nm) dd (nm) ǫd ǫs dg (nm) dd (nm) ǫd ǫs dg (nm) dd (nm) ǫd ǫs 20 20 1 - 5 5 1 - 5 20 1 1.1 1000 1000 16 - 1000 1000 16 - 1000 1000 16 16 20 695 16 set: 1.24 5 705 16 set: 1.24 5 600 16 1.1 (W/m·K) 0.512 [1.8%] 0.974 [2.7%] 1.27 [4.5%] IV. CONCLUSIONS In this work, the dispersion relations of SEWs propagating along the GD, DG, and GDG structures have been derived and analyzed along with their corresponding analytical solutions for the in-plane wavevector. The impact of various factors affecting the propagation length and thermal conductivity has been discussed based on the analytical solutions. It has been found that the frequency interval where the SEWs exist is almost independent on the dielectric function of the glass layer. Furthermore, the SEW propagation length of each structure has been compared at different conditions and the indicator for determining the structure where SEWs with the longest propagation length can be supported has been found. The propagation lengths of structure GDG is found to be between those of the structures DG and GD. The optimization to get maximum SEW thermal conductivity is also conducted with genetic algorithm, yielding SEW thermal conductivity of 1.27 W/m·K 15 similar to the thermal conductivity of bulk glass. It is worthwhile to mention that thermal energy transport by SEWs has a different nature than conventional thermal conduction; that is, its relatively long propagation length indicates that the energy is weakly absorbed by the substrate. Accordingly, it is a quite effective channel to spread heat from hot spots. Because Si, which is the most widely used material in semiconductor industry, has a dielectric function around 11.7 at 100 Trad/s and is almost lossless [27], the analysis conducted in this work provides guidelines for enhancing SEW thermal conductivity of nano-sized glass layers combined with Si ones, mitigating the hot-spot issue in electronic devices. ACKNOWLEDGMENTS This research was supported by the Basic Science Research Program (Grants No. NRF- 2017R1A2B2011192 and NRF-2019R1A2C2003605) through the National Research Foun- dation of Korea (NRF) funded by Ministry of Science and ICT. The stay of M. Lim in CentraleSuplec has been supported by the Erasmus Mundus EASED programme (Grant 2012-5538/004-001) coordinated by CentraleSuplec. [1] D.-Z. A. Chen, A. Narayanaswamy, and G. Chen, Phys. Rev. B 72, 155435 (2005). [2] D.-Z. Chen and G. Chen, Front. Heat Mass Transf. 1 (2010). [3] J. Ordonez-Miranda, L. Tranchant, T. Tokunaga, B. Kim, B. Palpant, Y. Chalopin, T. Antoni, and S. Volz, J. Appl. Phys. 113, 084311 (2013). [4] J. Ordonez-Miranda, L. Tranchant, Y. Chalopin, T. Antoni, and S. Volz, J. Appl. Phys. 115, 054311 (2014). [5] J. Ordonez-Miranda, L. Tranchant, B. Kim, Y. Chalopin, T. Antoni, and S. Volz, Phys. Rev. Lett. 112, 055901 (2014). [6] L. Tranchant, J. Ordonez-Miranda, T. Matsumoto, S. Gluchko, T. Antoni, S. Volz, and K. Miyazaki, in Thermal Investigations of ICs and Systems (THERMINIC), 2015 21st Inter- national Workshop on (IEEE, 2015) pp. 1 -- 5. [7] Z. M. Zhang, Nano/microscale heat transfer (McGraw-Hill Professional, 2007). [8] O. Semenov, A. Vassighi, and M. Sachdev, IEEE Trans. Device Mater. Rel. 6, 17 (2006). 16 [9] H. Raether, Surface Plasmons on Smooth and Rough Surfaces and on Gratings (Springer- Verlag, 1988). [10] J. J. Burke, G. I. Stegeman, and T. Tamir, Phys. Rev. B 33, 5186 (1986). [11] F. Yang, J. R. Sambles, and G. W. Bradberry, Phys. Rev. B 44, 5855 (1991). [12] T. Kume, T. Kitagawa, S. Hayashi, and K. Yamamoto, Surf. Sci. 395, 23 (1998). [13] A. Dereux, J.-P. Vigneron, P. Lambin, and A. A. Lucas, Phys. Rev. B 38, 5438 (1988). [14] J. Mendialdua, A. Rodriguez, M. More, A. Akjouj, and L. Dobrzynski, Phys. Rev. B 50, 14605 (1994). [15] M. Lim, J. Song, J. Kim, S. S. Lee, I. Lee, and B. J. Lee, J. Quant. Spectrosc. Radiat. Transf. 210, 35 (2018). [16] P. Ben-Abdallah, K. Joulain, and A. Pryamikov, Appl. Phys. Lett. 96, 143117 (2010). [17] S.-A. Biehs, M. Tschikin, R. Messina, and P. Ben-Abdallah, Applied Physics Letters 102, 131106 (2013). [18] Y. Guo, C. L. Cortes, S. Molesky, and Z. Jacob, Appl. Phys. Lett. 101, 131106 (2012). [19] R. Wang, J. Chen, Y. Xiang, Y. Kuai, P. Wang, H. Ming, J. R. Lakowicz, and D. Zhang, Phys. Rev. Applied 10, 024032 (2018). [20] S. Gluchko, B. Palpant, S. Volz, R. Braive, and T. Antoni, Appl. Phys. Lett. 110, 263108 (2017). [21] M. Lim, J. Song, S. S. Lee, and B. J. Lee, Nat. Commun. 9, 4302 (2018). [22] B. Song, D. Thompson, A. Fiorino, Y. Ganjeh, P. Reddy, and E. Meyhofer, Nat. Nanotechnol. 11, 509 (2016). [23] H. Iizuka and S. Fan, Phys. Rev. Lett. 120, 063901 (2018). [24] R. Messina and P. Ben-Abdallah, Sci. Rep. 3, 1383 (2013). [25] Y. De Wilde, F. Formanek, R. Carminati, B. Gralak, P.-A. Lemoine, K. Joulain, J.-P. Mulet, Y. Chen, and J.-J. Greffet, Nature 444, 740 (2006). [26] C. Yeh and F. I. Shimabukuro, The Essence of Dielectric Waveguides (Springer, 2008). [27] E. D. Palik, Handbook of Optical Constants of Solids, Vol. 1 (Academic Press, New York, 1985). [28] C. Yeh and G. Lindgren, Appl. Opt. 16, 483 (1977). 17
1810.00221	1	1810	2018-09-29T15:19:32	Fractal-Based Electrolytic Capacitor Electrodes: Scaling Behavior with Respect to Fractal Order and Complexity	[ "physics.app-ph" ]	In past decades, the application of fractals to electrode design for enhanced signaling and electrochemical performance was a popular subject and enabled the growth of consumer micro-electronics. Supercapacitors, which are energy storage devices with many promising characteristics, have largely grown alongside of such developments in electronics, but little work has been done to use fractal electrodes in supercapacitors. In this work, plane-filling and fractal patterns were used in designing laser scribed graphene supercapacitor electrodes, allowing the scaling laws of capacitance with respect to fractal order and complexity to be examined for the first time. An interesting exponential relationship between capacitance and fractal order for the more open structured fractals was observed, the exponent of which was proportional to the Hausdorff dimension. Additional non-linear relationships between capacitance and order were observed for other structures which was correlated with inter-plate repulsion and differences in path length. These findings provide the first step in maximizing the efficiency of fractal-based electrolytic devices by exploring the non-intuitive trends in capacitance with respect to fractal order and complexity.	physics.app-ph	physics	Fractal-Based Electrolytic Capacitor Electrodes: Scaling Behavior with Respect to Fractal Order and Complexity Benjamin Barnes1,2, Othman Suleiman1, JeanPaul Badjo1 and Kausik S Das1 1University of Maryland Eastern Shore, 1, Backbone Road, Princess Anne, MD 21853 USA 2Department of Chemistry and Biochemistry, University of Maryland College Park, MD 20742 USA In past decades, the application of fractals to electrode design for enhanced signaling and electrochemical performance was a popular subject and enabled the growth of consumer micro- electronics. Supercapacitors, which are energy storage devices with many promising characteristics, have largely grown alongside of such developments in electronics, but little work has been done to use fractal electrodes in supercapacitors. In this work, plane-filling and fractal patterns were used in designing laser scribed graphene supercapacitor electrodes, allowing the scaling laws of capacitance with respect to fractal order and complexity to be examined for the first time. An interesting exponential relationship between capacitance and fractal order for the more open structured fractals was observed, the exponent of which was proportional to the Hausdorff dimension. Additional non-linear relationships between capacitance and order were observed for other structures which was correlated with inter-plate repulsion and differences in path length. These findings provide the first step in maximizing the efficiency of fractal-based electrolytic devices by exploring the non-intuitive trends in capacitance with respect to fractal order and complexity. Due to their rapid charge/discharge rate, long operational lifetime1, and potentially flexible and environmentally benign designs2, electrolytic supercapacitors (ESCs) are poised to become an integral component of a range of electronic systems ranging from the acceleration boosters in electric cars to flexible cellular phones3. The primary focus of recent ESC research has been in developing new active materials, which include activated carbon,1 carbon nanotubes 4,5, carbon nano onions6, and carbide-derived carbons6. Graphene, however, received the greatest attention6,7,8,9,10 due to its high surface area and conductivity and chemical stability 9. A number of recent papers have propelled the method of fabricating ESCs from laser-reduced graphene oxide (GO) films into the realm of commercial feasibility11,12. This method is attractive as it serves to both exfoliate and reduce the GO resulting in the liberation of oxygen as CO, CO2, and H2O gas11,12. By tuning such parameters as laser intensity, scribing speed, and substrate temperature, the conductivity and overall quality of the laser scribed graphene (LSG) layer can be precisely controlled12,13,14,15, but by far, the greatest benefit of this technology is the unprecedented control over the detail of structures fabricated thanks to the high precision of modern laser scribing instruments16,17. This precision has not been fully exploited though, and most researchers continue to produce simple interdigitated, spiral, concentric, or parallel-bar ESC electrodes in their prototypes18,Error! Bookmark not defined.,19,19. The benefit of using more complex electrode designs, namely fractal patterns, remains largely a virgin field, with no work done to determine the relationship between fractal order and electrolytic capacitance. This is despite the success of applying fractal patterns to other fields in electronics20,21 and the strong theoretical case for fractal ESCs22 In this work, we demonstrate the impact of using intricate LSG fractal patterns in ESC electrodes. We examine the scaling laws which govern the relationship between fractal order and capacitive performance for several different fractal and plane-filling curves. Results and Discussion Due to external pressures such as natural selection, nature is expert in fully exploiting available surfaces by the formation of intricate designs, for example by the repetition of similar patterns over a great range of length scales: such structure is described by fractal geometry. As a result, fractals, and related space-filling structures, can be found throughout nature in the patterns of clouds23, neuronal and vessel networks24, as well as coasts, trees, galaxy clusters, leaves, and the paths of particles undergoing Brownian motion, to name but a few of the diverse topics investigated and popularized by Mandelbrot25. Fractals are defined as objects possessing a Hausdorff dimension greater than that of their geometric dimension. The Hausdorff dimension of a fractal is calculated based on scalability and self-similarity using the following formula: 𝐻𝑎𝑢𝑠𝑑𝑜𝑟𝑓𝑓 𝐷𝑖𝑚𝑒𝑛𝑠𝑖𝑜𝑛 = log (𝑅) log (𝑀) Where R is the repetition of self-similar patterns for each iteration of the fractal order, and M is the factor by which the self-similar pattern decreases in size for each iteration. The Hausdorff dimension, therefore, can be viewed as a rough measure of the compactness and complexity of the particular fractal structure under investigation. The Hausdorff dimension is shown in Figure 1a for each fractal investigated in this work. It is also shown in Figure 1b that the Hausdorff dimension is directly proportional to the path-length function of the fractal pattern. Figure 1: a) Hausdorff dimension for a number of fractal and plane-filling curves; b) relationship between path length and fractal order for the same structures From Figure 1 it can be concluded that path length increases exponentially with order for all of the fractal structures investigated in this work, but does so more significantly for those structures with a larger Hausdorff dimension, namely the Sierpinski and Hilbert curves. Area should follow the same trend as it is simply the product of path length and the constant width of the line used to draw the fractal. Therefore, since electrolytic capacitance is related to area through the following formula: 𝐶 = 𝑘𝑆 𝑑 where C is capacitance, k is the dielectric constant for the electrolyte, S is surface area, and d is the electrolytic double layer (EDL) thickness, it would be expected that capacitance would also increase exponentially for capacitors made with fractal electrodes. This would be true in an ideal case, but in the real world, other factors are involved which are detrimental to capacitance, such as path resistance, and the effect of coulombic repulsion between adjacent portions of the like- plates of the fractal pattern. These trends were investigated by fabricating electrolytic capacitors of different orders for the fractal and space-filling curves in Figure 1. The fractals chosen for this work are popular and representative structures in mathematics, and are shown below in Figure 2. These are the Hilbert curve, Koch curve, Sierpinski curve, Minkowski sausage, Koch snowflake, and the Sierpinski arrowhead. Figure 2: Fractal electrode designs and detail of the laser scribed fractal pattern. (L-R): detail of scribed pattern, Sierpinski arrowhead and Hilbert curves, Koch and Koch snowflake curves, Minkowski and Sierpinski curves. The fractal patterns were scribed in a ~30 µm film of graphene oxide which had been drop- casted on a PET substrate. The scribing was accomplished using a 450 nm laser scribing system. The reduction of GO to LSG is evidenced by a dramatic reduction in resistance from 12 000 Ω/mm2 for the GO sheet to 670 Ω/mm for the laser reduced graphene pattern. The scribing system is capable of great detail and continuity as seen in the inset above, making it an ideal means of fabricating fractal electrodes. The ESC was completed by depositing a film of solid state electrolyte over its surface. This process is summarized in Figure 3. Figure 3: Method of fabricating fractal electrodes for electrolytic supercapacitors. (Clockwise from top left): A PET film was cleaned with isopropanol and masked; GO solution was deposited on the PET film and allowed to dry; the desired fractal pattern was scribed into the GO film using a x-y plotting system; the electrode sheet was removed from the x-y plotter; a film of solid- state PVA-phosphoric acid electrolyte was deposited on the electrode; the electrode was evaluated using an LCR meter and by analyzing its CV behavior. FTIR and Raman spectroscopy also showed the expected changes indicative of laser reduction of GO to RGO, namely a loss of the -- OH band at 3150 cm-1 and the -O- band at 1045 cm-1 in the FTIR and a shift in the relative heights of the D and G bands in the Raman spectrum. Both spectra are shown below in Figure 4. Figure 4: (L-R): FTIR and Raman spectra showing evidence of GO reduction to conducting RGO. The first ESCs to be evaluated were the Sierpinski curve and Minkowski sausage. These fractals have a Hausdorff Dimension of 2.00 and 1.50, respectively, indicating that these two structures scale with similar complexity. It is also notable that for lower fractal orders (1st, and 2nd), the two structures are visually similar, with a large open space in the centers of the first order structure, which is divided into roughly a cross shape in the second order (Figure 5). The similarity of the two patterns then diverges for higher orders, with the Sierpinski curve becoming more complex overall and the Minkowski Sausage remaining roughly cross- or sauwastika-shaped with a more open center. These changes can be easily visualized in Figure 5. Figure 5: Relative geometry of the four orders of the Minkowski curve (top) and the Sierpinski curve (bottom) This drastic increase in the complexity of the Sierpinski curve is due to its space-filling nature, meaning that as the order tends to infinity, the curve tends to cross every point in the x-y plane. Figure 6 shows the respective relationships between capacitance and fractal order for the Sierpinski curve (top) and the Minkowski curve (bottom). Figure 6: Changes in capacitance with respect to changes in fractal order for a) the Sierpinski curve and c) the Minkowski curve. A log-log analysis of the exponential growth of capacitance of both curves in seen in b) and c) respectively. The exponential nature of the capacitance-order relationship for both of these fractals is seen in plots a) and c), and is further investigated by plotting the respective natural logs of the ordinate and abscissa shown in plots b) and d). Such near-exponential relationships are conveniently described by a power law such as follows: 𝑦 = 𝑎𝑥𝑘 Where, in this case, y is capacitance, x is fractal order, k is the scaling factor, and a is a constant. The slope of the log-log plot corresponds to k, while the x intercept is taken to be ln(a). From this information, a scaling law can be formed which describes the relationship between capacitance (y) and fractal order (x) for the Sierpinski curve, And the Minkowski curve. 𝑦 = 6.11𝑥2.52 𝑦 = 8.33𝑥1.40 It is notable that the power in the scaling law in this case is proportional to the Hausdorff dimension, with the Sierpinski curve having a larger exponent and a larger Hausdorff dimension. It is tempting to assume that this relationship holds true in general, but there are likely many other factors involved in determining capacitance, such as repulsion due to the proximity of like- charged plates, and the effect of increasing resistance with increasing path length. To determine how these two factors influence the capacitance-order relationship, two additional fractal structures were fabricated: the Koch curve and the Koch snowflake. The benefit of using these two structures is that he Hausdroff dimension and dimensional scaling law of these two are identical, and the path length differs by a constant factor of 3. The only change with increasing order then, is in the magnitude of repulsions between like plates; the Koch curve remains a highly open structure with minimal repulsion despite changes in order, whereas the Koch snowflake is cyclic, so additional increases in order result in increasingly close contact between any two points on the like plates of the capacitor and therefore increased coulombic repulsion. Figure 7: Comparison of the geometry of the first four iterations of the Koch curve (top) and the Koch snowflake (bottom). The Koch curve is clearly a more open structure with little repulsion, whereas the Koch snowflake's cyclic structure is more repulsive. This repulsion would be expected to manifest itself in increased capacitance for changes in order for the Koch curve, but decreased capacitance for changes in order for the Koch snowflake. Figure 8: Comparison of the capacitance-order relationship for a) the Koch curve c) the Koch snowflake, b) the Sierpinski arrowhead, and d) the Hilbert curve As shown in Figure 8a, and c above, the predicted trend was generally observed, with capacitance increasing in a more linear fashion for the Koch curve (a) with each change in order. The results were much more variable for the Koch snowflake (b), as expected; there was a promising initial increase in capacitance between order 1 and 2 which then declined and plateaued for orders 3 and 4. This indicates that initially there was a similar relationship between capacitance and order as for the Koch curve, but as order increased, the repulsions between adjacent portions of the like plates overcame the growth in capacitance for the Koch snowflake. Next the effect of surface area was investigated in two intermediate fractal patterns: these were the Hilbert curve and the Sierpinski arrowhead. Both curves possess large Hausdorff dimensions (2.00 and 1.58, respectively), and are therefore relatively compact and complex structures. The primary difference between the two structures is the rate of change in path length, and therefore surface area, with increasing order. As seen in Figure 1b, Hilbert curve has a greater exponential growth in length than does the Sierpinski arrowhead. Figure 9: Comparison of the geometries of the first four orders of the Sierpinski arrowhead (top) and the Hilbert curve (bottom) The capacitance and order data Figure 8 (b and d) are again non-linear, and prima facie, non- intuitive. In both cases, capacitance begins at a relatively large value for the first order structures. This is expected as both structures possess large surface area and in the first order, there is little repulsion due to the great distance between like plates. However, when the order is increased to 2, there is a large decline in capacitance for both fractals due, presumably, to the increased repulsions as anticipated. However, the more interesting result for this set is the increase in capacitance following the initial slump. This is especially evident for the Hilbert curve, capacitance of which begins increasing again after the 2nd order. This is likely due to the large exponential increase in surface area with order Figure 1b which may be capable of overcoming the repulsions due the compact structure. This exponential increase in surface area with order is not as large for the Sierpinski arrowhead, and therefore there is a delay in the recovery of capacitance which doesn't begin until the 3rd order. Taken together, these results illustrate an interesting circumstance: for some simple fractal structures with high surface area such as the Sierpinski curve, or with large open structures as in the Minkowski sausage, there exists a linear or exponential relationship between electrochemical capacitance and fractal order. This relationship is shown to be roughly proportional to the Hausdorff dimension of the particular fractal under investigation. As the structure becomes more complex, though, other forces seem to limit the increase in capacitance. This was shown by comparing the Koch curve and Koch snowflake, which share the same Hausdorff dimension and length scaling law, but differ in the magnitude of interaction between like plates. Thus the capacitance scales differently between these two structures. Additionally, by comparing the capacitance scaling of the Hilbert curve with the Sierpinski arrowhead, it was shown that in some cases at least, the repulsive effects can be overcome at higher orders by a structure which increases its surface area at a sufficient rate. Methods GO films were prepared on thin PET substrates by drop casting 6.7 mL of aqueous GO (Graphenea, 4 g/L) on a ~14 400 mm2 sheet (cleaned with isopropyl alcohol). The GO film and PET substrate was incubated at 50oC until dry, cut into sheets, and stored. The areal density of GO on the substrate was calculated to be ~2.7x10-6 g mm-2. The average thickness of these films was ~30 µm. The GO films appeared to be dark gray to ochre, depending on the thickness, and had a transmittance of ~45% of the solar spectrum. The GO film had a sheet resistance of ~12 000 Ω/mm2. The adhesion between the GO film and the PET substrate was poor, but adhesion between the LSG and PET was much better, possibly due to GO being fused into the partially molten plastic during the laser scribing process. Resistance in the LSG path was ~670 Ω/mm. FTIR and Raman analysis of the GO film revealed the strong OH stretching band centered at 3150, G and D bands at 1720 and 1615, respectively, and an ether peak at 1045. These bands and peaks were largely flattened when the GO film was laser scribed, indicating a successful reduction of the film as described in previous literature26; further evidence of the reduction process can be seen in the Raman spectrum as a shift in the relative heights of the D and G bands. Sierpinski, Minkowski, Koch, Koch snowflake, Hilbert and Sierpinski arrowhead curves of the 1st through 4th order were plotted using Wolfram Mathematica's built-in functions of the same name. The resulting graphics items was copied and opened in a drawing software (Microsoft Paint) where simple current collectors were added to the image. The electrode image was then opened in BenBox laser scribing software which was linked to an x-y plotter (MakeBlock x-y plotter V 2.0). A laser setting of 200 units of intensity, and a scribing speed of 6.7 mm/s was used. The laser used was supplied by MakeBlock and was a 450 nm blue-violet laser with a maximum power output of 450 mW -- 500 mW. By using the "scan by outline" setting in BenBox, the inner and outer outline of the fractal image was scribed, resulting in a double fractal pattern which is parallel at each point. The inner and outer lines could then be attached to positive and negative current collectors to complete the capacitive circuit. Theoretically this design would enable both electrolytic capacitance between the LSG and electrolyte in addition to traditional capacitance by the storage of opposite charges on parallel LSG regions. The distance between the parallel lines was easily controlled by the Thickness function in Mathematica. The GO layer was scribed 4 times, as previous work has shown that multiple laser passes improve conductivity14. A polymer gel electrolyte (PVA-phosphoric acid) was prepared as described elsewhere27 by combining PVA powder, water, and concentrated phosphoric acid in a 10:100:8 mass ratio. The resulting thick solution was pipetted across the scribed electrode (excluding only the current collector region), which was then stored away from dust until the gel was dry, a process which took up to 24 hours depending on temperature, humidity, and airflow. The final resistance of the electrolyte (measured by 4-point probe) was ~80 Ω/mm2 and the thickness of the layer (measured by Keyence LK Laser Displacement Meter) was approximately 35 µm. The average volume of these devices was approximately 0.32 cm3 (measured by water displacement) and mass was about 0.32 g. The completed ESC was evaluated using a BK Precision 891 LCR Meter, and a Keithley 2450 SourceMeter. Capacitance (C), Q factor (Q), equivalent series resistance (R), and dissipation factor (D) were determined at 200 Hz, 1 VRMS, in FAST mode using the LCR meter. The cyclic voltammetry (CV) behaviour of the capacitors was determined with the SourceMeter at different sweep frequencies between 0 and 1 V and 10 or 50 double sweeps. Acknowledgement KD and J.P. Badjo would like to thank Maryland Technology Enterprise Institute for their generous support through MIPS grant. 1 Vangari, M.; Pryor, T.; Jiang, L. (2013). "Supercapacitors: Review of Materials and Fabrication Methods." Journal of Energy and Engineering. (72 -- 79). 2 Lee, H-H.; et al. (2015). "Direct and environmentally benign synthesis of manganese oxide/graphene composites from graphite for electrochemical capacitors." Journal of Power Sources. 281 (44 - 48). 3 Simon, P.; Gogotsi, Y. (2011). "Capacitive Energy Storage in Nanostructured Carbon- Electrolyte Systems." Accounts of Chemical Research. 46 (1094 - 1103). 4 Kaempgen, M.; Chan, C.K.; Ma, J.; Gruner, G. (2009). "Printable thin film supercapacitors using single-walled carbon nanotubes." Nano Letters. 9 (1872 -- 1876). 5 Lu, W.; Qu, L.; Henry, K.; Dia, L. (2009). "High performance electrochemical capacitors from aligned carbon nanotube electrodes and ionic liquid electrolytes." Journal of Power Sources. 189 (1270 - 1277). 6 Simon, P.; Gogotsi, Y. (2011). "Capacitive Energy Storage in Nanostructured Carbon- Electrolyte Systems." Accounts of Chemical Research. 46 (1094 - 1103). 7 Liu, C.; Yu, Z.; Neff, D.; Zhamu, A.; Jang, B.Z. (2010). "Graphene-Based Supercapacitor with an Ultrahigh Energy Density." Nano Letters. 10 (4863 -- 4868). 8 Zhu, Y.; el at. (2011). "Carbon-based supercapacitors produced by activated graphene." Science. 332 (1537 - 1541). 9 Shao, Y.; et al. (2015). "Graphene-based materials for flexible supercapacitors." Chem Soc. Review. 44 (3639 -- 3665). 11 Trusovas, R.; et al. (2012). "Reduction of graphite oxide to graphene with laser irradiation." Carbon. Get Info. 12 Strong, V.; et al. (2012). "Patterning and Electronic Tuning of Laser Scribed Graphene for Flexible All-Carbon Devices." ACS Nano. 6 (1395 -- 1403). 13 Guan, Y.C.; Fang, Y.W.; Lim, G.C.; Zheng, H.Y.; Hong, M.H. (2016). "Fabrication of laser reduced graphene oxide in liquid nitrogen environment." Scientific Reports. 6 (get - numbers). 14 El-Kady, M.F.; Kaner, R.B.; (2013). "Scalable fabrication of high-power graphene micro- supercapacitors for flexible and on-chip energy storage." Nature Communications. 4 (1 -- 9). 15 Evlashin, S.; et al. (2016). "Controllable laser reduction of graphene oxide films for photoelectric applications." Applied Materials and Interfaces. 8 (28880 -- 28887). 16 "XY Plotter Robot Kit." MakeBlock. http://store.makeblock.com/xy-plotter-robot-kit/ 17 "LPKF ProtoMat E34/E44." LPKF. http://www.lpkf.com/_mediafiles/3866-brochure-lpkf- protomat-e34-44-en.pdf 18 El-Kady, M.F.; Strong, V.; Dubin, S.; Kaner, R.B. (2012). "Laser scribing of high- performance and flexible graphene-based electrochemical capacitors." Science. 335 (1326 - 1330). 19 Gao, W.; et al. (2011). "Direct laser writing of micro-supercapacitors on hydrated graphene oxide films." Nature Nanotechnology. 6 (496 -- 500). 20 Cohen, N. (1997). "Fractal antenna applications in wireless telecommunications." IEEE. (43 -- 49). 21 Puente, C.; Romeu, J.; Pous, R.; Ramis, J.; Hijazo, A. (1998). "Small but long Koch fractal monopole." Electronics Letters. 34 (9 - 10). 22 Park, B.; Zaouk, R.; Wang, C.; Madon, M.D. (2007). "A case for fractal electrodes in electrochemical systems." Journal of the Electrochemical Society. 154 (1 - 5). 23 Barnosky, A.D. (1982). "Area-perimeter relation for rain cloud areas." Science. 216 (185 - 187). 24 Panico, J.; Sterling, P. (1995). "Retinal neurons and vessels are not fractal but space-filling." The Journal of Comparative Neurology. 361 (479 - 490). 25 Mandelbrot, B.B. (1977). The Fractal Geometry of Nature. W.H. Freeman and Company. 26 Das, A.K.; et al. (2014). "Iodide-mediated room temperature reduction of graphene oxide: a rapid chemical route for the synthesis of a bifunctional electrocatalyst." Journal of Materials Chemistry A. 5 (1332 - 1340). 27 Kaempgen, M.; Chan, C.K.; Ma, J.; Gruner, G. (2009). "Printable thin film supercapacitors using single-walled carbon nanotubes." Nano Letters. 9 (1872 -- 1876).
1712.01390	1	1712	2017-12-04T22:11:56	Characterizations of the Nonlinear Optical Properties for (010) and (-201) Beta-phase Gallium Oxide	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We report, for the first time, the characterizations on optical nonlinearities of beta-phase gallium oxide (\b{eta}-Ga2O3), where both (010) \b{eta}-Ga2O3 and (-201) \b{eta}-Ga2O3 were examined for two-photon absorption (TPA) coefficient, Kerr refractive index, and their polarization dependence. The wavelength dependence of the TPA coefficient and Kerr refractive index was estimated using a widely used analytical model. \b{eta}-Ga2O3 exhibits a TPA coefficient of 1.2 cm/GW for (010) \b{eta}-Ga2O3 and 0.58 cm/GW for (-201) \b{eta}-Ga2O3. The Kerr refractive index is -2.1410^(15) cm2/W for (010) \b{eta}-Ga2O3 and -2.8910^(15) cm2/W for (-201) \b{eta}-Ga2O3. In addition, \b{eta}-Ga2O3 shows stronger nonlinear optical anisotropy on the (-201) plane than on the (010) plane, possibly due to highly asymmetric crystal structure. Compared with that of gallium nitride (GaN), the TPA coefficient of \b{eta}-Ga2O3 is 20 times smaller, and the Kerr refractive index of \b{eta}-Ga2O3 is also found to be 4 to 5 times smaller. These results indicate that \b{eta}-Ga2O3 has the potential for use in ultra-low loss waveguides and ultra-stable resonators and integrated photonics, especially in the UV and visible wavelength spectral range.	physics.app-ph	physics	Characterizations of the Nonlinear Optical Properties for (010) and (𝟐01) Beta-phase Gallium Oxide Hong Chen, Houqiang Fu, Xuanqi Huang, Jossue A. Montes, Tsung-Han Yang, Izak Baranowski and Yuji Zhao* School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85287, U.S.A. *[email protected] Keywords: Nonlinear optics; Beta-phase gallium oxide; Multi-photon absorption Abstract: We report, for the first time, the characterizations on optical nonlinearities of beta- phase gallium oxide (β-Ga2O3), where both (010) β-Ga2O3 and (201) β-Ga2O3 were examined for two-photon absorption (TPA) coefficient, Kerr refractive index, and their polarization dependence. The wavelength dependence of the TPA coefficient and Kerr refractive index was estimated using a widely used analytical model. β-Ga2O3 exhibits a TPA coefficient of 1.2 cm/GW for (010) β-Ga2O3 and 0.58 cm/GW for (201) β-Ga2O3. The Kerr refractive index is −2.14×1015 cm2/W for (010) β-Ga2O3 and −2.89×1015 cm2/W for (201) β-Ga2O3. In addition, β-Ga2O3 shows stronger nonlinear optical anisotropy on the (201) plane than on the (010) plane, possibly due to highly asymmetric crystal structure. Compared with that of gallium nitride (GaN), the TPA coefficient of β-Ga2O3 is 20 times smaller, and the Kerr refractive index of β- Ga2O3 is also found to be 4–5 times smaller. These results indicate that β-Ga2O3 has the potential for use in ultra-low loss waveguides and ultra-stable resonators and integrated photonics, especially in the UV and visible wavelength spectral range. 1. Introduction As an emerging wide bandgap semiconductor material, beta-phase gallium oxide (β-Ga2O3) has attracted considerable attention in catalysis[1,2], gas sensors[3], power electronics[4] and potential optical devices such as waveguides and resonators. Due to its wide bandgap, β-Ga2O3 possesses a board transparent spectrum from ultraviolet (UV) to visible wavelengths, ideal for optical applications at this wavelength range. Due to the compatibility of β-Ga2O3 with the III-nitride material system[9], β-Ga2O3 optical devices can also be actively-integrated with III-nitride based UV–visible light sources[10–14] and detectors[15–16]. These unique properties make β-Ga2O3 a very promising candidate for emerging integrated photonics applications at elusive UV to visible wavelengths, which are critical for various optical applications such as biochemical sensing, UV Raman spectroscopy, frequency up/down conversion, and quantum emitters, etc[17]. In order to realize the full potential of β-Ga2O3 materials in these optical applications, a comprehensive investigation of the material properties of β-Ga2O3 is of crucial importance. However, previous studies on β-Ga2O3 have been mainly focused on the electronic[5] and thermal properties[9] of β-Ga2O3, focusing on power electronics applications, with only very limited reports on the simple characterizations of basic optical properties on β-Ga2O3, such as transmission[6] and refractive index[7,8]. The fundamental nonlinear optical properties of β- Ga2O3, which are vital for various optical applications such as integrated photonics and quantum photonics, have never been investigated. Nonlinear optical processes such as the two-photon absorption (TPA) coefficient and the Kerr effect play a significant role in determining the performance of optical devices. For example, the TPA process is one of the major optical loss mechanisms for optical waveguides and resonators under high optical power density[18,19]. For resonators operating by critical coupling from bus waveguides, the refractive index shifting, mainly governed by the Kerr refractive index at high power density, will affect the coupling efficiency especially for ultra-high quality factor resonators[21]. To develop high performance β-Ga2O3 optical devices, it is therefore of paramount importance to characterize and understand the nonlinear optical properties of β- Ga2O3 including the TPA coefficient and Kerr refractive index. To investigate and evaluate the nonlinear optical properties of β-Ga2O3 in the visible spectral range, we performed a typical Z-scan characterization to study the TPA coefficient and Kerr refractive index of β-Ga2O3. The results show that β-Ga2O3 has a much smaller TPA coefficient (20 times smaller) and Kerr refractive index (4–5 times smaller)[19] compared to GaN, which is ideal for high performance waveguides and resonators in visible and potential UV wavelengths. Furthermore, due to the highly asymmetric crystalline structure of β-Ga2O3, the optical nonlinearity is found to be highly anisotropic. Relatively stronger nonlinear optical anisotropy was observed on the (201) plane than on the (010) plane. These results will serve as important references and guidelines for the design and fabrication of future photonic devices based on β- Ga2O3. This paper is organized as the following: In Section 2, we describe the methods used in this study, including experimental setup and theoretical models. In Section 3, we show the experimental results on β-Ga2O3 materials and discuss their impacts on the photonic devices. In section 4, we provide a brief summary of the work. 2. Methods Figure 1(a) schematically shows the experimental setup of the Z-scan measurement[19,21,22] used in this study. Light source was an ultrafast titanium-sapphire laser operating at 808 nm. A 𝜒2 crystal was used to generate the second harmonic wave at the wavelength of 404 nm. A half wavelength plate working at 404 nm was placed after the 𝜒2 crystal for light polarization tuning. Light beam was expanded by a set of lenses before being sent into optical objectives in order to fully utilize the numerical aperture. The samples were positioned between two optical objectives. An aperture was implemented in front of a power meter to perform open and closed aperture testing. The unintentionally doped (UID) β-Ga2O3 samples used in this work were provided by Tamura Corporation with a carrier concentration on the order of ~1017 cm-3 and a thickness of ~ 500 µm. The backside of the samples was polished by a hand-grinding method with diamond lapping film of 0.5 µm grade. The polishing process was carefully controlled so that the thickness of samples after polishing was reduced by less than 50 µm. Figure 1(b) depicts the crystal structure of β-Ga2O3, where the (010) and (201) crystal planes are also illustrated. To test the reliability of the setup, the beam size of the output light is used as an indicator. Based on Gaussian optics, with an incident beam diameter of 4 mm and a numerical aperture (NA) of 0.25, the calculated diameter of the output light is ~5 μm (at 1/e2) as shown in Figure 1(c). This value is consistent with beam diameter we extracted from open aperture Z-scan measurement in next section. The TPA coefficient and Kerr refractive index can be obtained from open/closed aperture scanning based on Equation 1–3, 𝛼𝑇𝑃𝐴𝐼0𝐿𝑒𝑓𝑓 2√2 𝑇 ≈ 1 − 𝑍0 = 2 𝑛𝜋𝜔0 λ × 1 2⁄ 1 + 𝑍2 𝑍0 𝑛𝑘𝑒𝑟𝑟 = λΔ𝜙 2𝜋𝐼0𝐿𝑒𝑓𝑓 (1) (2) (3) where T is the normalized transmission, 𝐼0 is the peak beam power density, 𝐿𝑒𝑓𝑓 indicates effective sample length, 𝑍0 indicates Rayleigh range of the beam, 𝑛 is the refractive index, 𝜔0 is the beam size at the focal plane, and 𝜆 is the wavelength. 𝐿𝑒𝑓𝑓 refers to the sample thickness, Δ𝜙 the nonlinear phase shift due to Kerr effect[21]. More information about these equations can be found in References 18, 20–24. It should be noted that the samples are tested at 404 nm which is above the half bandgap energy of β-Ga2O3 (506 nm). Therefore, a three-photon absorption modification to Equation 1 is not required in our case[26]. The wavelength- dependence of the TPA coefficients is also theoretically calculated using Equation 4, 𝛼𝑇𝑃𝐴(ω) = 𝐾 √𝐸𝑝 2𝐸𝑔 𝑛0 3 𝐹2(ℏ ∙ ω 𝐸𝑔⁄ ) (4) in which 𝐸𝑔 is the direct bandgap energy, 𝐸𝑝 = 2𝑃𝑣𝑐2/𝑚0 is a material-independent parameter for direct bandgap semiconductors obtained by the 𝑘 ∙ 𝑝 model[27], 𝑛0 is the refractive index, 𝜔 is the frequency, and K is a material-independent constant. 𝐹2 is a fitting function with the form 𝐹2(𝑥) = (2𝑥 − 1)1.5 (2𝑥)5 ⁄ . It should be noted that β-Ga2O3 exhibits a direct bandgap energy of 4.9 eV and a slightly smaller indirect bandgap of 4.85 eV[1]. In our theoretical analysis, we do not consider phonon assisted TPA because it only contributes to optical nonlinearity at a narrow bandwidth. 3. Experimental results and discussions Figure 2 shows the open aperture scanning results for (010) and (201) β-Ga2O3. Fitting the experimental data with Equation 1–3 yielded the TPA coefficients and Kerr refractive index, which were summarized in Table 1. (010) β-Ga2O3 showed a TPA coefficient αTPA = 1.2 cm/GW and nkerr = −2.14 × 10-15 cm2/W when 𝐸⃗ ⊥ [102], while an αTPA of 0.58 cm/GW and nkerr of −2.89 × 10-15 cm2/W were obtained on (2 01) β-Ga2O3 when 𝐸⃗ ∥ [102]. Due to the smaller TPA coefficient on (201) β-Ga2O3 than that on (010) β-Ga2O3, optical devices such as waveguides and resonators on (201) β-Ga2O3 will have lower optical loss induced by TPA at the visible spectral regime. In comparison, (201) β-Ga2O3 showed a larger Kerr refractive index than that on (010) β-Ga2O3. In addition, the output beam diameters were also obtained, which are in good agreement with theoretical calculations based on Gaussian optics. This indicates the setup is reliable and accurate. There are several physical models that can be implemented to understand this highly anisotropic optical nonlinearity[28–33]. A quantum mechanical approach[28,29] calculates third order susceptibility accurately but requires intensive computing resources. Therefore it is only utilized in simple crystal structures[29]. A simplified bond-orbital model developed in Reference 30–32 successfully explained the optical nonlinearities for various kinds of materials including metal-oxide crystals[33]. We employed this model in this work to qualitatively understand the anisotropic nonlinearity of (010) and (201) β-Ga2O3. From the bond-orbital model, the optical anisotropy of a crystal is mainly contributed from bonding electrons between adjacent atoms[30], while for other electrons that screen around individual atoms, the contribution to the optical anisotropy is less significant. As investigated comprehensively in Reference 34, β-Ga2O3 is constructed by two types of gallium ions and three types of oxygen ions as shown in Figure 3(a) and 3(b). Type I gallium ion Ga(I) is surrounded by a distorted tetrahedron of oxygen ions with atomic distances from 1.80 A to 1.85 A (1.83 A in average), type II gallium ion Ga(II) is surrounded by a distorted octahedron with atomic distances from 1.95 A to 2.08 A (2.00 A in average). On the (010) plane of β-Ga2O3, the bonds are not orderly oriented and therefore electrical fields in the (010) plane are more likely to interact with most of the bonds. However, on the (201) plane, bonds between Ga(I) and O(III) are highly oriented along the (010) direction. Such bonds have a relatively weak interaction with electric fields polarized in (201) plane. The consequence is that the optical nonlinearity on the (201) plane is relatively weaker than that on the (010) plane. We previously estimated the TPA coefficients and Kerr refractive index for GaN[19] at same wavelength range. The TPA coefficient of (010) and (201) β-Ga2O3 are ~10 and ~20 times smaller than that of GaN at 404 nm, respectively. The high TPA coefficient observed in GaN might due to its exciton effects[22]. This result implies that the β-Ga2O3 material is more capable of handling high optical power density applications in visible wavelength spectral range. Furthermore, Kerr coefficients of β-Ga2O3 are also 4–5 times smaller than that of GaN. For optical applications that require critical coupling, e.g., coupling from bus waveguide to ring or disk resonators, the Kerr effect modifies the coupling efficiency at high optical power density, especially in high resonance quality factor resonators[20]. Therefore, β-Ga2O3-based resonators will exhibit extreme coupling stability under high power operation compared to GaN-based resonators. Figure 4 represents the polarization dependences of the TPA and Kerr coefficients of (010) and (201) β-Ga2O3. Since β-Ga2O3 has a monoclinic crystal structure (Figure 1), 41 independent nonzero elements are required to fully describe its third order nonlinearity dil. Therefore, it is very difficult and inconvenient to find an explicit expression for deff using polarization angle and individual nonlinear component dil [20–22,25]. To describe the anisotropic nonlinearity clearly, we used ∆Tmax/∆Tmin in this work. Relatively higher nonlinear optical anisotropy was found on (201) β-Ga2O3 with ∆Tmax/∆Tmin of 1.93 while (010) β-Ga2O3 had a ∆Tmax/∆Tmin of 1.29. Figure 5 shows the wavelength dependence of the TPA coefficient using Equation 4 for (010) and (201) β-Ga2O3. For (010) β-Ga2O3, E⊥ has larger TPA coefficient than E∥. For (201) β- Ga2O3, E⊥ has a smaller TPA coefficient than E∥. The highest TPA coefficient is observed in (010) β-Ga2O3 when the electric field is perpendicular to the [102] direction, while the lowest TPA coefficient is obtained on (201) β-Ga2O3 when the electric field is parallel to the [102] direction. For each plane, the polarization dependence of the TPA coefficient is a function of multiple physical parameters[34] such as bond iconicity, covalent radii, etc. The physical mechanism of this polarization dependence on different crystal orientations of β-Ga2O3 is a topic of on-going investigation. For both (2 01) β-Ga2O3 and (010) β-Ga2O3 samples, the maximum TPA process occurs at the ~360 nm wavelength. For wavelengths above 360 nm, the TPA coefficient decreases as wavelength increases, which can be attributed to the decreased excitation energy with increasing wavelength. For wavelengths below 360 nm, the TPA coefficient increases as the transition approaches its resonance wavelength. 4. Conclusions We characterized the TPA coefficient and Kerr refractive index of both (010) and (𝟐01) β- Ga2O3. The TPA coefficient of Ga2O3 was found to be 10 to 20 times smaller than that of GaN at 404 nm. The Kerr refractive index of Ga2O3 was 4 to 5 times lower than that of GaN. Therefore, due to its ultra-low TPA coefficient and its small Kerr refractive index, β-Ga2O3 has the potential to serve as a more efficient platform for integrated photonic applications in UV and visible spectral range. Furthermore, the optical nonlinearities of β-Ga2O3 are highly anisotropic due to its asymmetric crystal structure.. These results can serve as guidelines for the design and fabrication of β-Ga2O3-based integrated photonic devices at UV–visible wavelengths for various optical applications such as biochemical sensing, UV Raman spectroscopy, frequency up/down conversion, and quantum photonics. Acknowledgements The authors thank Dr. Michael Gerhold of US Army Research Office for the helpful discussion. This work is supported by the Bisgrove Scholar Program from Science Foundation Arizona. We gratefully acknowledge the use of facilities within the LeRoy Eyring Center for Solid State Science and CLAS Ultra Fast Laser Facility at Arizona State University. [1] S.I. Stepanov, V.I. Nikolaev, V.E. Bougrov, and A.E. Romanov, Rev. Adv. Mater. Sci. 2016, 44. [2] T. Oshima, K. Kaminaga, H. Mashiko, A. Mukai, K. Sasaki, T. Masui, A. Kuramata, S. Yamakoshi, and A. Ohtomo, Jpn. J. Appl. Phys. 2013, 52, 11R. [3] K. Sasaki, M. Higashiwaki, A. Kuramata, T. Masui, and S. Yamakoshi, IEEE Electron Device Lett. 2013, 34, 4. [4] K. Matsuzaki, H. Hiramatsu, K. Nomura, H. Yanagi, T. Kamiya, M. Hirano and H. Hosono, Thin Solid Films 2006, 496, 1. [5] C. Janowitz, V. Scherer, M. Mohamed, A. Krapf, H. Dwelk, R. Manzke, Z. Galazka, R. Uecker, K. Irmscher, R. Fornari, and M. Michling, New J. Phys. 2011, 13, 8. [6] Z. Galazka, K. Irmscher, R. Uecker, R. Bertram, M. Pietsch, A. Kwasniewski, M. Naumann, T. Schulz, R. Schewski, D. Klimm, and M. Bickermann, J. Crys. Growth 2014, 404. [7] I. Bhaumik, R. Bhatt, S. Ganesamoorthy, A. Saxena, A.K. Karnal, P.K. Gupta, A.K. Sinha, and S.K. Deb, Appl. Opt. 2011, 50, 31. [8] H. He, R. Orlando, M.A. Blanco, R. Pandey, E. Amzallag, I. Baraille, and M. Rérat, Phys. Rev. B 2006, 74, 19. [9] M. D. Sing, W. S. Liang, R. H. Horng, P. Ravadgar, T. Y. Wang, and H. Y. Lee, Proc. SPIE 2012, 8263. [10] Y. Zhao, S. Tanaka, C.C. Pan, K. Fujito, D. Feezell, J.S. Speck, S.P. DenBaars, and S. Nakamura, Appl. Phys. Express 2011, 4, 8. [11] H. Chen, H. Fu, Z. Lu, X. Huang, and Y. Zhao, Opt. Express 2016, 24, 10. [12] H. Fu, Z. Lu, X.H. Zhao, Y.H. Zhang, S.P. DenBaars, S. Nakamura, and Y. Zhao, J. Disp. Technol. 2016, 12, 7. [13] H. Chen, H. Fu, X. Huang, Z. Lu, X. Zhang, J.A. Montes, and Y. Zhao, IEEE Photon. J. 2017, 9, 3. [14] Y. Zhao, Q. Yan, C.Y. Huang, S.C. Huang, P. Shan Hsu, S. Tanaka, C.C. Pan, Y. Kawaguchi, K. Fujito, C.G. Van de Walle, and J.S. Speck, Appl. Phys. Lett. 2012, 100, 20. [15] H. Fu, H. Chen, X. Huang, Z. Lu, and Y. Zhao, Y, J. Appl. Phys. 2017, 121, 1. [16] J.B. Limb, D. Yoo, J.H. Ryou, W. Lee, S.C. Shen, R.D. Dupuis, M.L. Reed, C.J. Collins, M. Wraback, D. Hanser, and E. Preble, Appl. Phys. Lett. 2006, 89, 1. [17] M. Soltani, R. Soref, T. Palacios, and D. Englund, "AlGaN/AlN integrated photonics platform for the ultraviolet and visible spectral range", Opt. Express 2016, 24, 22. [18] H. Chen et al. "Low Loss GaN Waveguides at the Visible Spectral Wavelengths for Integrated Photonics Applications", submitted. [19] H. Chen, X. Huang, H. Fu, Z. Lu, X. Zhang, J.A. Montes, and Y. Zhao, Appl. Phys. Lett. 2017, 110, 18. [20] A. Yariv, IEEE Photon. Technol. Lett. 2002, 14, 4. [21] M. Sheik-Bahae, A.A. Said, T.H. Wei, D.J. Hagan, and E.W. Van Stryland, IEEE J. Quantum Electron 1990, 26, 4. [22] C.K. Sun, J.C. Liang, J.C. Wang, F.J. Kao, S. Keller, M.P. Mack, U. Mishra, and S.P. DenBaars, Appl. Phys. Lett. 2000, 76, 4. [23] Y. Fang, F. Zhou, J. Yang, X. Wu, Z. Xiao, Z. Li, and Y. Song, Appl. Phys. Lett. 2015, 106, 13. [24] R. DeSalvo, M. Sheik-Bahae, A.A. Said, D.J. Hagan, and E.W. Van Stryland, Opt. Lett. 1993, 18, 3. [25] K. Wang, J. Zhou, L. Yuan, Y. Tao, J. Chen, P. Lu, and Z.L. Wang, Nano Lett. 2012, 12, 2. [26] V. Pačebutas, A. Krotkus, T. Suski, P. Perlin, and M. Leszczynski, J. Appl. Phys. 2002, 92, 11. [27] M. Sheik-Bahae, D.C. Hutchings, D.J. Hagan, and E.W. Van Stryland, IEEE J. Quantum Electron. 1991, 27, 6. [28] R.W. Boyd, Nonlinear optics, Academic press 2003. [29] Y.R. Shen, The principles of nonlinear optics, New York Wiley-Interscience 1984. [30] J.C. Phillips, Phys. Rev. Lett. 1968, 20, 11. [31] B.F. Levine, J. Chem. Phys. 1973, 59, 3. [32] B.F. Levine, Phys. Rev. B, 1973, 7, 6. [33] K. Koynov, P. Mitev, I. Buchvarov, and S. Saltiel, Pure Appl. Opt. Europ. Opt. Soc. P. A 1996, 5, 1. [34] S. Geller, J. Chem. Phys. 1960, 33, 3. Figure 1. (a) Experimental setup of the Z-scan measurement used in this work. AT indicates the attenuator, λ/2 is the half-wavelength plate, PL is the polarizer, OB is the optical objective lens, A is the aperture, PM is the power meter. L1, L2, L3 are lenses. (b) Crystalline structure of β-Ga2O3 showing (101) and (201) crystal orientations. (c) The estimated field intensity at the focal point, which is in good agreement with our open aperture fitting. Figure 2. Typical open aperture scanning curve obtained in this work. (a) Scan curve for (010) β-Ga2O3 sample, (b) for (201) β-Ga2O3 sample. Figure 3. Schematic views for the arrangement of ions inside β-Ga2O3, different types of ions are marked out by numbers. (a) Schematic view for (010) β-Ga2O3 sample, (b) for (201) β- Ga2O3 sample. Figure 4. Polarization dependence of minimum transmittance obtained in this work. (a) Polarization dependence for (010) β-Ga2O3 sample, (b) for (201) β-Ga2O3 sample. Figure 5. The estimated wavelength dependence of the TPA coefficient for (010) and (201) β- Ga2O3 samples. "E⊥" indicates that the electrical field intensity is perpendicular to [102] direction, while "E∥" indicates that field intensity is parallel to [102] direction Table 1. TPA coefficients and Kerr coefficients obtained in this work for (010) and (201) β- Ga2O3 samples. Beam diameter from fitting parameter is also shown. (010), 𝐸⃗ ⊥ [102] (201), 𝐸⃗ ∥ [102] αTPA(cm/GW) nkerr(cm2/W) R(μm) αTPA(cm/GW) nkerr(cm2/W) R(μm) 1.20 −2.14×10-15 5.8 0.58 −2.89×10-15 5.68
1809.06952	2	1809	2018-11-05T17:09:41	Reduced order derivation of the two-dimensional band structure of a mixed-mode resonator array	[ "physics.app-ph" ]	In this paper, the 2D band structure of a mixed-mode metamaterial resonator array for in-plane waves is investigated. The band structure in the interior and on the boundary of the irreducible Brillouin zone as well as 1D dispersion diagrams for different propagation angles are calculated numerically and presented. Additionally, a reduced order analytical method is established to compare and approximate the band structure. The studied metamaterial, with a T-shaped cantilever beam as resonator in its square array repeating unit cells, exhibits branches with mixed P and SV waves except at exactly one angle of propagation. This paper also reports on the occurrence of avoided level crossings, which are related to the existence of exceptional points in the complex domain. A reduced order analytical approach is used that can generate partial (low branches) band structure with relatively little computational effort. The reduced order model agrees well with the numerical results for these low branches and can provide support in mode identification and band sorting. With proper adjustments in parameters, this analytical method will be applicable for other metamaterials that have similar unit cell structure.	physics.app-ph	physics	Reduced order derivation of the two-dimensional band structure of a mixed-mode resonator array Alireza V. Amirkhizia) and Weidi Wang Department of Mechanical Engineering, University of Massachusetts, Lowell, Lowell, Massachusetts 01854, USA (Dated: 6 November 2018) In this paper, the 2D band structure of a mixed-mode metamaterial resonator array for in-plane waves is investigated. The band structure in the interior and on the boundary of the irreducible Brillouin zone as well as 1D dispersion diagrams for different propagation angles are calculated numerically and presented. Additionally, a reduced order analytical method is established to compare and approximate the band structure. The studied metamaterial, with a T-shaped cantilever beam as resonator in its square array repeating unit cells, exhibits branches with mixed P and SV waves except at exactly one angle of propagation. This paper also reports on the occurrence of avoided level crossings, which are related to the existence of exceptional points in the complex domain. A reduced order analytical approach is used that can generate partial (low branches) band structure with relatively little computational effort. The reduced order model agrees well with the numerical results for these low branches and can provide support in mode identification and band sorting. With proper adjustments in parameters, this analytical method will be applicable for other metamaterials that have similar unit cell structure. Keywords: mechanical metamaterials; band structure; mode mixing, level repulsion 8 1 0 2 v o N 5 ] h p - p p a . s c i s y h p [ 2 v 2 5 9 6 0 . 9 0 8 1 : v i X r a a)Electronic mail: alireza [email protected] 1 I. INTRODUCTION Acoustic metamaterials and phononic crystals1 have attracted a lot of attention in the past two decades mostly because of their potential ability to manipulate waves and their unusual overall dynamic properties. These novel dynamic properties lead to many pro- posed applications, such as wave filtering2,3, attenuation4,5, negative refractive index6 -- 8, and cloaking9,10. Due to the local resonance in their periodic structure, such media exhibit frequency band gaps, within which there are no propagating waves. The determination of band structure is of prime importance as the natural first step in deriving the overall properties of metamaterials. In particular the overall dynamic properties are generally close to what one might estimate using quasi-static micromechanical methods, except near resonances and band gaps. The material properties and geometrical layouts are the main factors affecting the band structure. Claeys et al.11 investigated various metamaterial layouts for reducing vibrations along a known transmission path both numerically and experimentally. Wang et al.12 stud- ied the influence of geometry on band gap properties of phononic crystals and developed spring-mass models of different vibrational modes to predict band gap frequencies. Even though most applications are meant for control of acoustic waves, the solid structure can inherently maintain shear and mixed modes. The present work is particularly interested in understanding the nature of such mixed modes in mechanical metamaterials. Various methods have been developed for band structure calculation. Transfer Matrix (TM) method is widely used5,13 -- 18 in 1D periodic beams and multi-layered structure. How- ever, the TM method usually can only be applied in one direction. Plane Wave Expansion (PWE) method, which expands the material property and amplitude of Bloch wave response into Fourier series, is one of the most commonly used approaches19 -- 23 for calculating the dis- persion relations of periodic structures including both phononic and photonic crystals. The PWE method is easy to apply and has wide applications, but it has slow convergence rate for systems of large elastic mismatch. Nemat-Nasser et al.24 introduced a mixed variational method which has high accuracy and quick convergence for calculating dispersion relation and effective elastodynamic parameters of periodic elastic composites. For phononic crys- tals or metamaterials with complex microstructures, finite element method (FEM) provides efficient and precise numerical solutions. 2 Beams are commonly found in locally resonant media as they could provide a compact "spring" element. Wang and Wang25 developed spring-mass models for three-dimensional phononic crystals and evaluated the band gap edge frequencies based on effective stiffness of beams. On the other hand, as the main medium for 1D propagation with attached res- onators, Timoshenko beam theory has been used by Yu et al.13,26 to study the band structures of metamaterial beams with the transfer matrix method. Extensive studies14,16,21,27,28 have been done on locally resonant structures based on Euler-Bernoulli and Timoshenko beam theories. Wu et al.20 studied band gaps of surface waves in two-dimensional phononic structures consisting of general anisotropic materials based on PWE and reported that some of the apparent crossing points in dispersion curves are indeed sharp bends around which the modes exchange suddenly. What appears incorrectly as a crossing point could lead to incorrect sorting of the band structure. Such level repulsion in dispersion curves has been studied and reported in literatures. Wu and Huang29 analyzed level repulsions of bulk acoustic waves and pointed out that the polarizations of different modes could be used as the criterion to identify real or apparent crossing points. Numerical and experimental studies performed by Yeh et al.30 showed that the band gap induced by level repulsion strongly depend on the geometry of phononic structure. Level repulsions occur in the vicinity of exceptional points31 in complex wavenumber domain. Maznev32 analytically studied the complex band structure and eigenvectors of locally resonant media in the vicinity of exceptional point and discussed the effect of damping ratio. Lu and Srivastava33 introduced a method based on the mode shape continuity around the exceptional point to identify the crossing points in the band structure, distinguish them from level repulsion or avoided crossings, and sort the bands correctly. In this paper, the band structure of a resonator array for any wave with oblique in-plane propagation direction is investigated. The schematic of repeating unit cell (RUC) of the studied metamaterial is shown in Fig. 1. The RUC includes a T-shaped cantilever beam as the resonator and the hollow frame as the cell. It is further assumed that the system is in plane strain conditions with the thickness of the array much larger than cell and wall length scales. Such structure is easy to produce by 3D printing and can be analyzed as an assemblage of elastic beams and masses. It is expected to have local resonance at around 3 kHz based on preliminary estimates of the resonator. The incident angles of P 3 (a) (b) FIG. 1: (a) Schematic drawing of the studied 2D infinitely periodic resonator array. (b) The detailed geometry of the RUC with coordinates of each line shown in mm. and SV waves are varied from 0◦ (x direction) to 90◦ (y direction). The propagating P and SV waves are assumed to be plane harmonic Bloch waves. In what follows, we first present the band structure results obtained from numerical approach (FEM), including the dispersion diagrams and mode shapes. Next we establish an analytical method for band structure calculation based on identifying the major reduced order degrees of freedom (DOFs). Then we compare the results from two approaches and discuss how the dispersion curves change with respect to different propagation directions. Finally we present and discuss the occurrence of the avoided and real crossing points in the band structure. II. NUMERICAL CALCULATION OF BAND STRUCTURE The numerical simulation is performed using finite element software COMSOL. The dy- namic problem of elastic waves propagating in a periodic medium can be described formally as the eigenvalue problem (cid:0)K − ω2M(cid:1) U = 0, (1) where ω is the angular eigenfrequency, and the real part of u(x, y, t) = U(x, y, ω)exp [−iωt] represents a harmonic displacement field at this frequency, and K and M are the stiffness and mass operators in Fourier domain, respectively. The spatial domain of this problem is reduced to a single unit cell based on Bloch-Floquet periodicity which requires the function 4 ........................................................................yx(cid:20)(cid:19)(cid:27)(cid:17)(cid:24)(cid:19)(cid:26)(cid:17)(cid:24)(cid:19)(cid:24)(cid:17)(cid:24)(cid:19)(cid:20)(cid:17)(cid:24)(cid:19)(cid:19)(cid:19)(cid:20)(cid:17)(cid:24)(cid:19)(cid:23)(cid:17)(cid:24)(cid:19)(cid:24)(cid:17)(cid:24)(cid:19)(cid:21)(cid:17)(cid:24)(cid:19)(cid:26)(cid:17)(cid:24)(cid:19)(cid:27)(cid:17)(cid:24)(cid:19)(cid:20)(cid:19)(cid:92)(cid:91)(a) (b) FIG. 2: First Brillouin zone and complete dispersion curves of first three modes along paths connecting high symmetry k points. u0 in u = u0(x, y, ω)exp [−i (ωt − kxna − kymb)] , (2) to be periodic. Here n, m ∈ Z, and a and b are unit cell dimensions in x and y directions (in this case a = b =10 mm), respectively, and k is the wavevector. The eigenfrequencies can be obtained as functions of k by solving this system, in particular, by applying a discretized mesh and converting the operator equation above to a finite matrix over the unit cell domain. In the present study, the structure is modeled based on the material properties of VeroGray, a widely available 3D printing material (density ρ = 1161 kgm/s3, Young's modulus E = 2.978 GPa, and Poisson's ratio ν = 0.35). We define non-dimensionalized wavevector components kxa = Qx and kya = Qy in x and y directions, respectively, with Qx = Q cos θ, Qy = Q sin θ, (3) (4) where θ is the propagation angle and Q is a dimensionless parameter. The 2D band structure can be obtained by solving eigenfrequencies with parametric sweep of Qx and Qy from 0 to π. The first three mode frequencies calculated along paths connecting high symmetry k points in the Irreducible Brillouin Zone (IBZ) are shown in Fig. 2. The 3D partial band structure is shown in Fig. 3a. The frequency contour plots of the first three modes are shown in Figs. 3b to 3d, respectively. Figure 3b show the frequency contour of the first mode (SV acoustic branch). Its first quadrant (Qx > 0, Qy > 0) is 5 ΓΧΜ(cid:1863)(cid:3051)(cid:1863)(cid:3052)Y(cid:2024)(cid:1853)(cid:3398)(cid:2024)(cid:1853)(cid:3398)(cid:2024)(cid:1853)(cid:2024)(cid:1853)MXMY00.511.52Frequency (Hz)1041st2nd3rdpartially shown as the red surface in Fig. 3a. The contour curves a sharp drop in the phase around 45◦ angle of propagation. This is expected as one should note the cooperative response of the vertical and horizontal walls in bending leading to higher stiffness in the cell structure against in-plane shear at middle angles. Similar trend can also be seen in Fig. 3c, where the contour curves for higher frequencies (> 3 kHz) represent SV optical branch. The trend of P wave speed can be seen from the two inner curves in Fig. 3c, representing its acoustic branch. Unlike SV wave, the P wave speed first decreases then increases as the propagation angle changes from 0◦ to 90◦. The underlying structural resistance to in-plane axial deformation changes from stretch-dominated near 0◦ and 90◦, to bending-dominated at middle angles, representing lower stiffness and phase speeds in contrast to shear response. Figure 3d shows the contour of the third mode, which is mostly in a range beyond the considerations of this paper, but the central portion (lower frequencies) also match the P wave acoustic branch behavior and may be identified by its optical branch. Note that this representation provides more information compared to Fig. 2 as it potentially allows for inspection of interesting points in the reciprocal space not on the high symmetry paths. This is in particular important for the mechanical metamaterials as substantial mode mixing is inherent to the physics of the problem. The dispersion curves along specific directions can be obtained by fixing the propagation angle θ and plot the mode frequencies with respect to Q. Figure 4 shows the dispersion curves for 0◦, 45◦, 60◦, and 90◦ directions. Typical points are identified by circles and labeled. The mode shapes of these points are shown in Fig. 5. Based on the observation of mode shapes, it can be concluded that the unit cell and its resonator have three main DOFs: the cell displacements in x and y directions and the lateral deflection of the resonator. Other DOFs can be observed to be independent in higher frequency ranges, which is beyond the considerations of the present work. The dispersion curves in Fig. 4 and the mode shapes in Fig. 5 indicate that there are mode mixing due to the coupling in the three main DOFs. Points a1, b1, c1, and d1 are examples of primarily acoustic shear vertical (SV) modes, while points a2, b2, c2, and d2 may be considered as optical SV modes. Points a3 and a4, b3 and b4, c3 and c4, d3 and d4 are examples of primarily longitudinal wave (P) modes. Points a5, b5, c5, and d5 represent a strong relative vibration between the resonator and cell, with opposing phases. Points a6, b6, c6, and d6 represent the local resonance modes where the T shaped cantilever absorbs most of the energy. Except for 90◦ direction shown in Fig. 4d, 6 (a) (c) (b) (d) FIG. 3: Partial band structure showing first 3 modes of the studied metamaterial. Red, green, and blue surfaces in (a) represent the first, second, and third modes, respectively (color online). (b) through (d) show frequency contours of the first 3 modes, in order. the P wave acoustic branches are mixed together with SV wave optical branches. III. ANALYTICAL APPROACH BASED ON REDUCED ORDER MODEL The general idea of the reduced order analytical approach is to discretize the repeating unit cell (RUC) into a finite number of structural elements and select degrees of freedoms based on the physical expectation of the resonator behavior. An oversimplified model for the T-resonator arrays in 1D is represented in Fig. 6. Here mc is associated with the mass of the frame in each cell. There is a spring with stiffness βc between two neighbor cells. A resonator with mass mi is connected to the each cell mass by a spring with stiffness βi. The cell and resonator have horizontal DOFs uc n, respectively. The integer n represents the horizontal location of the cells and resonators in the chain and superscripts c and i n and ui 7 0100020001300040005000Frequency (Hz)60000.810.80.60.60.40.40.20.2 0-3-2-10123-3-2-101230.511.522.53Frequency (kHz)-3-2-10123-3-2-101232468101214Frequency (kHz)-3-2-10123-3-2-101235101520Frequency (kHz)(a) θ = 0◦ (b) θ = 45◦ (c) θ = 60◦ (d) θ = 90◦ FIG. 4: Dispersion curves for different propagation directions. Selected typical points are shown by small circles and labeled; see Fig. 5. represent cell/crystal and inclusion/resonator, respectively. The dynamic equations can be established for the nth cell and resonator: mc mi n + uc n−1) + βi(ui n − uc n), ∂2uc n ∂t2 = βc(uc ∂2ui n ∂t2 = βi(uc n+1 − 2uc n − ui n). (5) (6) The band structure of the 1D chain model in normal incident P wave can be obtained by solving the characteristic equation based on Eqs. (5) to (6) and plotting the eigen- frequencies. The high number of variables in these equations is naturally reduced by using Bloch-Floquet periodicity. The conceptual approach may be expanded to 2D and 3D as 8 00.511.522.530100020003000400050006000Frequency (Hz) a6 a2 a5 a3 a1 a400.511.50100020003000400050006000Frequency (Hz) b1 b2b3 b4b5 b600.20.40.60.810100020003000400050006000Frequency (Hz) c1 c2 c3 c4 c5c600.511.522.530100020003000400050006000Frequency (Hz) d1 d2 d3 d4d5 d6(a) (c) (b) (d) FIG. 5: Mode shapes of selected points in Fig. 4. FIG. 6: Schematic of an infinite 1D discrete elastic chain with local resonators in each cell. well as more complex unit cell structures. However, in the case studied here a number of further assumptions are made based on the physical understanding of the system. The RUC is re-selected as a cross-shaped cell with a T-shaped cantilever attached, as shown in Fig. 7 (in contrast with numerical approach, where the a square frame and internal T- 9 𝑥𝑦𝐐0°𝑥𝑦𝐐45°𝑥𝑦𝐐60°𝑥𝑦𝐐90°𝑚𝑐𝑚𝑐𝑚𝑐𝑚𝑐𝑚𝑖𝑚𝑖𝑚𝑖𝑚𝑖𝛽𝑐𝛽𝑐𝛽𝑐𝛽𝑐𝛽𝑖𝛽𝑖𝛽𝑖𝛽𝑖shaped cantilever were analyzed). With this selection, each cell is connected to neighbors through cuts transverse to the beam axes. The cross-shaped cell has two independent DOFs: displacement in x direction is denoted by uc n,m represents its displacement in y direction. The same could be done for the tip mass of the T-shaped cantilever, but as n,m, and vc it turns out (based on the expected physics of the resonator) the vertical displacement may be eliminated geometrically and the only independent resonator DOF is the horizontal displacement ui n,m at the center of tip mass in x direction. Index m represents the vertical location of the cells and resonators in the array. (a) (b) FIG. 7: (a) Schematic drawing of the studied metamaterial with re-drawn RUC for reduced order analytical modeling. The (n, m)th cell is colored blue in the center. (b) shows the geometry of the RUC with DOFs and reference points c and f . The trial displacement solutions that inherently satisfy Bloch-Floquet periodicity are: n,m = uc uc 0 exp [−i(ωt − kxna − kyma)] , 0 exp [−i(ωt − kxna − kyma)] , 0 exp [−i(ωt − kxna − kyma)] , ui n,m = ui vc n,m = vc (7) (8) (9) where uc 0, ui 0 and vc 0 are the complex displacement amplitudes to be determined along with ω for each pair values of kx and ky. The relationship between forces and displacements of the RUC can be established by breaking it into elastic beams and rigid bodies. To obtain the force balance equations for the beams, one could introduce more extra dependent DOFs 10 ..............................n+1, m-1..............................n-1, m............yxn, mn+1, mn+1, m+1n-1, m-1n-1, m+1n, m-1n, m+1𝑢𝑛,𝑚𝑐𝑢𝑛,𝑚𝑖𝑣𝑛,𝑚𝑐𝑣𝑛,𝑚𝑓𝑢𝑛,𝑚𝑓𝜃𝑛,𝑚𝑓𝜃𝑛,𝑚𝑐𝑎𝑓𝑐at reference points c and f , as shown in Fig. 7b. Point c is chosen at center of the cross. In addition to two primary DOFs uc and vc, it also has a rotational DOF θc as the counter clockwise rotation angle. Point f is defined at the intersection of the T-shaped resonator beam and cell frame's horizontal wall. Point f has three dependent DOFs uf , vf , and θf defined in positive x, y, and counter clockwise directions, respectively. The four dependent DOFs defined at reference points will also need to satisfy Bloch-Floquet conditions. In the following calculations, all structural elements are treated as Timoshenko beams where shear deformations are allowed. In certain cases (where physically expected) the axial stiffness of the beams are also included in the process. The force balance equations of a Timoshenko beam element (in plane strain conditions) with four DOFs can be expressed   u0 φ0 u1  =  F0 M0 F1  , φ1 M1 (10) (11) as: EI (1 − ν2)(1 + Φ)L3  −12 6L 6L 12 6L (4 + Φ)L2 −6L (2 − Φ)L2 −12 6L (2 − Φ)L2 −6L (4 + Φ)L2 −6L −6L 12 where the parameter Φ is defined as Φ = 12EI κL2GA(1 − ν2) . Here κ = 5/6 is the shear correction factor for rectangular cross section, E is Young's modulus of the material, L is the length of beam, A is the cross sectional area, I is the second moment of inertia, and G is the shear modulus. In Eq. (10), the displacement u0 and u1 denote the transverse deflections at the two ends of the beam, and the angles φ0 and φ1 represent the rotation angles of cross section at these ends. The moments and angles at the two ends of beam are positive in counter clockwise direction. a. Resonator effective mass and stiffness For mechanical metamaterials, the stop band location is mainly affected by the natural frequency of the resonator, which can be deter- mined by resonator mass and its effective stiffness. The T-shaped resonator can be regarded as a rigid tip that stays on a vertical elastic beam, and its mass per unit thickness (mi = 0.1161 g/cm in the example studied here) is chosen to be the resonator effective mass. To determine an equivalent spring-mass model, consider an imaginary concentrated force Fi applied at the center of the rigid tip in positive x direction as in Fig. 8a. In order to find the effective resonator stiffness βi, the balance equations for the vertical elastic beam can 11 (a) (b) FIG. 8: (a) Schematics of the resonator for effective stiffness calculation. All the parts are assumed rigid, except for the elastic Timoshenko vertical beam. Only Fi and the loads on the main beam are shown. Note that the bottom part is assumed to have zero thickness and is presented to streamline the process of geometric elimination of extra DOFs. (b) Axial forces on the (n, m)th cell induced by relative displacements between points c and f as well as vertical neighbor cells. be written based on Eq. (10):  −12 12 12 6Le 6Le (4 + Φe)L2 −12 6Le (2 − Φe)L2 −6Le EIe (1 − ν2)(1 + Φe)L3 e   ut θt ub θb  =  Fi −FiLr −Fi Mb  , (12) e −6Le (2 − Φe)L2 e 6Le −6Le e −6Le (4 + Φe)L2 e where subscript e refers to the properties of the elastic beam under the rigid tip, ut and ub represent the displacements at top and bottom ends of the beam in positive x direction, θt and θb are the rotation angles, Lr (= 1 mm) is the distance from center of tip mass to top of the vertical beam, and Mb is the resultant moment at the bottom. Note the value of moment at top of the beam is known and given based on the moment equilibrium of the tip mass and the assumption that its moment of inertia is negligible in this problem. As will be seen in the following, the unbalanced moment induced in the beam is in fact an important and fundamental component of the unusual behavior of the system. However, the simplifying assumption of the local equilibrium for the tip appears not to limit this effect in 12 𝐹𝑖𝐿𝑟𝐿𝑒𝑢𝑖𝐹𝑖−𝐹𝑖𝑢𝑡𝑢𝑏𝑢𝑓𝜃𝑓ℎ−𝐹𝑖𝐿𝑟𝜃𝑡𝜃𝑏𝑀𝑏(cid:1831)(cid:1827)(cid:1858)(cid:1858)(cid:1855)(cid:1840)(cid:3051)(cid:2869)(cid:1840)(cid:3051)(cid:2868)(cid:1840)(cid:3052)(cid:2868)(cid:1840)(cid:3052)(cid:2869)(cid:1853)any significant way. The kinematic relations among the DOFs are ui = ut − θtLr , ub = uf − θf h, θb = θf , (13) (14) (15) where h (= 1.5 mm) is half of the cell wall thickness. Substituting Eqs. (13) to (15) into Eq. (12) yields: Fi = βi (cid:2)ui − uf + θf (Le + Lr + h)(cid:3) , (16) (17) (18) . where the effective resonator stiffness is: Mb = Fi (Le + Lr) , βi = ((4 + Φe)L3 e + 12L2 12EIe eLr + 12LeL2 r)(1 − ν2) For the cell geometry studied here (assuming 1 cm unit thickness) βi ≈ 66 945.4 (N/cm)/m. it's assumed that its mass (mc = b. Frame dynamics For the cross-shaped cell, 0.638 55 g/cm for the example studied here) is concentrated at the center point c. To study the horizontal dynamics of the cell, we assume that its structure is fully elastic and the cross-sectional area A is uniform along the length between reference points by neglecting the lateral constraints at the ends of beams. The axial forces applied to the (n, m)th cell , see Fig. 8b, can be expressed as Ny1 = EA (1 − ν2)a Ny0 = Nx1 = Nx0 = EA (1 − ν2)a 2EA (1 − ν2)a 2EA (1 − ν2)a (cid:0)vc (cid:0)vc (cid:16) (cid:0)−uf n,m+1 − vc n,m n,m − vc n,m−1 n+1,m − uc uf n,m , n,m + uc n,m (cid:1) , (cid:1) , (cid:17) (cid:1) . (19) (20) (21) (22) Note that the vertical DOF on the top and bottom of the cross shaped cell has been elimi- nated based on the neighboring cell DOFs; See Fig. 8b. The DOF at f , however, may not be eliminated like this due to the influence of the T-shaped resonator. There are also shear forces applied to the cell due to beam deformations, as shown in Fig. 9. Note the selection of rigid portions in this model. Subscript c denotes the vertical 13 (a) (b) FIG. 9: The shear forces of (n, m)th cell in y and x directions, associated with (a) horizontal and (b) vertical cell walls, respectively. The parts that include points c and f in this model are assumed to be rigid. The narrow part with point f in (a) is the same as the one shown in Fig. 8a. beam (cell wall) and has length Lc = a − 2h. Subscript f denotes the horizontal beam between the narrow part including node f and the square at te center of the cross, containing node c, with length Lf = a/2−h. The shear forces applied to the (n, m)th cell can be written based on Timoshenko beam theory:     ,  ,  , Vy1 = BV f n,m − θc vc n,mh vf n,m θf n,m θc n,m Vy0 = −BV f vf n+1,m −θf n+1,m vc n,m + θc −θc n,m n,mh Vx1 = BV c uc n,m+1 + θc n,m+1h θc n,m+1 n,m − θc uc n,mh θc n,m 14 (23) (24) (25) (cid:1831)(cid:1835)(cid:3033)(cid:884)(cid:1860)(cid:1874)(cid:3041)(cid:481)(cid:3040)(cid:3030)(cid:2016)(cid:3041)(cid:481)(cid:3040)(cid:3030)(cid:1874)(cid:3041)(cid:481)(cid:3040)(cid:3033)(cid:2016)(cid:3041)(cid:481)(cid:3040)(cid:3033)(cid:1831)(cid:1835)(cid:3033)(cid:1857)(cid:2879)(cid:3036)(cid:3038)(cid:3299)(cid:3028)(cid:1848)(cid:3052)(cid:2868)(cid:1848)(cid:3052)(cid:2869)(cid:1848)(cid:3052)(cid:2869)(cid:1848)(cid:3052)(cid:2869)(cid:1848)(cid:3052)(cid:2869)(cid:1848)(cid:3052)(cid:2868)(cid:1848)(cid:3052)(cid:2868)(cid:1848)(cid:3052)(cid:2868)(cid:133)(cid:1)(cid:15)(cid:3)(cid:15)(cid:1858)(cid:884)(cid:1860)(cid:2016)(cid:3041)(cid:481)(cid:3040)(cid:3030)(cid:1873)(cid:3041)(cid:481)(cid:3040)(cid:3030)(cid:1848)(cid:3051)(cid:2869)(cid:1848)(cid:3051)(cid:2869)(cid:1848)(cid:3051)(cid:2869)(cid:1848)(cid:3051)(cid:2868)(cid:1831)(cid:1835)(cid:3030) Vx0 = −BV c  , n,m−1h n,m−1 − θc uc −θc n,m−1 n,m + θc uc −θc n,m n,mh (cid:104) (26) (27) (cid:105) , where BV α = EIα (1 + Φα)(1 − ν2)L3 α 12 6Lα −12 6Lα with α = f, c. Note that in Eqs. (23) to (26), the displacements at the beam ends are expressed not only in terms of uc and vc, but also θc and h. This is due to kinematic relations based on rigidity of the relevant parts. Such assumption will inevitably increase the calculated SV wave velocity (the slope of SV wave branch). Another simpler assumption is that every point in the rigid portion have the same DOFs, and consequently in Eqs. (23) to (26) the terms with h in the displacement vectors will no longer appear. This will lead to a further overestimation of the SV wave speed. On the other hand, if we assume the effective beam lengths reach all the way through previously assumed rigid parts, e.g. to point c at the center of square, the formulations may be rewritten with h = 0 and the SV wave speed will become much lower than numerical simulation results. The presented model leads to relatively accurate results while keeping the process as simple as possible. (a) (b) FIG. 10: The moments applied on the square edges in (n, m)th cell are shown in (a) with directions labeled. (b) shows the free body diagram for the vertical rigid line where point f stays in the middle. The yellow line shown is at the left side of (n, m)th cross-shaped cell. The moments applied at the left, right, top, and bottom sides of the rigid square contain- ing point c in (n, m)th cell are denoted as McL, McR, McU , and McD, respectively, as shown 15 (cid:1839)(cid:3030)(cid:3022)(cid:1839)(cid:3030)(cid:3005)(cid:1839)(cid:3030)(cid:3013)(cid:1839)(cid:3030)(cid:3019)(cid:1858)(cid:1839)(cid:3029)(cid:1832)(cid:3036)(cid:1839)(cid:3033)(cid:3019)(cid:1839)(cid:3033)(cid:3013)(cid:1848)(cid:3052)(cid:2869)(cid:1857)(cid:2879)(cid:3036)(cid:3038)(cid:3299)(cid:3028)(cid:1848)(cid:3052)(cid:2868)(cid:1840)(cid:3051)(cid:2868)(cid:1857)(cid:2879)(cid:3036)(cid:3038)(cid:3299)(cid:3028)(cid:1840)(cid:3051)(cid:2869)in Fig. 10a. Using the similar process as the one used for shear forces, the moments applied to the square can be written as McL = BM f McR = −BM f McU = BM c McD = −BM c where BM α = EIα (1 + Φα)(1 − ν2)L3 α (cid:104)   vf n,m θf n,m n,m − θc vc n,mh n+1,m θc n,m −vf θf n+1,m n,m − θc θc n,m −vc n,mh n,m+1 + θc uc n,m+1h θc n,m+1 n,m − θc uc n,mh θc n,m −uc n,m−1 + θc θc n,m−1 n,m − θc θc n,m −uc n,m−1h n,mh  ,  ,  ,  ,    (28) (29) (30) (31) (32) (cid:105) , (33) (34) 6Lα (2 − Φα)L2 α −6Lα (4 + Φα)L2 α with α = f, c. The moments applied at the left and right sides of reference point f are denoted as Mf L and Mf R, as shown in Fig. 10b, and they can be written as: Mf L = BM f Mf R = −BM f  ,  . n−1,mh vc n−1,m + θc θc n−1,m vf n,m θf n,m −vc n,m + θc n,mh θc n,m −vf θf n,m n,m  16 All previous calculations are based on elements in the (n, m)th cell, and all the loads required for deriving dynamic equations of the RUC have been expressed. Note that the yellow line shown in Fig. 10b is massless and has no width, therefore the forces and moments applied to it should be balanced statically. We also assume that the moments around point c are balanced, i.e. ignore its moment of inertia. Combined with Bloch-Floquet periodicity: Vy1 = e−ikxaVy0 , the dependent DOF vf can be eliminated and expressed as vf 0 = e−ikxa + 1 2 vc 0 + e−ikxa − 1 4 (2h + Lf )θc 0 . (35) (36) The equations of motion for the three primary DOFs uc, ui, and vc can now be written as mc mi (mc + mi) ∂2uc ∂2ui n,m n,m ∂t2 = Nx1 − Nx0 + Vx1 − Vx0 , ∂t2 = −Fi , ∂t2 = Ny1 − Ny0 + Vy1 − Vy0 , n,m ∂2vc (37) (38) (39) The remaining DOFs, uf , θf , and θc , may be eliminated using other static balance equations at the reference points: Fi + Nx0 − e−ikxaNx1 = 0, Mf L − Mf R + Mb + Fih = 0, McL + McU − McR − McD + h(Vx0 + Vx1 + Vy0 + Vy1) = 0. (40) (41) (42) Alternatively and in general, the displacement vectors for the three primary DOFs and three dependent DOFs can be denoted as Up and Ud, respectively: Combining the three equations of motion Eqs. (37) to (39) and the three balance equations Eqs. (40) to (42), we obtain: . , (cid:105)(cid:62) (cid:105)(cid:62)  = 0, uc 0 ui 0 vc 0 0 θf uf 0 θc 0 Up Ud Up = Ud = (cid:104) (cid:104) Dpp Kpd Kdp Kdd 17 (43) (44) (45) where Kpd, Kdp, and Kdd are 3 × 3 stiffness sub-matrices, and Dpp is the 3 × 3 diagonal dynamic stiffness sub-matrix, which includes −ω2Mpp in addition to the structural stiffness Kpp. Note that all the components in Ud are assumed to be dependent, and we only need Up to describe the mode shapes fully. Hence Eq. (45) can also be written as: DUp = 0, (46) where D = Kpp − KpdK −1 dd Kdp is the system's overall dynamic stiffness matrix. After substituting the parameters, it's easy to prove that D is Hermitian, thus having only real eigenvalues. For any given wavenumber and incident angle, the first three mode frequencies (cid:12)(cid:12)(cid:12)D (cid:12)(cid:12)(cid:12) = 0. of the reduced order model can be obtained by solving the characteristic equation Plotting the mode frequencies with respect to wavenumber Qx and Qy yields the 3D band structure of oblique propagating waves. The eigen-vectors associated with these eigen- frequencies represent the mode shapes in the reduced order model. IV. RESULT AND DISCUSSION The asymmetric nature of this unit cell has created a coupling between P and SV modes that leads to quite interesting phenomena. While it may appear that the resonators only interact with P waves, the induced moment from the motion of the head interacts with the shear wave in the cell walls. The resulting band structure, therefore, demonstrates inherent and fundamental mode mixing. The first 3 modes of the band structure for oblique in-plane waves are shown in Fig. 11 and comparison is made between the numerical COMSOL results and those of the reduced order analytical approach. Using the symmetry, the band structure is plotted only in the region where Qx and Qy are positive and vary from 0 to π. Since the frequency band gap is expected to be located at around 3 kHz, the frequency range of interest will be 0 to 6 kHz. Generally the results from analytical method agree well with the band structure from numerical approach. FE results are shown as colored surfaces while the dots represent the mode frequencies calculated using the analytical method. Figures 11a to 11c show the first, second and third modes, respectively. Figure 11a shows the acoustic branch of oblique SV wave. Figure 11b shows the mixed branch of P and SV wave modes. Figure 11c shows the optical branch of oblique P wave. In the quasi-static limit, as wavenumber Q 18 (a) (b) (c) FIG. 11: Comparison of the 3D band structure under 6000 Hz. The colored surfaces are results from numerical method. The dots are obtained from analytical approach. (a) through (c) show the first, second and third modes, in order. approaches the origin, the slopes of P wave and SV wave acoustic branches are proportional to the wave phase velocities in the media, and they are well approximated by the results from analytical calculation. However, it can be clearly observed that the entire calculated band gap is about 700 Hz higher than expected. The reason for such discrepancy is that the calculated resonator stiffness βi is overestimated by assuming part of the structure as rigid, thus rendering the resonance frequency higher than numerical simulation result. Furthermore, the rotational moment of inertia, particular at the resonator head is neglected. Including this rotational DOF at the resonator head should improve the results, though. 19 9/11/2018f1.svgfile:///C:/Users/Weidi_Wang/Desktop/Tangle%20july/3d%20compare/f1.svg1/19/11/2018f2.svgfile:///C:/Users/Weidi_Wang/Desktop/Tangle%20july/3d%20compare/f2.svg1/19/11/2018f3.svgfile:///C:/Users/Weidi_Wang/Desktop/Tangle%20july/3d%20compare/f3.svg1/1While easily doable, the 4 DOF analysis is more convoluted without adding major clarity to the physical understanding of the problem. Instead, one can introduce a correction by recalculating βi in such a way that the resonance frequency of an isolated discrete resonator, ωi =(cid:112)βi/mi, matches any estimate of the actual isolated resonator, i.e. with fixed boundary conditions at the root. With this correction, the band structure along paths connecting the corners of the irreducible Brillouin zone was calculated. Figure 12 shows the results of this calculation, which turn out to be very close with FE results near resonances (reproduced from Fig. 2). In the remainder of this paper, however, we have not used this correction to maintain consistency. Figures 13a to 13f show partial dispersion diagrams for 0◦, 10◦, 45◦, 60◦, 89.5◦, and 90◦ propagation directions in order. FIG. 12: Comparison of dispersion curves along paths connecting high symmetry k points from FE simulations (Fig. 2) and based on the 3 DOF reduced order approach with adjusted βi. The pattern of transition in dispersion curves can be seen in Fig. 13. The slopes of the two branches at the origin are related to wave velocities. The P wave velocity in the metamaterial reaches its minimum while SV wave velocity reaches its maximum when the waves are propagating at 45◦. As shown in Fig. 13c, the mixed branch of P and SV wave modes becomes almost a straight line. As the angle increases, P wave velocity increases after 45◦ while SV wave velocity decreases. As the angle increase from 60◦ towards 90◦, avoided crossing, or level repulsion in band structure becomes more evident. The SV wave optical and P wave branches for 60◦ incidence almost intersect each other, as shown in Fig. 13d. However, the two branches exchange 20 MXMY0200040006000800010000Frequency (Hz)NumericalAdjusted Analytical(a) (c) (e) (b) (d) (f) FIG. 13: Partial band structure comparison. The red curves represent analytical results while the black ones are numerical results. (a) through (f) are the band structures for 0◦, 10◦, 45◦, 60◦, 89.5◦, and 90◦ incidences, in order. their slopes at the apparent crossing point and avoid each other due to level repulsion. If the numerical approach is performed on coarse wavenumber discretization33, the avoided branches will be easily mistaken as real crossing branches. We note that the anti-crossing 21 01230100020003000400050006000Frequency (Hz)Numerical Analytical01230100020003000400050006000Frequency (Hz)Numerical Analytical01230100020003000400050006000Frequency (Hz)Numerical Analytical01230100020003000400050006000Frequency (Hz)Numerical Analytical(cid:19)(cid:20)(cid:21)(cid:22)(cid:19)(cid:20)(cid:19)(cid:19)(cid:19)(cid:21)(cid:19)(cid:19)(cid:19)(cid:22)(cid:19)(cid:19)(cid:19)(cid:23)(cid:19)(cid:19)(cid:19)(cid:24)(cid:19)(cid:19)(cid:19)(cid:25)(cid:19)(cid:19)(cid:19)(cid:41)(cid:85)(cid:72)(cid:84)(cid:88)(cid:72)(cid:81)(cid:70)(cid:92)(cid:3)(cid:11)(cid:43)(cid:93)(cid:12)N(cid:88)(cid:80)(cid:72)(cid:85)(cid:76)(cid:70)(cid:68)(cid:79)(cid:3) A(cid:81)(cid:68)(cid:79)(cid:92)(cid:87)(cid:76)(cid:70)(cid:68)(cid:79)(cid:3)01230100020003000400050006000Frequency (Hz)Numerical Analyticalof dispersion curves can always be found until the propagation angle reaches exactly 90◦. For instance, at 89.5◦ the dispersion curves seem to have a real crossing point in Fig. 13e. However, zooming in the vicinity of the apparent crossing point indicates the opposite, as shown in the bottom right corner in Fig. 13e. Figure 13e implies mode anti-crossing instead of true intersection at 89.5◦ incidence, based on both numerical and reduced order analytical approaches. The reason for such avoided crossing is the small but non-negligible coupling between P and SV modes. As the P wave propagates at 89.5◦, although it is almost in y direction, there is still a wave component in x direction that causes the local resonance. In comparison, P wave is propagating in pure y direction at 90◦ incidence and no local resonance occurs within the frequency range of interest. P wave and SV wave become decoupled only at 90◦ incidence and P wave acoustic branch has a real crossing point with SV wave optical branch. This can be further confirmed by the dynamic matrix in Eq. (46) from the analytical method. Only when propagation angle is exactly 90◦, Eq. (46) takes the form: In other words since kx = 0 at 90◦ angle, the stiffness components D13, D23, D31 and D32 in Eq. (46) also vanish, thus making the cell displacement in y direction decoupled from the other two DOFs in x direction. Therefore, the eigenvectors must have the form: 0 are completely decoupled with uc where C1 and C2 are non-zero displacement components. Therefore at 90◦, the eigenvectors associated with vc 0. Such real crossing point at 90◦ incidence has two degenerate eigenvalues but one can still find two associated linearly independent eigenvectors. However once the propagation angle changes from exactly 90◦, the coupling terms D13, D23, D31 and D32 appear again. Close to 90◦, D13, D23, D31 and D32 have smaller values, and shaper bending occurs in the dispersion curves. 0 and ui The correct sorting of band structure is of prime importance as pointed out elsewhere33. One way to identify different branches is zooming in the vicinity of apparent crossing point, as shown in Fig. 13e. However, if the band structure is calculated by finite element method, 22 D11 D12 D12 D22  = 0.  uc 0 ui 0 vc 0 0 0 0 0 D33  = uc 0 ui 0 vc 0 C1 C2  or 0 0  , 0 1 (47) (48) the zoom in method requires a very high level of resolution in wavenumber, which is com- putationally inefficient. Another technique proposed in literature29,30 uses the polarizations of modes as the criterion to distinguish real or apparent cross points. This method is based on the fact that the displacement fields change rapidly around the apparent crossing point. In Fig. 14a we labeled four points e1, e2, e3, and e4. The corresponding mode shapes of these 4 points are obtained from finite element simulation and shown in Fig. 14b. (a) (b) FIG. 14: (a) The partial band structure in the vicinity of the apparent crossing point at propagation angle 89.5◦. The primed points en' are associated with the reduced order analytical solution. (b) The mode shapes at points e1, e2, e3, and e4 based on FE simulation results. The corresponding band structure and mode shape figures for 90◦ incidence are not shown here as they appear identical to (a) and (b), though with the important difference that the branches indeed cross in the real plane. Since these four points are chosen far away from the level repulsion area and are based on coarse discretization of wavenumber, it is very difficult to distinguish the difference between mode shapes at e1 and e2, or between e3 and e4, as shown in Fig. 14b. The similar mode shapes may make one believe that points e1 and e2 belong in the same branch and points e3 and e4 are on the other presumably continuous branch, which are both incorrect conclusions. In other words, when the points are chosen not close enough to the apparent crossing point, the difference in their eigenvectors could be very small, and the simulated mode shapes could lead to incorrect sorting. To further demonstrate the changes in mode shapes around apparent crossing points, we picked the corresponding points e1', e2', e3', and e4' from the analytical result, as shown in Fig. 14a, and calculated their eigenvectors, see Table I. The 23 00.20.40.60200040006000Frequency (Hz) e1 e1' e3 e2 e2'e4'e4 e3'TABLE I: Calculated displacement eigenvectors for 89.5◦ and 90 incidences. 89.5◦ 90◦ uc 0 ui 0 vc 0 uc 0 ui 0 vc 0 e1' e2' e3' e4' 5.04e-2-i1.76e-1 1.37e-1-i1.59e-1 7.91e-4+i1.45e-4 1.37e-3-i6.36e-4 -2.58e-1+i9.49e-1 -6.09e-1+i7.64e-1 1.05e-3-i7.31e-4 -1.63e-3+i3.08e-3 1.36e-4 -8.43e-4 9.99e-1 9.99e-1 1.80e-1+i4.80e-3 1.89e-1+i1.36e-2 -9.84e-1 -9.81e-1 0 0 0 0 1 0 0 1 eigenvectors at 89.5◦ show that the displacement of the cell in y direction, vc 0, changes its sign between points e1' and e2', and the change in sign also occurs for the displacement of resonator in x direction, ui for which the dominant components of the the eigenvectors (ui 0, between e3' and e4'. Note that we ensured to select eigenvectors, 0 in the former case and vc 0 in the latter) are essentially unchanged, so the sign or phase changes in other components have physical meaning in terms of change in mode shape. The small amplitudes of the components that change signs compared to the dominant components make identifying such differences from the simulated mode shapes difficult. The third component of the 89.5◦ eigenvector is not significantly affected in the transition from e1' to e2', but in the case of transition from e3' to e4', it moves one quadrant in the complex domain (sign of the imaginary part changes). These sign flips (in 89.5◦ case) prove that the points e1 and e2 are not on the same branch and neither are e3 and e4, i.e. we are observing an avoided level crossing. Furthermore, for 89.5◦ incidence, all the eigenvalue components have non- zero value, which indicates coupling between displacement components, in contrast to the decoupled case of 90◦ as shown in Eq. (48). These points are located in the 90◦ incidence band structure at the same Q values with e1' to e4' and their eigenvectors are shown in Table I. The eigenvector components with zero values indicate decoupling of motions. The fact that there is no change of sign in eigenvectors also proves real branch crossing at 90◦ incidence. A similar analysis may be performed on the mode shapes or eigenvectors calculated in the FE calculations. To demonstrate this, one can define aggregate cell and inclusion 24 displacements by averaging the point-wise quantities over the associated domains. Then, normalized displacement eigenvector components may be defined as j = ¯uc d1 j/¯vc j, d2 j = ¯ui j/¯vc j, j = ¯uc d3 j/¯ui j, (49) (50) (51) where ¯ denotes the average numerical value, and j represents the mode number in the order of increasing frequency. Note that these three values are obviously inter-dependent and are depicted here only for clarity of presentation. Further, define amplitude and argument of these quantities as dk j = rk j eiφk j . (52) Results are plotted in Fig. 15 for θ = 85◦, 89.5◦, 90◦ propagation angles from Q = 0.15 to Q = 0.2, where the second and third mode branches (j = 2, 3) have apparent (85◦ and 89.5◦) or real (90◦) crossings. The amplitudes rk j could become very big or very small, which makes it difficult to really distinguish mode shapes that may have very different phases. The transformed trajectories shown in Fig. 15 allows one to separate them. Nevertheless, near the apparent or real crossings, one would still need relatively high resolution to distinguish the modes. In other words, the mode shape analysis presented here does not resolve the issues that prompted the sorting algorithm proposed by Lu and Srivastava33. The case of θ = 90◦ requires further analysis for two reasons. First, regardless of the resolution, the trajectories of the two modes j = 2, 3 appear to intersect. Furthermore, the amplitudes become extremely small (see amplitudes in Fig. 15d), and the phases appear to be very close as well (when r → 0 an apparent π jump in phase as seen in Fig. 15d does not indicate discontinuity). It is fair to say that in these cases dk j for k = 1, 2 is essentially zero and j ≈ const. In other words, at the exact intersection of the two branches, the eigenvalues of two very different eigenvectors coalesce at one point. the mode shape is a pure d3 j = ¯uc j/¯ui Therefore, in order to continuously follow the same physical modes (i.e. the one with very small r1 j and r2 j versus the one with very large values of these quantities) across this point, one has to switch from j = 2 branch to j = 3 branch and vice versa, as also seen in Fig. 15d j → −∞ by an inconsequential ∆ϕ = π shift in phase (as by following the paths when log rk it is also the case, analytically, when log rk j → ∞). 25 (a) θ = 85◦ (b) θ = 89.5◦ (c) θ = 90◦ (d) θ = 90◦ FIG. 15: Representation of mode shape behavior near apparent (θ = 85◦, 89.5◦) and real (θ = 90◦) crossings based on quantities defined in Eqs. (49) to (52). The high resolution insets in (b) show that the mode shapes do not intersect. The different representation in (d) depicts the difference between the modes in the real crossing at θ = 90◦ more clearly as the transformation used in (c) artificially connects the neighborhoods r → 0 and r → ∞ through a π change in phase. V. SUMMARY AND CONCLUSIONS This paper presents a numerical and analytical study of the band structure of an acoustic metamaterial for oblique in-plane waves. An analytical method with reduced order model is established to approximate the band structure. Dispersion curves and mode shapes for different incident angles are shown. The studied metamaterial exhibits mixed branch of P and SV wave modes except when the propagation angle is exactly 90◦. We also report on the 26 50.16500.1800.2-5-5(cid:24)(cid:20)(cid:19)(cid:19)(cid:17)(cid:20)(cid:25)(cid:19)(cid:17)(cid:20)(cid:27)(cid:19)(cid:19)(cid:19)(cid:17)(cid:21)(cid:16)(cid:24)(cid:16)(cid:20)(cid:19)100.160.18100.200-10-10(cid:20)(cid:19)(cid:25)(cid:21)(cid:19)(cid:17)(cid:20)(cid:25)(cid:22)(cid:16)(cid:21)(cid:21)(cid:20)(cid:19)(cid:17)(cid:20)(cid:27)(cid:16)(cid:25)(cid:19)(cid:16)(cid:20)(cid:19)(cid:17)(cid:21)(cid:16)(cid:20)(cid:19)(cid:16)(cid:21)(cid:16)(cid:22)occurrence of avoided level crossings, which are related to the existence of exceptional points in complex domain33. With relatively high computational efficiency, the analytical method can calculate partial band structure which agrees well with numerical results, particularly when adjusted for the exact resonance frequency of a single discrete resonator (with rigid boundaries and not connected to the cell) and can provide support in mode identification and band sorting. With proper adjustments in parameters, this analytical method will be applicable for other metamaterials that have similar unit cell structure. REFERENCES 1Z. Liu, Science 289, 1734 (2000). 2J.-H. Sun and T.-T. Wu, Phys. Rev. B 74, 174305 (2006). 3J.-H. Sun and T.-T. Wu, Phys. Rev. B 76, 104304 (2007). 4M. S. Kushwaha, P. Halevi, G. Mart´ınez, L. Dobrzynski, and B. Djafari-Rouhani, Phys. Rev. B 49, 2313 (1994). 5S. Nemat-Nasser, H. Sadeghi, A. V. Amirkhizi, and A. Srivastava, Mech. Res. Commun. 68, 65 (2015). 6C. Ding, L. Hao, X. Zhao, C. Ding, L. Hao, and X. Zhao, J. Appl. Phys. 108, 074911 (2010). 7Y. M. Seo, J. J. Park, S. H. Lee, C. M. Park, C. K. Kim, and S. H. Lee, J. Appl. Phys. 111 (2012), 10.1063/1.3676262, arXiv:1102.1767. 8Y. Li, J. Lan, B. Li, X. Liu, and J. Zhang, J. Appl. Phys. 120 (2016), 10.1063/1.4964734. 9H. Chen and C. T. Chan, Appl. Phys. Lett. 91, 1 (2007). 10A. N. Norris and A. L. Shuvalov, Wave Motion 48, 525 (2011), arXiv:1103.6045. 11C. Claeys, N. G. Rocha de Melo Filho, L. Van Belle, E. Deckers, and W. Desmet, Extrem. Mech. Lett. 12, 7 (2017). 12Y. F. Wang, Y. S. Wang, and X. X. Su, J. Appl. Phys. 110 (2011), 10.1063/1.3665205. 13D. Yu, Y. Liu, G. Wang, H. Zhao, and J. Qiu, J. Appl. Phys. 100 (2006), 10.1063/1.2400803. 14Z. Wang, P. Zhang, and Y. Zhang, Math. Probl. Eng. 2013, 1 (2013). 15A. Sridhar, V. G. Kouznetsova, and M. G. Geers, Int. J. Mech. Sci. 133, 188 (2017). 16S. Zuo, T. Ni, X. Wu, and J. Fan, J. Vib. Control 23, 1663 (2017). 27 17A. V. Amirkhizi, Mech. Mater. 114, 76 (2017), arXiv:arXiv:1011.1669v3. 18A. Nanda and M. Karami, J. Sound Vib. 424, 120 (2018). 19D. J. Mead, J. Sound Vib. 190, 495 (1996). 20T.-T. Wu, Z.-G. Huang, and S. Lin, Phys. Rev. B 69, 094301 (2004). 21Y. Xiao, J. Wen, and X. Wen, Phys. Lett. A 376, 1384 (2012). 22M. Oudich, X. Zhou, and M. Badreddine Assouar, J. Appl. Phys. 116 (2014), 10.1063/1.4901997. 23H. Zhu and F. Semperlotti, J. Appl. Phys. 116 (2014), 10.1063/1.4894279. 24S. Nemat-Nasser, J. R. Willis, A. Srivastava, and A. V. Amirkhizi, Phys. Rev. B 83, 1 (2011). 25Y.-F. Wang and Y.-S. Wang, J. Vib. Acoust. 135, 41009 (2013). 26D. Yu, Y. Liu, H. Zhao, G. Wang, and J. Qiu, Phys. Rev. B 73, 064301 (2006). 27Y. Xiao, J. Wen, G. Wang, and X. Wen, J. Vib. Acoust. 135, 041006 (2013). 28G. Hu, L. Tang, and R. Das, J. Appl. Phys. 123 (2018), 10.1063/1.5011999. 29T.-T. Wu and Z.-G. Huang, Phys. Rev. B 70, 214304 (2004). 30S. L. Yeh, Y. C. Lin, Y. C. Tsai, T. Ono, and T. T. Wu, Ultrasonics 71, 106 (2016). 31W. D. Heiss and A. L. Sannino, J. Phys. A. Math. Gen. 23, 1167 (1990). 32A. A. Maznev, J. Appl. Phys. 123 (2018), 10.1063/1.5012999. 33Y. Lu and A. Srivastava, J. Mech. Phys. Solids 111, 100 (2018), arXiv:1706.03857. 28
1711.11172	1	1711	2017-11-30T00:48:22	Complementary logic operation based on electric-field controlled spin-orbit torques	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Spintronic devices as alternatives to traditional semiconductor-based electronic devices attract considerable interest as they offer zero quiescent power, built-in memory, scalability, and reconfigurability. To realize spintronic logic gates for practical use, a complementary logic operation is essential but still missing despite a recent progress in spin-based logic devices. Here, we report the development of a complementary spin logic device using electric-field controlled spin-orbit torque (SOT) switching. In heavy metal/ferromagnet/oxide structures, the critical current for SOT-induced switching of perpendicular magnetization is efficiently modulated by an electric field via voltage-controlled magnetic anisotropy (VCMA) effect in a non-volatile manner. Moreover, the polarity of the VCMA is tuned by the modification of oxidation state at the ferromagnet/oxide interface. This allows us to fabricate both n-type and p-type spin logic devices and to enable a complementary logic operation, paving the way for the development of non-volatile and reconfigurable logic devices.	physics.app-ph	physics	Complementary logic operation based on electric-field controlled spin-orbit torques Seung-heon Chris Baek1,2,, Kyung-Woong Park1,3,, Deok-Sin Kil3, Kyung-Jin Lee4,5 and Byong-Guk Park1,** 1Department of Materials Science and Engineering and KI for Nanocentury, KAIST, Daejeon 34141, Korea 2School of Electrical Engineering, KAIST, Daejeon 34141, Korea 3Research and Development Division, SK Hynix Semiconductor, Inc., Gyeonggi-do 17336, Korea 4Department of Materials Science and Engineering, Korea University, Seoul 02841, Korea 5 KU-KIST Graduate School of Converging Science and Technology, Korea University, Seoul 02841, Korea These authors equally contributed to this work. *Corresponding emails: [email protected] (B.-G.P.). Spintronic devices as alternatives to traditional semiconductor-based electronic devices attract considerable interest as they offer zero quiescent power, built-in memory, scalability, and reconfigurability. To realize spintronic logic gates for practical use, a complementary logic operation is essential but still missing despite a recent progress in spin-based logic devices. Here, we report the development of a complementary spin logic device using electric-field controlled spin-orbit torque (SOT) switching. In heavy metal/ferromagnet/oxide structures, the critical current for SOT-induced switching of perpendicular magnetization is efficiently modulated by an electric field via voltage- controlled magnetic anisotropy (VCMA) effect in a non-volatile manner. Moreover, the polarity of the VCMA is tuned by the modification of oxidation state at the ferromagnet/oxide interface. This allows us to fabricate both "n-type" and "p-type" spin logic devices and to enable a complementary logic operation, paving the way for the development of non-volatile and reconfigurable logic devices. Spintronic devices for information processing have attracted considerable interest owing to their potential advantages of improved computational capability and reduced power consumption [1-3]. Up to now, spin-based logic devices have been developed in two different approaches. The first is to add the electron spin as a variable parameter into existing semiconductor devices [4], such as the spin field-effect transistor (FET) proposed by Datta and Das [5]. Despite recent successes in the operation of the spin FET with a semiconducting channel [6-8], further developments are required to improve the signal at room temperature and to secure the device scalability. The second approach is to utilize metallic structures in which a logic operation is performed by using domain wall motion [9-12], magnetization switching [13,14], and spin wave propagation [15]. These metal-based spintronic devices are advantageous as they operate at room temperature. However, the screening effect of metals inhibits implementing electric-gating-induced field-effect functions, which are indispensable for complex logic operations. More importantly, a spintronic device performing complementary logic operation analogous to the conventional CMOS has not been developed yet. In this work, we experimentally demonstrate a spin logic device capable of performing complementary logic operation which is based on the spin-orbit torque (SOT) [16-21] and its controllability using an electric field [22,23]. The SOT is a spin-transfer torque originating from the spin-orbit coupling in heavy metal (HM)/ferromagnet (FM)/oxide structures and enables magnetization switching by an in-plane current. The critical current of the SOT- induced switching can be modulated by a perpendicular gate voltage (see Fig. 1a) which induces the change in the magnetic anisotropy of the trilayer structures via voltage-controlled magnetic anisotropy (VCMA) effect [24-32]. Therefore, the combination of SOT and VCMA allows us to control our device to be either switchable or non-switchable for a given input clock current. This is the first working principle of our spin logic device. The second principle is based on the finding that the polarity of the electric-field effect on the magnetic anisotropy (or VCMA effect) can be reversed by modulating oxidation state at the FM/oxide interface. As we show below, for an under-oxidized interface, a positive (negative) gate voltage decreases (increases) the magnetic anisotropy, whereas for an over-oxidized interface, the reversed VCMA effect occurs. This opposite dependence of VCMA effect on the sign of gate voltage allows us to form an "n-type" or a "p-type" spin logic device by controlling the oxidation state. Having these two types of devices (i.e., n-type and p-type), we realize the complementary functionality of spin logic devices as we demonstrate below. Electric-field controlled spin-orbit torque switching In order to prove the working principles in our devices, we carry out SOT-switching experiment combined with the application of a gate voltage for Ta(5 nm)/CoFeB(1 nm)/MgO(1.6 nm)/AlOx(1.8 nm) Hall bar structures, where perpendicular magnetic anisotropy (PMA) is established in the CoFeB (see Fig. 1a and Methods). We integrate the gate oxide (ZrO2) and electrode (Ru) on top of a square-shaped CoFeB/MgO/AlOx island for gating. We vary the oxidation time (tox) for AlOx formation from 25 sec to 125 sec in order to control the oxidation state at the CoFeB/MgO interface and to modify the dependence of magnetic anisotropy on the voltage polarity. In order to estimate VCMA effect, we measure the anomalous Hall resistance (Rxy) as a function of the in-plane magnetic field (Bx) under a gate voltage (VG), where Rxy represents the z-component of magnetization. Analysing the decrease in Rxy with increasing Bx allows us to estimate the perpendicular anisotropy field (HK). We find that the HK maximizes at tox = 75 sec, referred to as the optimal oxidation time tox,op, and slightly decreases for tox either longer or shorter than tox,op (See Supplementary Note 1). Figures 1b – 1d show the dependence of magnetic anisotropy on the gate voltage polarity for samples with different oxidation times. The sample with tox = 25 sec (< tox.op) shows that the Bx–dependent decreasing rate of Rxy is larger for VG of +22 V than for VG of -20 V (Fig. 1b), indicating that a negative (positive) VG increases (decreases) HK. Remarkably, the VCMA effect is reversed in its sign for the sample with tox = 125 sec (> tox,op; Fig. 1d). On the other hand, the VCMA effect is negligible in the sample with tox = 75 sec (= tox,op; Fig. 1c). The dependence of VCMA effect on tox is summarized in Fig. 1e. This dependence can be explained by the modification in the oxidation state at the CoFeB/MgO interface and the resultant change in PMA [33]. The application of an electric voltage induces the relocation of oxygen ions either towards or away from the interface depending on its polarity [34], which modulates HK accordingly. For example, in the under-oxidation condition (tox < tox.op), a negative bias supplies oxygen ions to the interface that is initially at the oxygen-deficient state, and consequently increases HK. The opposite VCMA effect occurs in the over-oxidation condition (tox > tox.op) for which a negative bias decreases HK. This voltage-induced change in oxidation state is supported by a non- volatile behaviour of the VCMA effect; HK remains the same even after VG is removed (Supplementary Note 2). We next investigate the effect of VG on SOT-induced switching performance. We note that in all SOT switching experiments, we apply an external in-plane field along the current direction (+x direction) for the deterministic switching. We note that this external field can be replaced by an internal exchange bias field [35,36]. The critical current (IC) for SOT-induced magnetization switching is known to be proportional to HK [37] so that VG is able to control IC through the modification of HK. As expected, the SOT-induced switching experiment under VG for the samples with different tox's demonstrates a clear dependence of IC on VG (Figs. 1f – 1h). The absolute value of IC of the under-oxidized sample (tox = 25 sec) is 6.5 mA for VG of +24 V, which is noticeably smaller than 8.8 mA for VG of -24 V (Fig. 1f). The over-oxidized sample (tox = 125 sec) shows the opposite behaviour; IC for a negative VG is smaller than that for a positive VG (Fig. 1h). On the other hand, the sample with tox = 75 sec shows a negligible modulation in IC by VG (Fig. 1g). Figure 1e shows that the modulation of the IC is closely correlated with that of the HK. We note, however, that the change in IC of ~30% is much greater than that in HK of ~2%, which implies the VCMA may not be the only source causing the electric-field controlled SOT. These results presented in Fig. 1 demonstrate two working principles required for the realization of the spin logic devices; the first is that IC for SOT-induced switching can be effectively tuned by VG and the second is that the dependence of IC on VG can be designed by the oxidation time. In particular, the second principle allows us to mimic p-type or n-type semiconductor device by defining a SOT device as p-type (n-type) when the device switches its magnetization by an in-plane input clock current IIN for a negative (positive) VG. Programmable logic operation of spin logic device With these working principles, we first demonstrate a logic operation using a single device in which the gate voltage (VG) and the input current (IIN) are used as two input parameters (Fig. 2a). The tox is 125 sec of the device (> tox.op; over-oxidation condition), which is thus p-type where IC for a negative VG is smaller than that for a positive VG (Fig. 2b). By setting the in-plane input clock current IIN of ±13 mA and VG of -24 V, the magnetization switches from the UP magnetization (//+z) to the DOWN magnetization (//-z) or vice versa, depending on the polarity of IIN. Figure 2c shows the operation of the given spin logic device, which allows us to construct a truth table (Fig. 2d). Here, we define the positive values of the inputs (VG = +24 V, IIN = +13 mA) as digital input "1" whereas their negative values (VG = -24 V, IIN = -13 mA) as digital input "0". Those inputs determine the output of the spin logic device, the magnetization direction which is monitored by measuring anomalous Hall resistance Rxy; UP magnetization (Rxy > 0) is equivalent to digital output "1" and DOWN magnetization (Rxy < 0) is equivalent to digital output "0". Figure 2d demonstrates that two logic operations are possible depending on the initial magnetization state; the device operates the OR function of IIN and VG when it is initialized as "1" state (magnetization UP), and the AND function of IIN and VG' (NOT VG) when initialized as "0" state (magnetization DOWN). This offers the realization of multi-functional or programmable spin logic devices. The logic operations can be also performed by the combination of two spin logic devices with the same type. The sample is fabricated by defining two nominally identical CoFeB/MgO/AlOx islands on a common Ta underlayer, which are designated as spin logic devices A and B, respectively (Fig. 3a). We note that both devices A and B are of p-type (tox = 125 sec) and IC is 12.5±0.5 mA (14.5±0.5 mA) for a VG -24 V (+24 V) [Fig. 3b]. Figure 3c shows the performance of the device for initial UP magnetization of both devices A and B (Rxy,A = Rxy,B = +2), where the IIN and VG is set to ±13 mA and -24 V, respectively. The magnetization or the Rxy,A (B) of the device A (B) is controlled by the IIN only when VG,A (B) of - 24 V is applied, confirming the p-type character. Using the results for the initial UP magnetization and IIN of -13 mA extracted from Figs. 3b and 3c, we construct a truth table (Fig. 3d), where two input parameters of VG,A and VG,B [the positive (negative) VG corresponds to the digital input "1" ("0")] determine the output of the device, magnetization configuration which is monitored by the sum of the Hall resistances (Routput = Rxy,A+Rxy,B). In order to describe the logic operation, we define the reference, Rref, UP (DOWN)= 𝑅𝑥𝑦,A +𝑅𝑥𝑦,B 2 , for UP (DOWN) magnetizations in both devices: Rref, UP=+2 and Rref, DOWN=-2. The truth table shows that the Routput is larger than the Rref, UP corresponding to the digital output "1", only when both VG,A and VG,B are positive, demonstrating the AND gate operation. On the other hand, the OR gate can be performed by taking the Rref, DOWN; the Routput is smaller than the Rref, DOWN corresponding to the digital output "0", only when both VG,A and VG,B are negative. Note that the different logic operation can be attained by changing the initial magnetization and the polarity of IIN as shown in Supplementary Note 3. Complementary functionality of spin logic devices The next demonstration is for the complementary operation of the spin logic devices, in which n-type device (tox = 25 sec; under-oxidation condition) and p-type device (tox = 125 sec; over-oxidation condition) are electrically connected and incorporated with a common gate electrode VG (Fig 4a). The n-type characteristic of the former device is shown in Supplementary Note 4. Before the operation, we initialize the devices to be UP magnetizations (Rxy,n = Rxy,p = +2) corresponding to the digital output "1" state and use the IIN of ±12 mA. Figure 4b shows the complementary operation of the spin logics; when we apply VG = +24 V (red-shaded region), the magnetization or digital state of the n-type device is manipulated by the IIN while that of the p-type device remains unchanged. In contrast, when we apply VG = -24 V (blue-shaded region), the digital state in the p-type device is selectively controlled by the IIN. Therefore, our devices perform the n-type (p-type) function exclusively for the positive (negative) VG, confirming the complementary functionality of the spin logic device. This is a spin analogue to the conventional CMOS, which can facilitate the spin-based logic devices. Furthermore, when we apply an IIN greater than 14 mA, both devices are simultaneously controlled regardless of the gate voltage (grey-shaded region). This is a unique function of the spin logic devices based on SOT, which allows for initializing or erasing all the information at the same time [38]. Discussion We finally discuss several advantages of our spin logic device over conventional CMOS- based device: (i) it does not require refreshing in quiescent mode because of its non-volatile characteristic of magnetic state as well as VCMA effect (Fig. S2), resulting in a substantial reduction in power consumption. (ii) The n-type or p-type character of the spin logic device is determined by the oxidation state at the FM/oxide interface, which in principle can be controlled by a gate voltage for an improved VCMA effect. This reconfigurable spin logic device offers great flexibility for various logic algorithms. (iii) The device does not include semiconductors, meaning that three-dimensional integration of the devices would be possible, which can result in considerably increased device density. (iv) It can also serve as a built-in memory cell within a logic circuit, reducing power dissipation required for transferring signals between commonly separate logic and memory cells. (v) The switching of the SOT-based device is fast [39], which are prerequisites for logic operations. We note that the signal of our device can be considerably improved by introducing a magnetic tunnel junction to read the output of the device [40, 41]. This together with above advantages promises that the further development of our spin logic device could provide an efficient logic unit with high density and fast speed. Methods Sample preparation The samples of Ta(5 nm)/Co32Fe48B20(CoFeB, 1 nm)/MgO(1.6 nm)/AlOx(1.8 nm) structure were prepared by magnetron sputtering on thermally oxidized Si substrates with a base pressure of less than 4.0×10-6 Pa (3.0×10-8 Torr) at room temperature. All metallic layers were grown by DC sputtering with a working pressure of 0.4 Pa (3 mTorr), while MgO layer is deposited by RF sputtering (150 W) from an MgO target at 1.33 Pa (10 mTorr). AlOx was formed by depositing 1.5 nm of a metallic Al layer followed by exposing to O2 plasma with a pressure of 4 Pa (30 mTorr) and a power of 30 W for various oxidation times tox. In order to promote the perpendicular magnetic anisotropy (PMA), samples were annealed at 250 °C for 40 min in vacuum condition. The Hall-bar structured devices including a square- shaped ferromagnetic island were fabricated using photo-lithography and Ar ion-beam etching. The width of the Hall bar is 5 μm or 10 μm and the size of the ferromagnetic island varies from 4×4 μm2 to 10×10 μm2. The gate oxide of ZrO2 (40 nm) was deposited by atomic layer deposition (ALD) at 150 °C using JUSUNG SDP ALD system. Measurements The anomalous Hall resistance (Rxy) was measured using a DC current of 100 μA while sweeping in-plane magnetic field (Bx), parallel to the current (Ix) direction, under various gate voltages (VG). The spin-orbit torque (SOT) switching experiments were carried out by applying a pulsed current (Ix) of 20 μs and a fixed in-plane magnetic field Bx of 10 mT under VG's. The logic operations were performed by controlling magnetization direction using a pulsed input clock current (IIN) of 20 μs and a gate voltage of ±24 V under an in-plane field (Bx) of 10 mT. All measurements were carried out at room temperature. Data availability Authors can confirm that all relevant data are included in the paper and/or its supplementary information files and data are available on request. References 1. Wolf, S. A. et al. Spintronics: A spin based electronics vision for the future. Science 294, 1488-1495 (2001). 2. Žutić, I., Fabian, J. & Das Sarma, S. Spintronics: fundamentals and applications. Rev. Mod. Phys. 76, 323-410 (2004). 3. Sinova, J. & Žutić, I. New moves of the spintronics tango. Nature Mater. 11, 368– 371 (2012) 4. Dery, H., Dalal, P., Cywinski, L. & Sham, L. J. Spin-based logic in semiconductors for reconfigurable large-scale circuits. Nature 447, 573–576 (2007). 5. Datta, S. & Das, B. Electronic analogue of the electro-optic modulator. Appl. Phys. Lett. 56, 665–667 (1990). 6. Koo, H. C. et al. Control of spin precession in a spin-injected field effect transistor. Science 325, 1515–1518 (2009). 7. Choi, W. Y. et al. Electrical detection of coherent spin precession using the ballistic intrinsic spin Hall effect. Nature Nanotechnol. 10, 666-670 (2015). 8. Wunderlich, J. et al. Spin Hall effect transistor. Science 330, 1801–1804 (2010). 9. Allwood, D. A. et al. Magnetic domain-wall logic. Science 309, 1688–1692 (2005). 10. Franken, J. H., Swagten, H. J. M. & Koopmans, B. Shift registers based on magnetic domain wall ratchets with perpendicular anisotropy. Nature Nanotechnol. 7, 499–503 (2012). 11. Murapaka, C., Sethi, P., Goolaup, S. & Lew, W. S. Reconfigurable logic via gate controlled domain wall trajectory in magnetic network structure. Sci. Rep. 6, 20130 (2016). 12. Currivan-Incorvia, J. A. et al. Logic circuit prototypes for three-terminal magnetic tunnel junctions with mobile domain walls. Nature Commun. 7, 10275 (2016) 13. Bhowmik, D., You, L. & Salahuddin, S. Spin Hall effect clocking of nanomagnetic logic without a magnetic field. Nature Nanotechnol. 9, 59–63 (2014). 14. Nikonov, D. E., Bourianoff, G. I. & Ghani, T. Proposal of a spin torque majority gate logic. IEEE Electron Device Lett. 32, 1128–1130 (2011). 15. Jamali, M., Kwon, J. H., Seo, S.-M., Lee, K.-J. & Yang, H. Spin wave nonreciprocity for logic device applications. Sci. Rep. 3, 3160 (2013). 16. Miron, I. M. et al. Perpendicular switching of a single ferromagnetic layer induced by in-plane current injection. Nature 476, 189-193 (2011). 17. Liu, L. et al. Spin-torque switching with the giant spin Hall effect of tantalum. Science 336, 555–558 (2012). 18. Kim, J. et al. Layer thickness dependence of the current-induced effective field vector in TaCoFeBMgO. Nat. Mater. 12, 240-245 (2013). 19. Garello, K. et al. Symmetry and magnitude of spin-orbit torques in ferromagnetic heterostructures. Nat. Nanotech. 8, 587-593 (2013). 20. Fan, X. et al. Quantifying interface and bulk contributions to spin-orbit torque in magnetic bilayers. Nat. Commun. 5, 3042 (2014). 21. Qiu, X. et al. Spin-orbit-torque engineering via oxygen manipulation. Nat. Nanotech. 10, 333-338 (2015). 22. Kaiming, C. et al. Electric field control of deterministic current-induced magnetization switching in a hybrid ferromagnetic/ferroelectric structure. Nature Mater. 16, 712-716 (2017) 23. Emori, S., Bauer, U., Woo, S. & Beach, G. S. D. Large voltage-induced modification of spin-orbit torques in Pt/Co/GdOx. Appl. Phys. Lett. 105, 222401 (2014) 24. Ohno, H. et al. Electric-field control of ferromagnetism. Nature 408, 944–946 (2000) 25. Matsukura, F., Tokura, Y., & Ohno, H. Control of magnetism by electric fields. Nature Nanotechnol. 10, 209-220 (2015) 26. Weisheti, M. et al. Electric field induced modification of magnetism in thin-film ferromagnets. Science 315, 349–351 (2007). 27. Maruyama, T. et al., Large voltage-induced magnetic anisotropy change in a few atomic layers of iron. Nature Nanotech. 4, 158-161 (2009) 28. Wang, W.-G., Li, M., Hageman, S. & Chien, C. L. Electric-field-assisted switching in magnetic tunnel junctions. Nature Mater. 11, 64–68 (2012). 29. Shiota, Y. et al. Induction of coherent magnetization switching in a few atomic layers of FeCo using voltage pulses. Nature Mater. 11, 39-43 (2012) 30. Kanai, S. et al. Electric field-induced magnetization reversal in a perpendicular- anisotropy CoFeB-MgO magnetic tunnel junction. Appl. Phys. Lett. 101, 122403 (2012) 31. Leistner, K. et al. Electric field control of magnetism by reversible surface reduction and oxidation reactions. Phys. Rev. B 87, 224411 (2013) 32. Bauer, U. et al. Magneto-ionic control of interfacial magnetism. Nature Mater. 14, 174-181 (2015) 33. Manchon, A. et al. Analysis of anisotropy crossover due to oxygen in Pt/Co/MOx trilayer. J. Appl. Phys. 104, 043914 (2008). 34. Bi, C. et al. Reversible control of Co magnetism by voltage-induced oxidation. Phys. Rev. Lett. 113, 267202 (2014). 35. Fukami, S., Zhang, C., DuttaGupta, S., Kurenkov, A. & Ohno, H. Magnetization switching by spin-orbit torque in an antiferromagnet/ferromagnet bilayer system. Nature Mater. 15, 535-541 (2016). 36. Oh, Y.-W. et al. Field-free switching of perpendicular magnetization through spin-orbit torque in antiferromagnet/ferromagnet/oxide structures. Nature Nanotechnol. 11, 878- 884 (2016). 37. Lee, K.-S., Lee, S. –W., Min, B. –C. & Lee, K. –J. Threshold current for switching of a perpendicular magnetic layer induced by spin Hall effect. Appl. Phys. Lett. 102, 112410 (2013). 38. Baek, S. C., Oh, Y. –W., Park, B. –G. & Shin, M. Novel Operation of a Multi-Bit SOT Memory Cell Addressed with a Single Write Line. IEEE Trans. Magn. (2017). 39. Fukami, S., Anekawa, T., Ohkawara, A., Zhang, C. & Ohno, H. A sub-ns three- terminal spin-orbit torque induced switching device. 2016 IEEE Symposium on VLSI Technology. 1-2 (2016). 40. Currivan-Incorvia, J. A. et al. Logic circuit prototypes for three-terminal magnetic tunnel junctions with mobile domain walls. Nature Commun. 7, 10275 (2016). 41. Fukami, S., Anekawa, T., Zhang, C. & Ohno, H. A spin-orbit torque switching scheme with collinear magnetic easy axis and current configuration, Nature Nanotechnol. 11, 621-625 (2016). Figure 1 Manipulation of magnetic anisotropy (HK) and critical switching current (IC) by gate voltage (VG). a, Schematic of the sample with Ta(5 nm)/CoFeB(1 nm)/MgO(1.6 nm)/AlOx (1.8 nm) structures, where the magnetization direction can be controlled by the combination of in-plane current (Ix) and gate voltage (VG). b,c,d, Voltage-controlled magnetic anisotropy (VCMA) for samples with oxidation time (tox) of 25 sec (b), 75 sec (c), and 125 sec (d), respectively. The normalized anomalous Hall resistance Rxy /Rxy, B=0 is plotted as a function of in-plane magnetic field (Bx) under a VG of +22 V or -20 V. Here VG of 20 V corresponds to an electric field of 5.0 MV/cm. e, The change in HK and IC for samples with various tox's by the VG, where ΔHK and ΔIC are HK (VG= +22V) - HK (VG= -20V) and IC (VG= +24V) - IC (VG= -24V), respectively. f,g,h, Spin-orbit torque (SOT)-induced switching experiments for samples with tox = 25 sec (f), tox = 75 sec (g), and tox = 125 sec (h). The Rxy vs in-plane current Ix under a VG of +24 V or -24 V. During the switching experiments, a magnetic field of 10 mT is applied along the direction of the Ix. Figure 2 Logic operation of a single spin logic device. a, Schematic of the device, in which the gate voltage (VG) and the input current (IIN) are used as two input parameters. b, SOT- induced switching characteristic of the spin logic device under the VG's of ±24 V. c, Input parameters and output of the device. Upper two panels: sequence of the VG (= ±24 V) and the IIN (= ±13 mA). Lower panel: The anomalous Hall resistance Rxy representing the magnetization direction that is determined by the VG and IIN. d, Logic true table constructed using the results in Figure 2c. Figure 3 Logic operation of p-type spin logic devices. a, Schematic of the device, consisting of two p-type spin logic devices which are denoted as spin logic device A and B, respectively. b, SOT-induced switching characteristic of device A (left panel) and B (right panel) under the VG's of ±24 V. c, Input parameters and output of the device. Upper three panels: sequence of the VG (= ±24 V) and the IIN (= ±13 mA). The VG,A (VG,B) is the VG applied on device A (B). Lower two panels: The anomalous Hall resistance Rxy representing the magnetization direction that is determined by the VG,A (VG,B) and IIN. The Rxy,A (Rxy,B) is the Rxy of device A (B). d, Truth table of the logic operations in the spin logic devices: input parameters of VG,A and VG,B, output of the device Routput (= Rxy,A+Rxy,B), and the reference Rref, UP (DOWN)= 𝑅𝑥𝑦,A +𝑅𝑥𝑦,𝐵 2 , for UP (DOWN) magnetizations of both devices. Both devices are initialized to have UP magnetization directions and IIN = -13 mA Figure 4 Demonstration of complementary functionality in spin logic devices. a, Schematic of the device consisting of n-type and p-type spin logic devices, which are electrically connected and incorporated with a common gate electrode VG. b, Input parameters and output of the device. Upper two panels: sequence of the input parameters: common gate voltage, VG = +24 V (red-shaded region), -24 V (blue-shaded region), and IIN = ±12 mA. The grey-shaded region is for IIN = ±14 mA. Lower two panels: The anomalous Hall resistance Rxy representing the magnetization direction that is determined by the VG and IIN. The Rxy of n-type and p-type devices are denoted as Rxy, n and Rxy, p, respectively. Supplementary Note 1. Dependence of magnetic anisotropy on oxidation time Figures S1a-S1e show the anomalous Hall resistance (Rxy), representing the z-component of magnetization (𝑚𝑧), as a function of the in-plane magnetic field (Bx) for the Ta(5 nm)/CoFeB(1 nm)/MgO(1.6 nm)/AlOx(1.8 nm) samples with different oxidation times tox's. The x-component of the magnetization, mx (=√1 − 𝑚𝑧 2) versus Bx is shown in the inset of each graph, from which the HK is extracted (Fig. S1f). Note that all the measurements are done without a gate voltage (as-deposited state). We find that the HK is the largest for tox = 75 sec, referring to as the optimal oxidation state, tox,op, and HK decreases for tox either longer or shorter than tox,op. Figure S1 Magnetic anisotropy field (HK) versus oxidation time (tox). a,b,c,d,e Normalized anomalous Hall resistance (Rxy) as a function of in-plane magnetic field (Bx) for Ta(5 nm)/CoFeB(1 nm)/MgO(1.6 nm)/AlOx(1.8 nm) samples of tox= 25 sec (a), 50sec (b), 75 sec (c), 100 sec (d), and 125 sec (e), respectively. Insets show mx (=√1 − 𝑚𝑧 2) versus Bx curves. f, Summary of the HK versus tox. The red line is a guide to the eye. Supplementary Note 2. Non-volatility of voltage-controlled magnetic anisotropy As shown in Fig. 1 of the main text, the magnetic anisotropy field (HK) of the over-oxidized sample (tox = 125 sec) decreases by a gate voltage VG of -20 V, which is demonstrated in Fig. S2, where the black (red) symbols represent the normalized Rxy versus Bx curve under 0 V (- 20V). We repeated the same measurement after removing the VG, which is indicated as the blue symbols in Fig. S2, demonstrating the HK remains the same as the previously-measured value under -20 V. This clearly shows non-volatile behavior of the voltage-controlled magnetic anisotropy in our samples. Figure S2 Non-volatility of voltage-controlled magnetic anisotropy. Normalized anomalous Hall resistance versus in-plane magnetic field for the over-oxidized sample (tox = 125 sec) under a sequence of gate voltage, 0 V (black symbols), -20 V (red symbols), and 0 V (blue symbols). Supplementary Note 3. Logic operation of spin logic device Table S1 summarizes the logic operation of the device of Fig. 3 on the main text when both devices are initialized as DOWN magnetization and the input clock current is positive (IIN > 0). By changing the initial magnetization states, the same device exhibits different Boolean equations from those in Fig. 3d of the main text; demonstrating that the logic operation can be programmed by the initial magnetization direction. Table S1. Logic operation of two p-type spin logic devices for initial magnetization DOWN and the positive IIN. Logic input Logic output VG,A VG,B Rxy,A+ Rxy,B Rref,UP (+2 Ω) Rref,UP (-2 Ω) -24V (0) -24V (0) +4 Ω -24V (0) +24V (1) +24V (1) -24V (0) 0 0 +24V (1) +24V (1) -4 Ω 1 0 0 0 1 1 1 0 Supplementary Note 4. Electric-field controlled spin-orbit torque switching in n-type spin logic device Figure S3 shows the dependence of the spin-orbit torque (SOT) switching on the gate voltage VG for the n-type device (tox= 25) in the complementary spin logic devices presented in Fig. 4 of the main text. The IC in the n-type spin logic device increases (decreases) when a negative (positive) VG of 24 V is applied, which is the opposite to that in p-type device, which is summarized in Table S2. Figure S3 Electric-field controlled spin-orbit torque switching. a, Schematic of n-type (tox= 25) devices. b, SOT switching characteristics of the devices under a positive and negative VG of 24 V. Table S2 Summary of critical current of SOT-induced switching for the complementary spin logic device consisting of n-type and p-type devices shown in Fig. 4 of the main text. VG IC, n-type IC, p-type +24V -24V +24V -24V UP→DOWN -12mA -13mA -14mA -12mA DOWN→UP +11 mA +13mA +14 mA +12mA
1801.06469	1	1801	2018-01-19T15:50:47	Highly selective dry etching of GaP in the presence of Al$_\textrm{x}$Ga$_{1-\textrm{x}}$P	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We present an inductively coupled-plasma reactive-ion etching process that simultaneously provides both a high etch rate and unprecedented selectivity for gallium phosphide (GaP) in the presence of aluminum gallium phosphide (Al$_\textrm{x}$Ga$_{1-\textrm{x}}$P). Utilizing mixtures of silicon tetrachloride (SiCl$_4$) and sulfur hexafluoride (SF$_6$), selectivities exceeding 2700:1 are achieved at GaP etch rates above 3000 nm/min. A design of experiments has been employed to investigate the influence of the inductively coupled-plasma power, the chamber pressure, the DC bias and the ratio of SiCl$_4$ to SF$_6$. The process enables the use of thin Al$_\textrm{x}$Ga$_{1-\textrm{x}}$P stop layers even at aluminum contents of a few percent.	physics.app-ph	physics	Highly selective dry etching of GaP in the presence of AlxGa1−xP with a SiCl4/SF6 plasma Simon Honl, Herwig Hahn, Yannick Baumgartner, Lukas Czornomaz and Paul Seidler IBM Research -- Zurich, Saumerstrasse 4, CH-8803 Ruschlikon, Switzerland E-mail: [email protected] and [email protected] Abstract. We present an inductively coupled-plasma reactive-ion etching process that simultaneously provides both a high etch rate and unprecedented selectivity for gallium phosphide (GaP) in the presence of aluminum gallium phosphide (AlxGa1−xP). Utilizing mixtures of silicon tetrachloride (SiCl4) and sulfur hexafluoride (SF6), selectivities exceeding 2700:1 are achieved at GaP etch rates above 3000 nm/min. A design of experiments has been employed to investigate the influence of the inductively coupled-plasma power, the chamber pressure, the DC bias and the ratio of SiCl4 to SF6. The process enables the use of thin AlxGa1−xP stop layers even at aluminum contents of a few percent. Keywords: inductively-coupled-plasma reactive ion etching. gallium phosphide, aluminum gallium phosphide, selective etching, 8 1 0 2 n a J 9 1 ] h p - p p a . s c i s y h p [ 1 v 9 6 4 6 0 . 1 0 8 1 : v i X r a Highly selective dry etching of GaP in the presence of AlxGa1−xP 2 1. Introduction III-V materials are well established in the semiconductor industry for applications ranging from RF amplifiers in cellular communication devices to light emitting and laser diodes [1 -- 4] to multi-junction concentrator solar cells [5, 6]. The integration of such devices onto silicon substrates is an emerging trend, offering the opportunity to combine, for example, low-power opto-electronic devices [7 -- 9] and high-electron- mobility transistors [10] based on III-V compound semiconductors with complex silicon microelectronic circuits. Numerous techniques have been developed for the microfabrication of such III-V devices [11], among which processes for selective etching are essential. Inductively coupled-plasma reactive-ion etching (ICP-RIE) is a particularly attractive method to achieve selectivity because the separation of plasma generation and ion bombardment offers additional flexibility for optimizing the etching of one material preferentially over another [12]. ICP-RIE processes have been developed for compounds such as GaAs, AlGaAs, InGaAs, InP and GaN [13 -- 18]. In contrast, GaP is a largely unexplored material for integrated nanophotonic and opto-electronic applications, despite its attractive combination of large refractive index and large electronic bandgap, and fabrication processes for this material are in general far less advanced. In particular, selective etching of GaP versus AlxGa1−xP has hardly been investigated. Although wet etch processes for the selective removal of AlxGa1−xP in the presence of GaP exist [19], there are no reports of the opposite selectivity by wet etching, and only one such dry etching process has been previously investigated [20]. The etch rate and selectivity of the latter process are too low for the practical removal of large amounts (many tens of microns) of GaP or for stopping on a thin layer (a few hundred nanometers) of AlxGa1−xP, as may be needed for process flows involving wafer bonding, for example. To address this shortcoming, we present here an ICP-RIE process based on SiCl4 and SF6 that etches GaP over AlxGa1−xP with unprecedented selectivity while maintaining an extremely high GaP etch rate. The process enables the use of thin AlxGa1−xP etch stop layers with aluminum content as low as a few percent, providing much greater process control and enabling new types of GaP-based devices [21]. We employ the design-of-experiments method [22 -- 24] to investigate the influence of pressure, ICP power, DC bias and gas composition on the etch rate of GaP and the selectivity of etching GaP with respect to two AlxGa1−xP compositions. 2. Experimental Methods 2.1. Sample preparation AlxGa1−xP samples were epitaxially grown by metalorganic chemical vapor deposition (MOCVD) (Veeco P125) from trimethylgallium (TMGa), trimethylaluminum (TMAl) and tertiarybutylphosphine (TBP) on undoped, [100]-oriented quarters of a 2-inch GaP wafer with a nominal thickness of 400 µm. Chips diced from the same epi- ready substrates were used as samples for the etching experiments on GaP itself. The growth temperature in the MOCVD system was 650 ◦C at the susceptor as measured with an optical pyrometer. Prior to the growth of the AlxGa1−xP layers, a 100 nm- thick homoepitaxial layer of GaP was grown at a V-to-III molar-flow ratio of 10:1. Subsequently, AlxGa1−xP layers with a target thickness of 100 nm were grown at Highly selective dry etching of GaP in the presence of AlxGa1−xP 3 a V-to-III molar-flow ratio of 6:1 with various proportions of TMGa and TMAl. The resulting film stoichiometries were determined by X-ray diffraction to have an aluminum content of x = 0.033 and x = 0.097 and are referenced below as x = 0.03 and x = 0.10, respectively. A SiO2 hard mask was employed for the etching experiments on both the GaP and AlxGa1−xP samples. Starting with a quarter GaP wafer and a quarter wafer of each AlxGa1−xP composition, 100 nm of SiO2 were deposited by plasma-enhanced chemical vapor deposition (PECVD) in an Oxford Plasmalab 100 system with SiH4 and N2O as precursors. The SiO2 was then photolithographically patterned using the positive resist AZ6612 (Microchemicals GmbH) and a CF4/Ar RIE process in an Oxford NGP 80 system. The photoresist was subsequently removed using acetone, isopropanol, and an oxygen plasma (GIGAbatch 310 M from PVA TePla), which should also eliminate non-volatile etch residues originating from the CF4-containing RIE process. Finally, the GaP and AlxGa1−xP quarter wafers were diced into 3.5-mm × 3.5-mm chips. The complete process flow is schematically illustrated in figure 1. Figure 1. Process flow for the fabrication of AlxGa1−xP samples: (a) epitaxial growth of GaP and AlxGa1−xP by MOCVD; (b) deposition of SiO2 by PECVD; (c) spin-coating and exposure of photoresist; (d) pattern transfer into SiO2 by RIE; (e) removal of photoresist. 2.2. Etching For the development of the selective ICP-RIE process with SiCl4 and SF6, the samples were etched for 300 s in an Oxford Instruments PlasmaPro 100 ICP system. The samples were placed side by side at the center of a 4-inch Si carrier wafer with no Highly selective dry etching of GaP in the presence of AlxGa1−xP 4 additional thermal conduction promoter such as oil, wax, or grease. The temperature of the carrier wafer was held at 30 ◦C with a helium backing flow. The other process parameters were determined by the design of experiments (see below). Following etching, the surface profile of the samples was measured with a Bruker DekTak XT profilometer, where the measured step height included the residual SiO2 hard mask. The hard mask was then stripped with a 90-s dip in buffered hydrofluoric acid (BHF), after which the etch profile was remeasured to determine both the etch rate and how much of the hard mask had been removed during the ICP-RIE process. The etch rate of GaP and AlxGa1−xP (x = 0.03 and x = 0.10) in BHF is negligible (measured to be < 0.3 nm/min for x = 0.10 by atomic force microscopy in a separate control experiment using an organic resist for the mask). If it was found that the SiO2 layer had been completely consumed during dry etching, the experiment was repeated for an etch duration of 120 s. To account for differences in the etch rate across the chip, the profile was measured at three positions from the edge towards the center in steps of 600 µm and the results averaged. 2.3. Design of Experiments Given the large parameter space of an ICP-RIE process [11], it is obviously time- consuming and laborious to analyze the process by investigating changes of one parameter at a time. To obtain a general understanding of the etch process more efficiently, i.e. without performing experiments for every possible parameter combination, we used a design-of-experiments (DOE) approach [22], probing a limited selection of points distributed over the parameter space. Specifically, a fractional factorial design comprising 24 experiments was created with the commercial statistical software JMP R(cid:13) (SAS Institute) [25]. We investigated the influence of the ICP power (PICP), the chamber pressure (pC), the DC bias voltage (VDC), and the SiCl4- to-SF6 ratio expressed as the fraction of SiCl4 (fSiCl4 ) on the etch rate of GaP and the selectivity between GaP and AlxGa1−xP. The parameters of the individual experiments performed as specified by JMP for an I-optimality criterion [22] are displayed in table 1. An I-optimal design was possible because preliminary experiments gave us confidence in having already obtained a general understanding of the effect of the process conditions, and it provides for better prediction than e.g. the D-optimal design [22]. Values for the ICP power, pressure, and DC bias were chosen by the software from a continuous range. The chamber pressure was varied between 5 mTorr and 60 mTorr and the ICP power between 25 W and 400 W. The DC bias was set to values in the range of 50 V to 250 V by adjusting the RF power for each experiment individually. For fSiCl4, six explicit values were chosen according to preliminary experiments, keeping the combined total flow of SiCl4 and SF6 constant at 20 sccm. 3. Results and Analysis The etch rates obtained from the 24 experiments comprising the DOE are presented in table 1. The results were analyzed separately with respect to two responses: the GaP etch rate and the selectivity between GaP and AlxGa1−xP. Because the observed etch rates span several orders of magnitude and the error is expected to increase with both etch rate and selectivity, the least-squares regression analyses with JMP were performed on the common logarithm of the response (either etch rate or selectivity). Much better fits to the data were thus achieved; a systematic skew of the distribution Highly selective dry etching of GaP in the presence of AlxGa1−xP 5 Table 1. Experimental parameters and resulting etch rates. ρi denotes the etch rate for the various materials with i indicating the aluminum content in AlxGa1−xP (i = 0 represents GaP). Becuase the signal-to-noise ratio of step heights less than 10 nm is low, the etch rates below 2 nm/min generally have a large relative error. Values noted as 0 correspond to no resolvable steps observed with the profilometer. Etch rates that were excluded from the DOE fit due to residue deposition or micomasking are marked with an asterisk, in which case the given value for the etch rate generally represents a lower limit. No. fSiCl4 [% SiCl4] pC [mTorr] VDC [V] PICP [W] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 37.5 75 50 50 50 25 37.5 66.7 25 100 37.5 25 37.5 50 66.7 0 25 37.5 37.5 0 100 100 50 66.7 51.8 5 49.1 16.1 35.3 49.1 60.1 51.7 5 60.1 5.1 60.1 46.2 60.1 18.8 32.6 21.5 13.3 18.7 5 5 41.3 5 49.1 97 60 255 207 166 77 211 122 202 108 250 260 183 51 232 230 100 50 100 69 209 49 100 233 25 64 347 25 194 403 403 308 308 25 103 45 103 140 403 271 121 347 289 25 233 403 403 103 ρ0 [nm] 2.39 114 3394 1209 2637 2.4 8.26 2527 398∗ 24.9 346 11.7 6.3 503∗ 1129 1.84 2.73 4.45 5.6 0 27.5 6.06 216 3008 ρ0.03 [nm] 4.06 33.7 1.5 53.3∗ 6.6 4.35 11.1 22.4 122∗ 19.9 327 9.86 4.15 11.5 98.6∗ 2.08 2.53 3.79 5.46 0 27.3 3.09 48.8∗ 133∗ ρ0.10 [nm] 4.51 10.6 1.25 11.9 3.59 4.34 10.1 12.9 111∗ 17.5 229 10.4 4.45 5.57 44.8∗ 1.89 1.89 3.34 3.91 0 26.2 1.7 17.4 57.4∗ of residuals as a function of the response was observed with a linear scale but not with a logarithmic scale. To find an accurate model of minimal complexity, we started in each case by describing the desired response as an over-specified polynomial function of the process parameters, including cross terms. The coefficient of determination R2, the adjusted R2, and the probability value p for both individual terms and the model as a whole were calculated. Terms with low statistical significance as gauged by a high p-value were sequentially removed until the highest remaining individual p-value was below 0.05 and removal of any more terms would lead to a substantial drop in the adjusted R2. The global maximum of the response within the explored parameter range was also determined as well as the variation of the response in the vicinity of the global maximum as a function of each process parameter individually. 3.1. GaP etch rate We first present a model for the GaP etch rate alone. (Selectivity is treated with a separate model to account for the additional mechanisms in effect in the etching of AlxGa1−xP that do not pertain to GaP.) We restrict the model to processes for Highly selective dry etching of GaP in the presence of AlxGa1−xP 6 Table 2. Model terms and their corresponding probability value p for the etch rate of GaP. Term p f 2 SiCl4 VDC pC · fSiCl4 0.00000 0.00041 0.01668 which the fraction of SiCl4 is within the range of 37.5% to 100% because the abrupt change from the low etch rates observed at lower fractions of SiCl4 would require a high-degree polynomial with more terms than can be justified with the size of the data set. In other words, we model a region of the parameter space for which GaP is etched at an appreciable rate. In addition, experiment 14 was excluded, as residues formed on this sample during the etch process leading to micromasking effects, so the etch rate could not be unambiguously determined. After thorough investigation of various alternative polynomial functions of the process parameters, we consistently found that only the three terms listed in table 2 are necessary to provide a suitable model. The calculated dependence of the GaP etch rate on each individual process parameter with the others fixed at the value that gives the maximum response is displayed in figure 2. The ICP power dependence is not shown because this parameter is not statistically significant over the parameter range investigated, a result that suggests that the plasma is already saturated at the lowest value of 25 W with those species involved in the etching of GaP. While the chamber pressure alone is also not significant, it does play a role in conjunction with the fraction of SiCl4. Indeed, it is not surprising that the product of pressure and SiCl4 fraction (pC · fSiCl4 ) appears in the model, as this term is related not only to the relative but also the absolute concentration of the etchants. The effect of pC is primarily to shift the value of fSiCl4 for which the maximum etch rate is achieved, not to increase the etch rate, as is evident from figure 2(a), in which the modest monotonic increase in etch rate with pressure stays within the 95% confidence interval. The model predicts that for high SiCl4 content a higher pressure is preferable, whereas at high SF6 content a lower pressure yields a higher etch rate. This behavior is evident in the response surface plot in figure 3. We interpret this finding as indicating that species originating from SiCl4 rather than SF6 dominate chemical reaction with GaP. Specifically, etching by ion bombardment of the surface is more important when the amount of chlorine-containing species is low, as this physical process should benefit from lower pressure. If the content of chlorine-containing species in the plasma is high, chemical reactions with the surface play a greater role and should be enhanced at high ion density, i.e. at high pressure. Of considerably more statistical significance is the DC bias. The monotonic increase in etch rate over one and a half orders of magnitude as the DC bias increases from 50 V to 250 V (figure 2(b)) is attributed to the change in kinetic energy of the ions bombarding the sample and the consequent enhancement of the physical component of etching. We note that the calculated optimum value for VDC corresponds to one end of the parameter range investigated. It might therefore be possible to achieve an even higher etch rate if the parameter range is extended. The third and statistically most significant term is the quadratic dependence on fSiCl4, which reflects the existence of an optimum ratio of SiCl4 to SF6 for achieving a Highly selective dry etching of GaP in the presence of AlxGa1−xP 7 Figure 2. GaP etch rate as predicted by the model as a function of (a) pC, (b) VDC, and (c) fSiCl4 , individually. In each panel, the other free parameters are fixed at pC = 60 mTorr, VDC = 250 V, and fSiCl4 = 73%, respectively. The blue shaded regions define 95% confidence intervals. The dark blue diamonds in (c) indicate the measured etch rate for additional etch experiments at the indicated SiCl4 content. high GaP etch rate (figure 2(c)). Individual experiments conducted within the scope of the DOE with a low fraction of SiCl4 had a very low etch rate, and processes containing only fluorine species showed virtually no etching, indicating that either SF6 is not a good etchant or it passivates the surface. On the other hand, the etch rate also declines at high SiCl4 content. Indeed, the etch rate for a plasma using only fSiCl4 [%]406080100507090ρ0 [nm/min]101102103104VDC [V]50100150200250ρ0 [nm/min]101102103104pC [mTorr]102030405060ρ0 [nm/min]101102103104(c)(b)(a)Highly selective dry etching of GaP in the presence of AlxGa1−xP 8 Figure 3. Predicted GaP etch rate plotted versus chamber pressure (pC) and fraction of SiCl4 (fSiCl4 ). SiCl4 is one and half orders of magnitude lower than the optimum etch rate. Clearly, species originating from SF6 play a role in accelerating the etching despite the fact that a plasma using SF6 alone hardly etches GaP at all. Over the range of parameters investigated, the model predicts a maximum GaP etch rate of 9900 nm/min at pC = 60 mTorr, VDC = 250 V, and fSiCl4 = 73%, with the lower bound of the 95% confidence interval at 3100 nm/min and the upper bound at 32 000 nm/min. To test the models capability to predict the etch rate with respect to fSiCl4, additional etch experiments were performed on GaP samples at flows of 8 sccm (12 sccm), 13 sccm (7 sccm), 15 sccm (5 sccm) and 20 sccm (0 sccm) for SiCl4 (SF6) with pC = 60 mTorr, VDC = 250 V, and PICP = 300 W. A high ICP power was chosen to ensure the saturation of dissociated and ionized species in the plasma. Although the observed etch rates are somewhat lower than predicted (figure 2(c)) (presumably because the additional runs were carried out at a later date than the DOE experiments and the environment in the ICP-RIE chamber may be slightly different), the results are within or near the 95% confidence interval of the fit and follow the shape of the curve nicely. The quality of the model can be summarized in an actual-by-predicted plot, in which the experimentally observed etch rates are compared with those calculated from the model (figure 4). The coefficient of determination R2 = 0.90 and the probability value p < 0.0001 indicate a good fit and a high statistical significance of the model. 3.2. Selectivity Applying the same DOE analysis method to the selectivity for etching GaP versus AlxGa1−xP for the stoichiometries x = 0.03 and x = 0.10, we again consider only those experiments for which the fraction of SiCl4 is within the range of 37.5% to 100% and exclude those samples that exhibited micromasking, namely samples 4, 14, 15, 23, and 24. In this case, we find a considerably more complicated dependence on the various process parameters (table 3). Despite the relatively large number of terms, all but two p-values are under 0.001, and the p-value for the model as a whole is below ρ0 [nm/min]101102103104pC [mTorr]102030405060fSiCl4 [%]406080100507090Highly selective dry etching of GaP in the presence of AlxGa1−xP 9 Figure 4. (a) Actual-by-predicted plot of the GaP etch rate ρ0 with the 95% confidence interval shaded in red, and (b) plot of residuals of log(ρ0) as a function of predicted ρ0. In the absence of errors, the black dots indicating the individual experiments would lie on the straight lines. Table 3. Model terms and their corresponding probability value p for the selectivity of etching GaP over AlxGa1−xP for x=0.03 and x=0.10. Term p (x = 0.03) p (x = 0.10) f 2 SiCl4 f 3 SiCl4 fSiCl4 VDC p2 pC · fSiCl4 C pC VDC · fSiCl4 V 2 DC PICP 0.00001 0.00035 0.00062 0.00013 0.00024 0.00053 0.00043 0.00082 0.00224 0.02276 0.00000 0.00001 0.00001 0.00001 0.00002 0.00003 0.00007 0.00019 0.00026 0.00039 0.0001 for both stoichiometries, indicating quite high statistical significance. The model predicts a maximum selectivity over the parameter range investigated of roughly 15000:1 for the x = 0.03 stoichiometry at PICP = 400 W, pC = 37.3 mTorr, VDC = 250 V, and fSiCl4 = 63%, and 17000:1 for the x = 0.10 stoichiometry at PICP = 400 W, pC = 35.6 mTorr, VDC = 250 V, and fSiCl4 = 61%, as is evident from figure 5. Such selectivity values are beyond the range measurable in our experiments log(ρ0) residual [log(nm/min)]Actual ρ0 [nm/min]10110210310410010-1Predicted ρ0 [nm/min]100101102103104Predicted ρ0 [nm/min]100101102103104(a)(b)Highly selective dry etching of GaP in the presence of AlxGa1−xP 10 using a 100 nm-thick SiO2 hard mask. Figure 5 also shows the calculated dependence of the selectivity on each individual process parameter with the others fixed at the value that gives the maximum response. The selectivity behavior is qualitatively the same for the two AlxGa1−xP stoichiometries. In contrast to the model for the GaP etch rate, the ICP power is statistically significant in determining the selectivity (figure 5(a)). Still, it plays the least impor- tant role. The modest gain in selectivity with ICP power, which is linear on this logarithmic scale, remains within or close to the 95% confidence interval. Again, the ICP power is expected to influence primarily the chemical component of the etching process, and the weak dependence suggests that the plasma composition is already close to being saturated at the lowest value of 25 W also for those species involved in the etching of AlxGa1−xP and not just GaP. The influence of chamber pres- sure on selectivity (figure 5(b)) is also different than the situation for the GaP etch rate; several terms containing pC are statistically significant. The selectivity varies over an order of magnitude for the higher aluminum content and even more for the lower aluminum content. In both cases, the selectivity reaches a maximum in the middle of the pressure range evaluated. Since pressure is statistically insignificant in determining the GaP etch rate, this implies that there is some trade-off governing the AlxGa1−xP etch rate and suggests that, in addition to physical and chemical etching of the surface, other effects such as passivation may play a role. The interdependence of pressure and SiCl4 fraction is illustrated in the response surface plots in figure 6. Highly selective dry etching of GaP in the presence of AlxGa1−xP 11 Figure 5. Selectivity as predicted by the model as function of (a) PICP, (b) pC, (c) VDC and (d) fSiCl4 (x = 0.03 in blue, x = 0.10 in red). In each plot, the remaining parameters are fixed at PICP = 400 W, VDC = 250 V, pC = 36.3 mTorr and fSiCl4 = 63%, respectively, which represents the global maximum of selectivity for x = 0.03. The shaded regions in the respective color indicate the 95% confidence intervals of the model. 102103Selectivity104Al0.10Ga0.90PAl0.03Ga0.97PpC [mTorr]102030405060102103Selectivity104Al0.10Ga0.90PAl0.03Ga0.97PVDC [V]50100150200250102103Selectivity104Al0.10Ga0.90PAl0.03Ga0.97PfSiCl4 [%]406080100507090102103Selectivity104PICP [W]100200300400Al0.10Ga0.90PAl0.03Ga0.97P(b)(c)(d)(a)Highly selective dry etching of GaP in the presence of AlxGa1−xP 12 Figure 6. Predicted selectivity for etching GaP over AlxGa1−xP plotted versus chamber pressure (pC) and fraction of SiCl4 (fSiCl4 ) for (a) x = 0.03 and (b) x = 0.10. For DC bias (figure 5(c)), selectivity appears to be determined by an interplay between the GaP etch rate and the AlxGa1−xP etch rate, both of which are affected by this parameter. Hence, a quadratic dependence is observed in addition to an overall monotonic increase in selectivity with DC bias. Keeping with the hypothesis of a competition between chemical and physical processes, we attribute the nearly flat selectivity at low DC bias to a relatively weak contribution of the physical component which varies similarly for all stoichiometries. However, with increasing DC bias, the kinetic energy of the impinging ions accelerates the etching of GaP much more than the etching of AlxGa1−xP. This behavior is consistent with passivation primarily of AlxGa1−xP. As was the case for the GaP etch rate, the fraction of SiCl4 has the most complex influence on the selectivity (figure 5(d)). The dependence is however not just quadratic fSiCl4 [%]406080100507090pC [mTorr]102030405060Selectivity x =0.1010110210310010-110-2(b)fSiCl4 [%]406080100507090pC [mTorr]102030405060Selectivity x =0.0310110210310010-110-2(a)Highly selective dry etching of GaP in the presence of AlxGa1−xP 13 now but includes linear and cubic terms. The result is a selectivity that peaks at a lower value of fSiCl4 than the GaP etch rate and that plateaus at high fSiCl4 values where the SF6 flow is low. This would again be consistent with preferential passivation of AlxGa1−xP, but where the passivation is provided specifically by the SF6. At high SF6 content, i.e. low fSiCl4 values, both GaP and AlxGa1−xP are etched slowly and selectivity is low because SF6 is a poor etchant even for GaP. With increasing fraction of SiCl4, GaP starts to etch while AlxGa1−xP is passivated. At even higher values of fSiCl4 , either the passivation mechanism becomes less effective for AlxGa1−xP or the plasma composition becomes less reactive or both, and selectivity begins declining even before the GaP etch rate peaks. The plateau at the highest values of fSiCl4 is then ascribed to a limiting situation where the differentiation between GaP and AlxGa1−xP is no longer changing due to passivation or plasma reactivity. Overall, the stoichiometry with 10% aluminum content yields a higher selectivity than the 3% composition, except in a portion of parameter space where the fraction of SiCl4 is high. As can be seen in figure 5(d), the selectivity is higher for x = 0.10 at its maximum but drops faster with increasing fSiCl4 than the selectivity for x = 0.03. This suggest that, in addition to the passivation of AlxGa1−xP by SF6, there may be a competing enhancement at higher aluminum content of etching by chlorine species. The actual-by-predicted plots for selectivity (figure 7) indicate that the model describes the observations well. For both x = 0.03 and x = 0.10, the coefficient of determination is R2 = 1.00 and the probability value is p < 0.0001. We attribute the higher value of R2 for selectivity in comparison to that for the GaP etch rate to the fact that the GaP and AlxGa1−xP samples were etched side by side for each parameter set, which may somewhat reduce any scatter due to run-to-run variations. Highly selective dry etching of GaP in the presence of AlxGa1−xP 14 Figure 7. Actual-by-predicted plots of selectivity for etching GaP over AlxGa1−xP for (a) x = 0.03 and (c) x = 0.10, with the 95% confidence intervals shaded in red, and plots of residuals of the common logarithm of selectivity as a function of predicted selectivity for (b) x = 0.03 and (d) x = 0.10. In the absence of errors, the black dots indicating the individual experiments would lie on the straight lines. The grey dots represent data that were excluded from the analysis due to micromasking on either the GaP or the AlxGa1−xP samples. 101102103100100101102103Predicted Selectivity (x = 0.03)Actual Selectivity (x = 0.03)100101102103-0.040.000.04(a)(b)(c)(d)log(Selectivity (x = 0.03)) residuallog(Selectivity (x = 0.10)) residual101102103100100101102103100101102103-0.040.000.04Predicted Selectivity (x = 0.03)Predicted Selectivity (x = 0.10)Predicted Selectivity (x = 0.10)Actual Selectivity (x = 0.10)Highly selective dry etching of GaP in the presence of AlxGa1−xP 15 With the set of parameters that gives the maximum selectivity, the model for the GaP etch rate predicts an etch rate of 7900 nm/min with the lower bound of the 95% confidence interval at 2500 nm/min and the upper bound at 25 000 nm/min. On the other hand, the process with the highest predicted GaP etch rate yields a selectivity of 2400:1 with the lower and upper bound of the 95% confidence interval at 1300:1 and 4300:1, respectively, for x = 0.03. For x = 0.10, the selectivity is predicted to be 1900:1 with the lower and upper bound of the 95% confidence interval at 1500:1 and 2300:1, respectively. The process can therefore be adjusted depending on whether high selectivity or high etch rate is required. The two models can be combined in a plot that shows both GaP etch rate and selectivity as a function of fSiCl4 for the specific values of PICP = 300 W, VDC = 250 V and pC = 60 mTorr (figure 8), from which it is again apparent that the maximum etch rate occurs at a lower fraction of SiCl4 than the maximum selectivity. Figure 8. GaP etch rate (black) and selectivity for x = 0.10 (red dashed) as a function of fSiCl4 at PICP = 300 W, VDC = 250 V and pC = 60 mTorr. The blue and red shaded regions define the 95% confidence intervals. 101fSiCl4 [%]406080100507090ρ0 [nm/min]101102103104102103104105106107Selectivity (x = 0.10)Highly selective dry etching of GaP in the presence of AlxGa1−xP 16 Finally, we note that even higher selectivities may be attainable with aluminum contents greater than those tested here. From a practical point of view, epitaxial growth of AlxGa1−xP on a GaP substrate with an aluminum content well above 50% is entirely feasible, as the lattice mismatch between GaP and AlP is only 0.01% [26]. 3.3. Surface Morphology The surface morphology of the etched samples varies greatly over the range of process parameters used. Most of the deeply etched GaP samples have a smooth horizontal top surface without any evidence of residues being deposited. Only samples etched in processes 5, 9 and 14 show micromasking to various extents. In both experiment 9 and experiment 14, the fraction of SiCl4 is at the threshold between the passivating domain and the etching domain, so the micromasking may be related to the onset of passivation. The sample for process 5 exhibits micromasking only locally. In contrast, all of the AlxGa1−xP samples that were not significantly etched exhibited some residue formation. The etch profile can depend strongly on crystal lattice orientation. Figure 9 shows SEM micrographs of a sample that was etched under conditions for which the GaP etch rate and selectivity are both high (PICP = 303 W, pC = 60 mTorr, VDC = 130 V, and fSiCl4 = 60%). Here, we examined a square, 100-µm × 100- µm, test feature aligned with the (110) and the (110) flats of the GaP wafer. This allows us to determine the orientation of the crystal lattice planes in the etch profile. Figure 9(a) displays a micrograph of one corner of the square, whereas figures 9(b) - (d) show the same structure after milling the corner along the dashed white line in figure 9(a) with a gallium focused ion beam (FIB). The colorized overlays in figures 9(a) and (b) indicate the orientation of the respective lattice planes. The process produces a smooth horizontal top surface as well as smooth, angled and somewhat curved sidewalls which converge to (111) and (111) planes in figures 9(c) and (d), respectively. This behavior is typical of chemical etching of III-V materials, for which etching of the densely packed, group-III-terminated {111} planes are often slowest to etch. [27, 28] 4. Discussion Literature on etching GaP selectively over AlxGa1−xP is scarce. The only other previously published method is a dry etching process that makes use of SiF4 and SiCl4 and etches GaP at 135 nm/min with a selectivity of 124:1 with respect to Al0.6Ga0.4P [20]. We have observed similar behavior in an ICP-RIE process with CF4 and Cl2 [29]. Etching with a mixture of SiCl4 and SF6 has never before been investigated. However, selective dry-etching of GaAs in the presence of AlxGa1−xAs has been repeatedly reported using etching mixtures such as SF6/SiCl4 [30], CCl2F2 [31 -- 34] and SF6/BCl3 [13]. Selectivity in both the GaAs/AlxGa1−xAs system and the GaP/AlxGa1−xP system has been attributed to the formation of AlF3 on the surface of the aluminum-containing material which inhibits further etching [20, 30]. The passivating effect apparently arises from the much lower volatility of AlF3 compared to that of both AlCl3 and GaCl3 (table 4). Formation of GaF3, which also has low volatility, is not observed on GaAs or AlxGa1−xAs and is presumed to be impeded by a high kinetic barrier to reaction [30, 35], explaining why fluorine-containing species provide selectivity but not chlorine species. In the case of the phosphides, Epple et Highly selective dry etching of GaP in the presence of AlxGa1−xP 17 Figure 9. SEM micrographs of an etched square test feature with the GaP crystal planes indicated by color overlays: (a) the as-etched structure with a dashed white line delineating the region subsequently removed by FIB milling; (b) the same structure after FIB milling; (c) and (d) profile views perpendicular to the (110) plane and (110) plane, respectively. al verified the presence of an aluminum/fluorine-based layer on Al0.6Ga0.4P by Auger electron spectroscopy after etching with SiF4 and SiCl4 [20]. Our models for both the GaP etch rate and the selectivity are fully consistent with the above interpretation. We see evidence for strong surface passivation of AlxGa1−xP as compared to GaP, where the passivation is associated with SF6. Although the results of Epple et al and those presented here demonstrate that fluorine precursors alone, be it SiF4 or SF6, do not etch GaP, our results indicate that the etching reaction is not simply controlled by the SiCl4; the presence of some SF6 5 μm(110)(110)(b)5 μm(111)(111)(a)2 μm2 μm(c)(d)Highly selective dry etching of GaP in the presence of AlxGa1−xP 18 Table 4. Possible reaction products in fluorine- and chlorine-based III-V plasma etch processes and their boiling or sublimation (indicated with s) temperature Tb at atmospheric pressure (101.325 kPa). [36] Product AlF3 AlCl3 GaF3 GaCl3 NF3 PF3 PF5 PCl3 PCl5 AsF3 AsCl3 Tb[◦C] 1276 s 180 s 950 s 201 −128.75 −101.8 −84.6 76.7 160 s 57.8 130 greatly enhances the etch rate (see figure 2(c)). Similar behavior has been observed both for the GaAs/AlxGa1−xAs system [30] and for GaN [35] when etching with SF6 and SiCl4, although it is not as dramatic. One contributing factor may be the rate of removal of the group V element, as argued by Karouta et al [35] for etching of GaN. The group V halides all have relatively low boiling or sublimation temperatures (table 4), but the temperatures for the fluorides are lower. PF3, PF5, and NF3 are especially volatile, evaporating well below room temperature. Thus, even though chlorine species may be primarily responsible for etching, fluorine atoms may contribute to the etch process, particularly in the absence of aluminum. It should be noted though that, if volatility of the reaction products were the only decisive factor, removal of gallium would become rate limiting. The effect of facile removal of phosphorous must therefore be to lower the reaction barrier to form the gallium containing products. But this explanation alone is not sufficient. We need to account for our etch rates and selectivity being orders of magnitude higher with a mixture of SF6 and SiCl4 than those previously published for etching with SiF4 and SiCl4. We believe the relative strength with which sulfur and silicon bind fluorine and chlorine may instead be responsible. Table 5 lists bond dissociation energies for a series of silicon and sulfur halide molecules. Although we do not know what species are present in the plasma, and there may be a variety of both ions and radicals generated, it is nevertheless clear that fluorine is much more strongly bound to silicon than chlorine, and both halogen atoms are much more weakly bound to sulfur than to silicon. The use of SF6 instead of SiF4 may thus lead to a transfer of fluorine atoms to silicon but not chlorine atoms to sulfur. The result would be an increase in reactive chlorine species impinging on the surface to be etched. In fact, during etching of GaAs, Salimian et al [30] detected an increase in chlorine atoms in emission spectra when SF6 was added to SiCl4 plasmas. A related explanation has been proposed to describe etching of GaN with SiCl4/SF6/Ar chemistry [35]. The net result is that both the rate for etching GaP and the selectivity are strongly dependent on the ratio for SF6 to SiCl4 and the maximum values are reached when both gases are present. We note though that a proper analysis would need to take into account activation energies and the corresponding reaction barriers. Highly selective dry etching of GaP in the presence of AlxGa1−xP 19 Table 5. Bond dissociation energies Do molecules. [36 -- 38] 298 for a series of silicon and sulfur halide Bond F − SiF3 F − SF5 Cl − SiCl3 Cl − SF5 Cl − SCl Do 298[kJ/mol] 669.44 391.6 464.4 < 272 293 5. Conclusions We successfully demonstrated highly selective etching of GaP in the presence of AlxGa1−xP with a plasma combining SF6 and SiCl4. A design of experiments implemented to model the parameter space yielded a predicted maximum selectivity of 15000:1 or more for etching of GaP over AlxGa1−xP with an Al content as low as 3% while simultaneously achieving GaP etch rates of several thousand nm/min. For the parameters tested, selectivities of up to 2700:1 and GaP etch rates above 3000 nm/min were measured experimentally. These results contrast with the previous work using SiF4 and SiCl4 claiming the need for high aluminum content in order to form an etch stop layer and for which etch rates and selectivity were two orders of magnitude lower. Use of a mixture of SiCl4 and SF6 instead of a purely chlorine-based plasma is essential for amplifying the GaP etch rate, which we predict can be tuned to values approaching 10 000 nm/min. Although these high-etch-rate and high-selectivity processes exhibit a crystal orientation-dependent morphology, we believe that it may be possible to find less aggressive conditions suitable for pattern transfer applications while maintaining sufficient selectivity. With this advancement in the state of the art of GaP etching, new processing schemes become possible, such as those involving bonding of GaP onto carrier substrates. This opens the door to a variety of new GaP-based integrated nanophotonic applications. Acknowledgments This work was supported by funding from the European Unions Horizon 2020 Programme for Research and Innovation under grant agreement No. 722923 (Marie Curie H2020-ETN OMT) and grant agreement No. 732894 (FET Proactive HOT). References [1] M Kneissl et al. Advances in group III-nitride-based deep UV light-emitting diode technology. Semiconductor Science and Technology, 26(1):014036, 2010. [2] M H Chang, D Das, P V Varde, and M Pecht. Light emitting diodes reliability review. Microelectronics Reliability, 52(5):762 -- 782, 2012. [3] F A Kish et al. Very high-efficiency semiconductor wafer-bonded transparent-substrate (AlxGa1−x)0.5In0.5P/GaP light-emitting diodes. Applied Physics Letters, 64(21):2839 -- 2841, 1994. [4] F A Ponce and D P Bour. Nitride-based semiconductors for blue and green light-emitting devices. Nature, 386(6623):351 -- 359, 1997. [5] M Yamaguchi, T Takamoto, K Araki, and N Ekins-Daukes. Multi-junction III -- V solar cells: current status and future potential. Solar Energy, 79(1):78 -- 85, 2005. Highly selective dry etching of GaP in the presence of AlxGa1−xP 20 [6] A W Bett, F Dimroth, G Stollwerck, and O V Sulima. III -- V compounds for solar cell applications. Applied Physics A, 69(2):119 -- 129, 1999. [7] A W Fang, H Park, O Cohen, R Jones, M J Paniccia, and J E Bowers. Electrically pumped hybrid AlGaInAs-silicon evanescent laser. Optics express, 14(20):9203 -- 9210, 2006. [8] P Dong, T-C Hu, T-Y Liow, Y-K Chen, C Xie, X Luo, G-Q Lo, R Kopf, and A Tate. Novel integration technique for silicon/III -- V hybrid laser. Optics express, 22(22):26854 -- 26861, 2014. [9] D L Mathine. The integration of III -- V optoelectronics with silicon circuitry. IEEE Journal of Selected Topics in Quantum Electronics, 3(3):952 -- 959, 1997. [10] T E Kazior, R Chelakara, W Hoke, J Bettencourt, T Palacios, and H S Lee. High performance mixed signal and rf circuits enabled by the direct monolithic heterogeneous integration of GaN HEMTs and Si CMOS on a silicon substrate. In Compound Semiconductor Integrated Circuit Symposium (CSICS), 2011 IEEE, pages 1 -- 4. IEEE, 2011. [11] S J Pearton. Reactive ion etching of III -- V semiconductors. International Journal of Modern Physics B, 8(14):1781 -- 1876, 1994. [12] M Shearn, X Sun, M D Henry, A Yariv, and A Scherer. Advanced plasma processing: etching, deposition, and wafer bonding techniques for semiconductor applications. Intech, 2010. [13] J W Lee, M W Devre, B H Reelfs, D Johnson, J N Sasserath, F Clayton, D Hays, and S J Pearton. Advanced selective dry etching of GaAs/AlGaAs in high density inductively coupled plasmas. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 18(4):1220 -- 1224, 2000. [14] F Karouta, B Jacobs, O Schoen, and M Heuken. Chemical and complementary role of fluorine in a chlorine-based reactive ion etching of GaN. physica status solidi (a), 176(1):755 -- 758, 1999. [15] M Volatier, D Duchesne, R Morandotti, R Ares, and V Aimez. Extremely high aspect ratio GaAs and GaAs/AlGaAs nanowaveguides fabricated using chlorine ICP etching with N2-promoted passivation. Nanotechnology, 21(13):134014, 2010. [16] T Maeda, J W Lee, R J Shul, J Han, J Hong, E S Lambers, S J Pearton, C R Abernathy, and W S Hobson. Inductively coupled plasma etching of III -- V semiconductors in BCl3-based chemistries: I. GaAs, GaN, GaP, GaSb and AlGaAs. Applied surface science, 143(1):174 -- 182, 1999. [17] T Maeda, J W Lee, R J Shul, J Han, J Hong, E S Lambers, S J Pearton, C R Abernathy, and W S Hobson. Inductively coupled plasma etching of III -- V semiconductors in BCl3- based chemistries: II. InP, InGaAs, InGaAsP, InAs and AlInAs. Applied surface science, 143(1):183 -- 190, 1999. [18] H Hahn, J B Gruis, N Ketteniss, F Urbain, H Kalisch, and A Vescan. Influence of mask material and process parameters on etch angle in a chlorine-based GaN dry etch. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 30(5):051302, 2012. [19] J W Lee, C J Santana, C R Abernathy, S J Pearton, and F Ren. Wet chemical etch solutions for AlxGa1−xP. Journal of the Electrochemical Society, 143(1):L1 -- L3, 1996. [20] J H Epple, C Sanchez, T Chung, K Y Cheng, and K C Hsieh. Dry etching of GaP with emphasis on selective etching over AlGaP. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena, 20(6):2252 -- 2255, 2002. [21] K Schneider, P Welter, Y Baumgartner, S Honl, H Hahn, L Czornomaz, and P Seidler. In Optomechanics with one-dimensional gallium phosphide photonic crystal cavities. Quantum Nanophotonics, volume 10359, page 103590K. International Society for Optics and Photonics, 2017. [22] R H Myers, D C Montgomery, and C M Anderson-Cook. Response surface methodology: process and product optimization using designed experiments. John Wiley & Sons, 2016. [23] L Ilzarbe, MJ ´Alvarez, E Viles, and M Tanco. Practical applications of design of experiments in the field of engineering: a bibliographical review. Quality and Reliability Engineering International, 24(4):417 -- 428, 2008. [24] R A Fisher. The design of experiments. Oliver And Boyd; Edinburgh; London, 1937. [25] SAS Institute Inc., Cary, NC. JMP R(cid:13), Version 13.1.0., 1989-2007. [26] A Addamiano. X-ray data for the phosphides of aluminium, gallium and indium. Acta [27] H C Gatos and M C Lavine. Characteristics of the {111} surfaces of the III -- V intermetallic Crystallographica, 13(6):505 -- 505, 1960. compounds. Journal of the Electrochemical Society, 107(5):427 -- 433, 1960. [28] Y Tarui, Y Komiya, and Y Harada. Preferential etching and etched profile of GaAs. Journal of The Electrochemical Society, 118(1):118 -- 122, 1971. Highly selective dry etching of GaP in the presence of AlxGa1−xP 21 [29] K Schneider, P Welter, Y Baumgartner, H Hahn, L Czornomaz, and P Seidler. Gallium Journal of Lightwave Technology, under phosphide-on-silicon dioxide photonic devices. review. [30] S Salimian and C B Cooper III. Selective dry etching of GaAs over AlGaAs in SF6/SiCl4 mixtures. Journal of Vacuum Science & Technology B: Microelectronics Processing and Phenomena, 6(6):1641 -- 1644, 1988. [31] K L Seaward, N J Moll, D J Coulman, and W F Stickle. An analytical study of etch and etch-stop reactions for GaAs on AlGaAs in CCl2F2 plasma. Journal of applied physics, 61(6):2358 -- 2364, 1987. [32] C M Knoedler and T F Kuech. Selective GaAs/AlxGa1−xAs reactive ion etching using CCl2F2. Journal of Vacuum Science & Technology B: Microelectronics Processing and Phenomena, 4(5):1233 -- 1236, 1986. [33] J Vatus, J Chevrier, P Delescluse, and J-F Rochette. Highly selective reactive ion etching IEEE Transactions on Electron applied to the fabrication of low-noise algaas gaas fet's. Devices, 33(7):934 -- 937, 1986. [34] K Hikosaka, T Mimura, and K Joshin. Selective dry etching of AlGaAs-GaAs heterojunction. Japanese Journal of Applied Physics, 20(11):L847, 1981. [35] F Karouta, B Jacobs, P Vreugdewater, N G H van Melick, O Schoen, H Protzmann, and M Heuken. High etch rate and smooth morphology using a novel chemistry in reactive ion etching of GaN. Electrochemical and solid-state letters, 2(5):240 -- 241, 1999. [36] D R Linde, editor. CRC Handbook of Chemistry and Physics. CRC Press, 84th edition, 6 2003. [37] R Walsh. Bond dissociation energy values in silicon-containing compounds and some of their [38] B D Gailbreath, C A Pommerening, S M Bachrach, and L S Sunderlin. Potential energy surface implications. Accounts of Chemical Research, 14(8):246 -- 252, 1981. of SCl− 3 . The Journal of Physical Chemistry A, 104(13):2958 -- 2961, 2000.
1909.06158	2	1909	2019-09-27T10:08:14	Exciton-Polaritons in Uniaxially Aligned Organic Microcavities	[ "physics.app-ph", "cond-mat.mes-hall", "cond-mat.soft" ]	Here we report the fabrication and optical characterization of organic microcavities containing liquid-crystalline conjugated polymers (LCCPs): poly(9,9-dioctylfluorene-co-benzothiadiazole) (F8BT), poly(9,9-dioctylfluorene) (PFO) and poly(2,7-(9,9-dihexylfluorene)-co-bithiophene) (F6T2) aligned on top of a thin transparent Sulfuric Dye 1 (SD1) photoalignment layer. We extract the optical constants of the aligned films using variable angle spectroscopic ellipsometry and fabricate metallic microcavities in which the ultrastrong coupling regime is manifest both for the aligned and non-aligned LCCPs. Transition dipole moment alignment enables a systematic increase in the interaction strength, with unprecedented solid-state Rabi splitting energies up to 1.80 eV for F6T2, the first to reach the visible spectrum; with an optical gap of 2.79 eV this also gives the highest-to-date organic microcavity coupling ratio, 65$\%$. We also demonstrate that the coupling strength is polarization-dependent with bright polaritons photoluminescence for TE polarization parallel to the transition dipoles and either no emission or weakly coupled emission from the corresponding TM polarization. The use of uniaxally aligned organic microcavities with switchable coupling strength offers exciting prospects for direct observations of ultrastrong coupling signatures, quantum simulation, polaritonics and condensation related phenomena.	physics.app-ph	physics	Exciton-Polaritons in Uniaxially Aligned Organic Microcavities F. Le Roux,∗,† R. A. Taylor,† and D. D. C. Bradley‡,‡ †Department of Physics, University of Oxford, Parks Road, Oxford, OX1 3PU, U.K. ‡Physical Science and Engineering Division, King Abdullah University of Science and Technology, Thuwal, 23955-6900, Saudi Arabia E-mail: [email protected] Abstract Here we report the fabrication and optical characterization of organic microcavities containing liquid-crystalline conjugated polymers (LCCPs): poly(9,9-dioctylfluorene- co-benzothiadiazole) (F8BT), poly(9,9-dioctylfluorene) (PFO) and poly(2,7-(9,9-dihexyl fluorene)-co-bithiophene) (F6T2) aligned on top of a thin transparent Sulfuric Dye 1 (SD1) photoalignment layer. We extract the optical constants of the aligned films using variable angle spectroscopic ellipsometry and fabricate metallic microcavities in which the ultrastrong coupling regime is manifest both for the aligned and non-aligned LCCPs. Transition dipole moment alignment enables a systematic increase in the in- teraction strength, with unprecedented solid-state Rabi splitting energies up to 1.80 eV for F6T2, the first to reach energies comparable to those in the visible spectrum; with an optical gap of 2.79 eV this also gives the highest-to-date organic microcavity coupling ratio, 65%. We also demonstrate that the coupling strength is polarization- dependent with bright polaritons photoluminescence for TE polarization parallel to the transition dipoles and either no emission or weakly coupled emission from the cor- responding TM polarization. The use of uniaxally aligned organic microcavities with 1 switchable coupling strength offers exciting prospects for direct observations of ultra- strong coupling signatures, quantum simulation, polaritonics and condensation related phenomena. 2 The strong coupling (SC) regime in the solid-state is entered 1 when the interaction be- tween the electric component of a confined electromagnetic field and the excitations present within a semiconductor becomes sufficiently intense that their original energy levels are re- placed by so-called polariton hybrid states of light and matter, separated by a Rabi splitting energy ΩR. Organic semiconductor Frenkel excitons are an interesting alternative to the more traditional Wannier excitons seen in III-V inorganic semiconductors for the study of exciton-polaritons thanks to their large binding energies (EB ∼ 0.5 ± 0.25 eV 2,3) which allow room-temperature observation of varied phenomena including Bose-Einstein conden- sation, 4,5 superfluidity of light 6 and optical logic. 7 Their large intrinsic oscillator strengths 8 combined with the small mode volumes Vm of metallic microcavities 9 have enabled ΩR ≥ 1 eV, 10 -- 13 with values up to ΩR = 1.12 eV. 14,15 This splitting is directly comparable to the ≥ 20%, thereby exciton transition energy ωex an yields normalized coupling ratios g = ΩR ωex crossing into ultrastrong coupling (USC), an interaction space that has received great recent attention, with attractive research perspectives and multiple emerging applications. 16,17 Ex- perimental realizations of increasingly higher coupling ratios have also been reported for inor- ganic semiconductor based intersubband polaritons, 18 and other physical systems, including superconducting circuits, 19 Landau polaritons 20 and plasmonic picocavities interacting with vibrational degrees of freedom of individual molecules. 21 For an ensemble of organic semiconductor excitons within a cavity, ΩR scales with the square root of ωex according to: 22 -- 24 ΩR = 2µ.E (cid:114) Nωex 2effVm , (1) where µ is the transition dipole moment, E the electric field, N the number of molecules, ωex the exciton transition energy, eff the cavity effective permittivity and Vm the cavity mode volume. One way to increase the value of the coupling ratio has then been to work with lower energy excitons, as done by Barachati et al., 25 resulting in a then record g = 62%. This 3 approach is inherently accompanied by a reduction of ΩR compared with the use of excitons lying at higher energies with equivalent oscillator strengths. The alternative is to look to increase ΩR; The most direct routes to achieve this include (i) increasing N , which although generally not straightforward can be done, for example, by reducing the bulkiness of con- jugated polymer solubilizing groups 26 (ii) increasing µ through conformational control 27,28 or a photo-switchable configuration change, 9 and/or (iii) increasing µ.E. In the latter case, uniaxial orientation has been shown to enhance conjugated polymer thin film refractive in- dex (and correspondingly transition dipole moment) in the direction parallel to the chain orientation axis 29,30 yielding an enhanced dot product for a suitable polarization of E. This offers a clear route to enhancing ΩR that is demonstrated below, using a photoalignment process to achieve thermotropic liquid crystalline conjugated polymer (LCCP) chain orien- tation. Two recent reports on the coupling of liquid crystal (LC) vibrational modes 31 and carbon nanotubes Wannier excitons 32 have also shown that µ.E can be maximized in this way and that polarization-dependence allows for applications discussed further in the text. Previous approaches to LCCP orientation typically used a traditional rubbed polyimide (PI) alignment layer onto which the polymer was spin coated prior to thermal treatment. 33 The clearing temperatures of LCCPs are relatively high (∼200-300◦C), 33,34 leading to the requirement for a high temperature stable PI, for which there are limited commercial op- tions. Precursor route poly(p-phenylenevinylene) has also been used as an alternative rubbed alignment layer, having the advantage of temperature stability and an electronic structure that more readily permits charge injection from the underlying electrode to the LCCP. 35 Other approaches to orientation include stretching and rubbing the conjugated polymer, 36 Langmuir-Blodgett deposition, 37 use of an aligned host matrix 38 or nanoimpriting. 39 Success- ful fabrication of polarized light emitting diodes 35,40,40 -- 43 and polarized photoluminescence structures 29 has resulted, together with intrachain mobility enhanced transistors. 44 For a variety of practical reasons, oriented LCCPs have, however, not been used before in strongly or ultrastrongly-coupled microcavities. 4 Non-contact photoalignment of LC mesophases has emerged 45 as a promising alternative to rubbing-induced alignment. Among photoalignment layer materials, the azobenzene- containing Sulfuric Dye 1 (SD1) has shown high temperature stability and remarkable quality for the alignment of low molecular weight LCs. 46,47 Recently, the orientation of poly(9,9- dioctylfluorene-co-benzothiadiazole) (F8BT) using SD1 has also been observed 48 in which a thermal treatment enables the orientation of the LCCP. Potential advantages of using SD1 photoalignment for photonics and polaritonics are threefold, namely that the SD1 layer can be very thin (≤ 5 nm), that it is almost transparent in the visible (peak absorption at ∼ 3.25 eV) and that it is a patternable process which allows a straightforward way to fabricate novel photonic structures. 47,49 We report here a detailed study of the use of oriented LCCP films within metallic mi- crocavities. F8BT, poly(9,9-dioctylfluorene) (PFO) and poly(2,7-(9,9-dihexylfluorene)-co- bithiophene) (F6T2) films are oriented with SD1 photoalignment layers and in all three cases we demonstrate a systematic enhancement of ΩRTE, for TE-polarized light parallel to the chain orientation direction, compared to non-aligned reference samples. The maximum ΩRTE = 1.80 ± 0.01 eV (689 nm) is for F6T2, a value that would sit within the visible spectrum. This structure also gives the largest normalized coupling ratio, g = 65%, reported to date for an organic semiconductor microcavity. Photoluminescence for TE-(parallel to the alignment direction) and corresponding TM-polarizations makes the changes in coupling strength between polarizations evident for all three polymers. Going beyond enhancement of the Rabi-splitting energy, we discuss the potential use of uniaxially aligned organic micro- cavities for demonstration of the elusive polaritonic NOT gate, 50 -- 52 for quantum simulation through complex energy landscapes and more generally its advantages for the realization of polarization sensitive devices, lasing and condensation related phenomena. 5 Results And Discussion The optical constants for thin films of PFO, F8BT, F6T2 (see Methods for all fabrication protocols) were extracted using Variable Angle Spectroscopic Ellipsometry (VASE). As the polymer chains tend to lie in the plane of the film, 53 -- 55 the resulting optical constants are well fitted 27,56 using an in-plane/out-of plane anisotropic model yielding the components nord, kord, nex, kex of the complex refractive index n = n + ik. Figure 1 shows (in green) the in-plane optical constants (nord, kord) for F8BT (b), PFO (c) and F6T2 (d) (The complete sets of optical components extracted are available in Supplementary Information). All spectra are comprised of either one or several inhomogeneously broadened distributions (EXPFO at around 3.23 eV, EXF8BT1 with excitation states lying above 5 eV associated to ring-localized fluorene states. 57 at respectively 3.82 eV and 2.70 eV and EXF 6T 2 at 2.79 eV) and EXF8BT2 Following photoalignment of an SD1 spincoated film in the in-plane y direction, the optical components nx, ny, nz, kx, ky, kz were extracted using a biaxial anisotropic model 58 and are shown in Figure 1(a). As expected, the alignment brings about intense ny, ky optical components compared to their perpendicular counterparts nx, kx. Polymer layers of F8BT, PFO and F6T2 were then spincoated on top of the photoaligned SD1, thermally annealed into their respective nematic phases (250◦C for F8BT, 160◦C for PFO and 220◦C for F6T2) and subsequently quenched to room-temperature causing and effectively freezing the in-plane uniaxial alignment in the y direction. The corresponding optical components of the obtained films were fitted and are shown in Figure 1(b),(c),(d) with ny, ky substantially larger than nx, kx. As the oscillator strengths in the x and y direction fx,y ∝(cid:82) kx,y(E)dE are proportional to the number of underlying contributing dipoles, we calculate the ratio R = 1− fx fx+fy where fx,y is the oscillator strength in the corresponding direction and estimate the total percentage of dipoles aligned in the y direction. For all excitons, R exceeds 83% (calculated for EXF8BT1 ) with a maximum value of 94% (for EXF6T2) underlining the remarkable alignment quality. 6 Figure 1: (a) Optical components for a thin film of SD1 aligned along the y direction. (b),(c),(d) In green: In-plane nord, kord optical components for spincoated F8BT (b), PFO (c) and F6T2 (d). The in-plane optical components of the films following alignment are shown in blue for ny, ky (parallel to the alignment direction) and in red for nx, kx (perpendicular to the alignment direction) for SD1 (a), F8BT (b), PFO (c) and F6T2 (d). Solid lines give the real component of the complex refractive index n = n + ik, dashed lines the imaginary component. Time-integrated photoluminescence (PL) spectra were recorded at normal incidence for the three polymers and are shown in Figure 2 (see Methods for the experiment geometry). The spectra were measured at normal incidence with the collection polarizer both in vertical (blue) and horizontal (red) positions; for the aligned films, the vertical direction matched the 7 direction of the alignment. In each case, we calculate the integrated ratio RVH =(cid:82) IV(E) IH(E) dE which reveals the presence of in-plane uniaxial alignment in the film. For F8BT, the spectrum of the non-aligned film (Figure 2 (a)) reveals an inhomogeneously broadened distribution with S1 − S0 (0-0), (0-1) vibronic peaks located at 2.29 eV (541 nm) and 2.15 eV (577 nm). We calculate RVH = 1.08, with a deviation from unity being fully accounted for by the degree of polarization of the excitation laser beam; the polymer chains have as expected no preferential in-plane orientation. For the aligned F8BT (2 (b)), the same spectral positions for the vibronic peaks are recorded with a difference in relative heights due to thickness variations between the aligned and non-aligned films. The integrated ratio RVH = 8.3 makes the in-plane preferential alignment of the emitting layer evident (in-depth studies relating the polarized emission to the microscopic parameters of aligned F8BT can be found in Ref 41). For non-aligned PFO (Figure 2(c)), the S1 − S0 well resolved (0-0), (0-1) and (0-2) PL vibronic peaks appear at 2.88 eV (430 nm), 2.71 eV (457 nm) and 2.58 eV (481 nm) with an integrated ratio RVH that increases from 1.09 to 6.9 from non-aligned to aligned film (Figure 2(d)). As the vibronic structure is this time well resolved, we calculate the peak ratios RpVH(E) = IV(E) IH(E) for RpVH(2.88) = 11.2, RpVH(2.71) = 9.5 and RpVH(2.58) = 8.2, results comparable to those obtained previously by using rubbed PI layers. 40 For non-aligned F6T2 (Figure 2(e)), the PL vibronic peaks appear at 2.28 eV (544 nm), 2.12 eV (585 nm) and 1.95 eV (636 nm) and RVH increases from 1.16 to 3.50 (Figure 2(g)), a ratio harder to calculate due to a weaker emission (in which the noise had to be removed using a fast Fourier transform algorithm) compared with F8BT and PFO. 8 Figure 2: PL spectra for non-aligned ((a),(c) & (e)) and aligned ((b),(d) & (f)) F8BT ((a) & (b)), PFO ((c) & (d)) and F6T2 ((e) & (f)) thin films. For a given polymer in a given alignment state, both polarizations spectra were normalized by the same value to allow for their comparisons. Solid blue lines show the PL collected using a polarizer in the vertical direction and red lines in the horizontal direction. In panels (b),(d) and (f): // and ⊥ indicate that the collection polarizer's direction respectively matches or is perpendicular to the direction of the polymer chains in the film. A fast Fourier transform algorithm was used for F6T2 ((e) & (f)) to reduce noise in the recorded signal. 9 Al-Polymer-Al microcavities were fabricated using non-aligned F8BT, PFO, F6T2 and for each cavity, a corresponding Al-Aligned(SD1-Polymer)-Al microcavity was fabricated. Angle-resolved polarized reflectivity maps were recorded for each microcavity by varying the angle θ formed between the vector normal to the microcavity plane and the incident light direction. For the aligned cavities, the measurement was performed at an angle Φ (formed between TE polarization and the polymer chain direction) equal to 0◦. All results were analysed using a Hopfield-Agranovich Hamiltonian 22,59,60 including either one (PFO, F6T2) or two (F8BT) separate excitons with all fitting results displayed in Table 1. For each microcavity, the experimental results are supported by transfer matrix reflectivity (TMR) calculations whose outputs are shown in Supporting Information. The reference results ob- tained for the non-aligned PFO and F8BT cavities are also shown in Supporting Information and agree with previous reports. 27,56 Measured and fitted results for the non-aligned F6T2 cavity for TE polarization are shown in Figure 3 (a) where we observe the Lower (LP) and Upper (UP) Polaritons avoiding crossing of the exciton (EXF6T2 = 2.72 eV) by more than 1 eV, providing clear evidence of USC. Even though the layer is non-aligned, the largest Rabi splitting energy to date Ω0TE = 1.38 ± 0.01 eV is measured, corresponding to a coupling ratio g = 49%. As expected, the results for TM polarization (Figure 4 (a)) displays a flatter angular dispersion for the polaritons as the effective refractive index neffTM of the microcavity increases due to polarization-dependent penetration depth through the metallic mirrors. 10,61,62 For this polarization, the in-plane/out-of-plane anisotropy of the active layer slightly reduces the Rabi energy down to Ω0TM = 1.27± 0.01 eV, as the contribution of the weaker out-of plane (kex) component to the overall interaction increases along with θ. 10 Results for the aligned cavities of PFO, F6T2 and F8BT are shown respectively in Fig- ure 3 (b), (c) and (d) for TE polarization at Φ = 0◦. In each case, USC is clearly observed with fitting results revealing a systematic increase for Ω0TE,Φ=0◦ of ∼ 40% compared to the non-aligned cavities. As the transition dipole moment µ rotates together with the polymer 10 chains in the y-direction, the relative increase in coupling strength can be calculated using 2 ∼ 41% for TE polarization. Different the term µ.E in Equation 1 to be: √ cos(0) cos( π 4 ) = factors can explain deviations to this theoretical value: (i) albeit thin, the SD1 layer can slightly decrease the overall coupling strength by diminishing the overlap between exciton and cavity modes, (ii) all the polymer chains are not perfectly aligned in the y direction with remainding optical activity in the x direction (see Figure 1) reducing the coupling strength, (iii) the reordering of the polymer can affect its microscopic properties. All the Ω0TE,Φ=0◦ values derived here, if not for the one we have just reported for non-aligned F6T2, exceed previous reports: the aligned F8BT cavity exhibits Ω01TE,Φ=0◦ = 1.18 ± 0.01 eV and Ω02TE,Φ=0◦ = 1.25 ± 0.01 eV, the aligned PFO cavity Ω0TE,Φ=0◦ = 1.47 ± 0.01 eV and the aligned F6T2 cavity reaches a value of Ω0TE,Φ=0◦ = 1.80 ± 0.01 eV. To our knowledge, this last value is the first Rabi splitting energy to reach values comparable to photons in the visible spectrum (∼ 689 nm) in a solid-state system, exceeding by more than 60% the previous record 14,15 and corresponds to a slightly higher coupling ratio g ∼ 65% than the one obtained in Ref 25, fact which is remarkable since this result is obtained for an exciton lying ∼ 1.6 eV higher in energy. 11 Figure 3: Experimental, angle-resolved, TE-polarized reflectivity maps for microcavities containing: (a) non-aligned F6T2, (b) aligned PFO, (c) aligned F6T2 and (d) aligned F8BT. For (b), (c) & (d), the measurements were performed at Φ = 0◦ (see definition in the text) as depicted in the experiment schematic shown above the maps. Overlaid solid white lines are the exciton EX and cavity EC modes, black dashed lines are polaritons fitted from the analytical model. 12 The results for TM polarization for the aligned cavities are shown in Figure 4 (b), (c) and (d). In those measurements, the weak coupling regime is clearly observed for F6T2 (c) and F8BT (d) as the LP and UP are no longer visible and replaced by a single photonic mode. For PFO (b), the reflectivity observed does not show proper anti-crossing around the exciton (EXPFO = 3.23 eV) with the PL measurements in the next section supporting the lack of evidence for SC. Figure 4: Experimental, angle-resolved, TM-polarized reflectivity maps for microcavities containing (a) non-aligned F6T2 (b) aligned PFO (c) aligned F6T2 and (d) aligned F8BT. For (b), (c) & (d), the measurements were performed at Φ = 0◦ (see definition in the text) and the overlaid solid white line is the exciton EX. In (a) in addition to the exciton EX, the second white line represents EC the cavity mode and the black dashed lines the polaritons fitted from the analytical model. 13 Table 1: Extracted and pre-set parameter values for microcavities containing non-aligned and aligned polymers, modelled using a Hopfield-Agranovich Hamiltonian. 22,59,60 Values are shown for both TE- and TM-polarization, at Φ = 0◦ (see definition in the text) for the aligned cavities. The initials A and NA respectively designate aligned and non-aligned polymer layers. PFO NA PFO A F8BT NA F8BT A F6T2 NA F6T2 A 3.23 3.23 3.82 3.82 2.79 2.79 1.02 ± 0.01 1.47 ± 0.01 0.84 ± 0.01 1.18 ± 0.01 1.38 ± 0.01 1.80 ± 0.01 2.70 2.70 - 22 0.80 ± 0.01 31 - 49 1.27 ± 0.01 0.84 ± 0.01 1.25 ± 0.01 31 0.80 ± 0.01 46 - 21 30 - - 46 - - - - - 65 - - - - - - Polymer: ω1(eV)1 ω2(eV)2 Ω01TE(eV)3 g01TE%4 Ω01TM(eV)5 g01TM%6 Ω02TE(eV)7 g02TE%8 Ω02TM(meV)9 g02TM%10 11 neffTE neffTM 12 32 0.97 ± 0.01 30 - - - 46 - - - - - - - - - 1.68 ± 0.01 1.70 ± 0.01 1.70 ± 0.01 1.87 ± 0.01 1.64 ± 0.01 1.71 ± 0.01 2.33 ± 0.01 2.37 ± 0.01 2.12 ± 0.01 - - - EcTE(0)(eV)13 3.13 ± 0.02 2.92 ± 0.02 2.54 ± 0.02 2.71 ± 0.02 2.48 ± 0.02 2.78 ± 0.02 EcTM(0)(eV)14 3.15 ± 0.02 1 Exciton oscillator 1 transition energy. 2 Exciton oscillator 2 transition energy. 3 TE-polarized Rabi energy associated with exciton 1 for ωcavTE = ω1 (see definition in the text). 4 TE-polarized normalized coupling ratio energy associated with exciton 1 (see definition in the 2.59 ± 0.02 2.50 ± 0.02 - - - text). 5 TM-polarized Rabi energy associated with exciton 1 for ωcavTM = ω1 (see definition in the text). 6 TM-polarized normalized coupling ratio energy associated with exciton 1 (see definition in the text). 7 TE-polarized Rabi energy associated with exciton 2 for ωcavTE = ω1 (see definition in the text). 8 TE-polarized normalized coupling ratio energy associated with exciton 2 (see definition in the text). 9 TM-polarized Rabi energy associated with exciton 2 for ωcavTM = ω1 (see definition in the text). 10 TM-polarized normalized coupling ratio energy associated with exciton 2 (see definition in the text). 11 Effective refractive index for TE polarization. 12 Effective refractive index for TM polarization. 13 TE-polarized energy of the bare cavity mode at normal incidence. 14 TM-polarized energy of the bare cavity mode at normal incidence. 14 Angle-resolved PL was recorded for each microcavity. The excitation laser used for the thin film PL measurement was focused onto the sample at an incidence of 75◦, the excitation geometry between films and microcavities experiments being identical. For each microcavity, the average power was kept low (≤ 10µW) and the excitation energy was chosen to optically pump one of the intense absorptions arising from the optical transitions of the underlying polymer. For the aligned cavities, the measurement was performed at Φ = 0◦. PL intensity maps plotted by energy vs emission angle for both TE- ((a), (c) and (e))) and TM-polarized ((b), (d) and (f)) emissions from the non-aligned F8BT ((a) and (b)), PFO ((c) and (d)) and F6T2 ((e) and (f)) microcavities are shown in Figure 5. In each case the emission is dominated by a narrow single peak originating from the LP which is relatively insensitive to angular-dispersion: a recognizable feature of USC 10 -- 12 (values including peak positions and full width at half maximum (FWHM) at normal incidence as well as the angular dispersion from 0 to 60◦ are shown in Table 2). The results for PFO agree with a previous report 27 and the variety of polymers used here allows for emissions across the visible spectrum (in the blue at ∼ 452 nm for PFO, green at ∼ 537 nm for F8BT and yellow/orange at ∼ 588 nm for F6T2). Corresponding PL intensity maps for the aligned cavities are shown in Figure 6. TE- polarized emission ((a), (c) and (f)) resembles the one observed from non-aligned microcav- ities with a single peak emitted from the LP. It however differs in energy for F8BT and PFO between non-aligned and aligned cavities: the peak emission is recorded at 2.13 eV at normal incidence for aligned F8BT compared to 2.31 eV when non-aligned, and at 2.59 eV for aligned PFO compared to 2.74 eV when non-aligned. These redshifts are not the result of different thicknesses between the cavities as TMR calculations show that for each pair, the polymer layer thicknesses are comparable (the aligned F8BT is 110 nm-thick compared to 118 nm when aligned, the aligned PFO 96 nm-thick compared to 97 nm when aligned) but are direct evidence of the increased interaction strength which repels the UP and LP to respectively higher and lower energies. The emission for aligned and non-aligned F6T2 is 15 closer in energy at normal incidence (2.11 eV aligned compared with 2.10 eV non-aligned) and is this time the result of a much larger thickness of the non-aligned F6T2 layer (the aligned F6T2 is 94 nm-thick compared to 123 nm non-aligned), resulting in lower energy cavity modes and LP (as can be observed in Figure 3) which makes up for the difference in interaction strength between the two microcavities. TM-polarized measurements ((b), (d) and (f)) show no distinguishable emission for F8BT (b) and F6T2 (g) as the only photonic mode in both those cavities lies too high in energy (at ∼ 3 eV for F6T2 and ∼ 2.9 eV for F8BT) to allow any emission from the underlying polymer (Figure 2 (b) and (f) shows the emission located at energies lower than 2.6 eV for both F6T2 and F8BT). For the microcavity containing aligned PFO (d), a 5-times weaker, broad, angle- independent TM-polarized emission was detected (f), with two maxima at 2.88 and 2.71 eV coinciding with the ones from the bare film (Figure 2(d)). This broad, structured, angle- insensitive dispersion confirms that the microcavity no longer operates under USC for this polarization. The emission itself is only allowed through the photonic mode at ∼ 2.88 eV (see 2(b)) and even though this photonic mode overlaps with the most intense part of the bare film's PL, the resulting signal is much weaker in TM than TE polarization as the emission in the direction perpendicular to the chain alignment is intrinsically much weaker. 16 Figure 5: Angle-resolved PL spectral intensity maps for microcavities containing non-aligned F8BT ((a) & (b)), PFO ((c) & (d)) and F6T2 ((e) & (f)), with TE ((a), (c), (e)) and TM ((b), (d), (f)) polarized spectra plotted separately. The excitation energy is overlaid in white for each measurement. Each polarization pair ((a) & (b)), ((c) & (d)), ((e) & (f)) is normalized by the same value to allow for comparison. 17 Figure 6: Angle-resolved PL spectral intensity maps for microcavities containing aligned F8BT ((a) & (b)), PFO ((c) & (d)) and F6T2 ((e) & (f)), with TE ((a), (c), (e)) and TM ((b), (d), (f)) polarized spectra plotted separately. The measurements were performed at Φ = 0◦ (see definition in the main text) . Each polarization pair ((a) & (b)), ((c) & (d)), ((e) & (f)) is normalized by the same value to allow for comparison. 18 Table 2: Peak positions and FWHMs at normal incidence for the TE- and TM-polarized emission s displayed in Figure 5 and 6. The angular dispersion of the emission for both polarizations from 0 to 60◦ is also reported. The initials A and NA respectively designate aligned and non-aligned polymer layers. Polymer: PFO NA PFO A F8BT NA F8BT A F6T2 NA F6T2 A Peak position (eV) FWHMTE (meV) FWHMTM (meV) TE dispersion (meV) 2.74 2.59 99 ± 1 140 ± 5 99 ± 1 130 50 - TM dispersion (meV) 110 - 2.31 89 ± 5 82 ± 5 30 20 - 80 - 2.13 2.11 127 ± 1 109 ± 1 109 ± 1 2.10 109 ± 1 110 110 - 60 - Several opportunities unique to our photoalignment technique for exciton-polaritons ex- ist: the possibility to align SD1 at mesoscopic scales 47 -- 49 allows local alignment of the polymer chains, which could lead to the fabrication of the missing NOT gate in optical logic thanks to spontaneous splittings between TE and TM polarizations, 50 -- 52 rapid switching of the coupling strength using LCPPs and LCs in microcavities could also be an opportunity to perform direct extraction of the USC ground state virtual contents 16,17,22 and the fabrication of complex energy landscapes, especially when combined with the tunability of the molecu- lar structure (for example by generating segments of β-phase in PFO 63) could help address challenges in quantum simulation. 64 On a more practical standpoint, we expect that a wide breadth of polarization sensitive devices and phenomena such as Bose-Einstein condensation and exciton-polariton lasing (which has recently been demonstrated using pentafluorene 65) will take advantage of the polarization-dependent coupling (the spontaneous polarization observed during lasing and condensation could for instance be controlled by molecular align- ment). Conclusions We have fabricated organic microcavities containing LCCPS (F8BT, PFO, F6T2) aligned using a thin photoalignment layer (SD1). The USC regime was first observed for the non- 19 aligned microcavities with a value of ΩRF6T2 = 1.34 eV exceeding previous reports. The alignment then allowed for a systematic increase of the coupling strength in the direction of the alignment with giant values of ΩR culminating at 1.80 eV for F6T2, a value comparable to photon energies in the visible spectrum, also corresponding to the highest coupling ratio g = 65% to date in the solid-state. 16,17 Angle-resolved PL for the TE polarization parallel to the alignment direction revealed red-shifted LP emissions compared to the non-aligned cavities, a signature of the increased interaction strength. In this geometry, the absence of polaritons in TM-polarized reflectivity and weak or no PL also demonstrated that the coupling strength was polarization dependent. By using three different polymers, we demonstrated that the alignment can be generalized to other LCPPs and that a real opportunity to reach coupling ratios close to 90% exists if the alignment can be applied at lower energies. Aligning LC- CPS in microcavities at microscopic scales using SD1 also offers further possibilities for the realization of polaritonic devices and rich energy landscapes. Methods Materials The three polymers used in this study were supplied by Cambridge Display Technology (F8BT), Sumitomo Chemical (PFO and F6T2) and used as received. Their peak molecular weights were: MpPFO = 50 × 103 g.mol−1, MpF8BT = 77 × 103 g.mol−1, MpF6T2 = 80 × 103 g.mol−1. The azo-dye photo-alignement layer SD1 was supplied by Dai-Nippon Ink and Chemicals, Japan. Anhydrous toluene (99.8%), anhydrous chloroform (≥99%) and anhy- drous 2-methoxyethanol(≥99.8%) were purchased from Sigma-Aldrich. Solvents were used as received. For the mirror fabrication, Aluminium pellets (99.999%) were purchased from Kurt J. Lesker. 20 Film Fabrication The bare films (used for ellipsometry and PL) of SD1, PFO, F8BT, F6T2 were spincoated from solutions in 2-methoxyethanol (SD1 at 1 mg.mL−1), toluene (PFO at 18 mg.mL−1 and F8BT at 18 mg.mL−1) and chloroform (F6T2 at 13 mg.mL−1). All solutions were prepared in an inert environment, left to stir overnight at a temperature of 55◦C except for F6T2 in chloroform which was left stirring at room temperature. All solutions were then filtered using a 0.45 µm PTFE filter. All samples were spincoated on top of fused silica substrates. The non-aligned polymer films were spun for 1 min at a speed of 2000 rpm with an acceleration of 1200 rpm.s−1. For the aligned polymer films, the SD1 layer was first spincoated by spinning 5 s at 500 rpm (acceleration 500 rpm.s−1) and 25 s at 2000 rpm (acceleration 1200 rpm.s−1). The film was then annealed for 6 minutes at a temperature of 150◦C to drive any traces of solvent away. The alignment of the SD1 layer was performed in air by exposing the sample to 5 mW of polarized UV light (emitted by a M365LP1 LED in front of a broadband WP25M-UB polarizer from Thorlabs) for 10 minutes; The SD1 chains aligning perpendicular to the direction of the polarized light. 46,47 The samples were then put back in an inert environment where the polymer layer was spincoated by spinning for 1 min at a speed of 2000 rpm with an acceleration of 1500 rpm.s−1. Each polymer was then thermally annealed into their respective nematic phases (160◦C for PFO, 220◦C for F6T2, 250◦C for F8BT) using a Linkam THMS600 heating stage with a heating rate of 30◦C.min−1. This temperature was held for 10 minutes and subsequent quenching to room temperature was realized by quickly placing the sample on the metallic floor of the glovebox while applying a gentle flow of nitrogen. Microcavity Fabrication The aluminium mirrors were evaporated at a rate of 10 A.s−1 at a pressure of 10−9 mbar. For the non-aligned cavities, the spincoating conditions used on top of the bottom mirror were identical to the ones described for the bare films. For the aligned cavities, the SD1 21 layer was spincoated on top of the bottom mirror. The structures fabricated had low Q factors ∼ 25, characteristic of metallic microcavities using Aluminium. The concentration of the SD1 solution in 2-methoxyethanol was increased to 3 mg.mL−1 as spincoating SD1 on a metallic surface results in lower thicknesses than on fused silica. A slightly thicker layer also acts as a protection layer to prevent the aluminium from reacting with the polymer upon annealing at high temperatures (using 1 mg.mL−1 SD1 solution resulted in samples unfit for measurement). The sample was then annealed 6 minutes at a temperature of 150◦C to drive any traces of solvent away and the rest of the alignment procedure was similar to the one used for the bare films. For the spincoating of the polymer layers, the solution concentrations were in some cases adjusted so as to adjust the thicknesses and therefore the cavity mode energy. The concentrations used in the non-aligned microcavities were: PFO at 19 mg.mL−1 , F8BT at 20 mg.mL−1 and F6T2 at 15 mg.mL−1. The concentrations for the aligned microcavities were: PFO at 18 mg.mL−1, F8BT at 18 mg.mL−1, F6T2 at 13 mg.mL−1. Optical Characterization The optical constants for the non-aligned and aligned films of PFO, F6T2 and F8BT were ex- tracted using a J.A. Woollam ESM-300 ellipsometer. For each sample, 8 reflection-geometry measurements were performed with light incident from 45◦ to 61◦ (angles of incidence are quoted relative to the plane normal) together with a normal incidence (0◦) transmission mea- surement, for the aligned films the measurement was performed at Φ = 0◦. The reflectivity maps obtained in Figure 1 were obtained using a home-built white light reflectivity setup. The microcavities were placed at the center of a stage with two independent rotating arms. A deuterium-halogen light source (DH-2000-DUV from Ocean Optics) was coupled into a fiber whose output was collimated onto a broadband polarizer (WP25M-UB from Thorlabs) and onto the sample (final spot size 1 mm). The reflected light was then coupled into a second fiber placed onto the second arm and analyzed using a spectrometer (HRS 500, 150 g/mm grating with blazing wavelength at 300 nm) and CCD (Pylon-2KB CCD from Prince- 22 ton Instruments). In all cases, a neat aluminium mirror with known reflectivity was used as reference. The acquisition angle was varied by 0.5◦ steps from 10 to 70◦. Photoluminescence All the time-integrated PL measurements were performed on the same home-built setup. The fiber coupled to the white light source was disconnected and the first arm was positioned at 75◦ relative to the sample plane normal. The pulsed laser beam from a supercontinuum white light laser (SuperK Extreme with its UV spectral extension unit Extend-UV, NKT Photonics) was used as excitation source and focused onto the sample (spot size < 1 mm). The excitation energy was tuned according to the optical transitions, the incident power was kept low (≤ 10µW ) with pulse widths of 20 to 30 ps and a repetition rate of 77.87 MHz. The broadband polarizer was placed on the collection arm, at a distance of 10 cm from the sample. Two nearly closed irises (∼ 1cm) at a distance of 5 cm from each other were then placed before the coupling lens of the collection fiber in order to ensure that the collected light was emitted at the desired angle in the horizontal plane. The light was then analysed using the spectrometer and CCD described before using this time a 300 g/mm grating blazed at a wavelength of 500 nm. For the polymer films, the light was collected at normal incidence. For the microcavities the acquisition angle was varied by 1◦ steps from -40 to 60◦. Data Analysis The minima of the reflectivity maps shown in Figures 3 and 4 were analyzed using a least- square fitting algorithm for the eigenvalue problem: Hqvi,q = ωi,qvi,q, (2) where Hq is an extension of the Agranovich/Hopfield Hamiltonian containing either one 22,59,60 (PFO, F6T2) or two excitonic resonances 27 (F8BT): 23 ωcav,q + 2Dq −i Ω1,q 2 −i Ω2,q ω1 0 2  Hq = i Ω1,q 2 i Ω2,q 2 2Dq −i Ω1,q −i Ω2,q 2 2  , (3) −i Ω1,q 2 −i Ω2,q 2 0 0 0 0 0 −ω2 −2Dq −i Ω1,q −i Ω2,q 2 i Ω1,q 2 i Ω2,q 2 −i Ω1,q 2 ω2 2 −i Ω2,q 2 −ωcav,q − 2Dq −i Ω1,q −ω1 0 0 2 −i Ω2,q 2 0 0 0 0 In the case of a single exciton oscillator, Hq reduces to the usual 4 x 4 Hopfield-like USC matrix. 10 In Eq. (2), (3), q is the in-plane wave vector, ωcavq the cavity mode energy, ωj the frequency for the j-excitons, Ωj,q is the associated Rabi frequency, and for a given angle θ: Ωj,q = Ωj(θ) = Ω0j ωcav(θ) where Ω0j is the Rabi frequency on resonance for the j-excitons. It was shown that in metal-organic semiconductor-metal cavities ωcav(θ) can be (cid:113) ωj approximated by: 10 (cid:33)− 1 ωcav(TE,TM),q = ωcav(TE,TM)(θ) = ωcav(0) (cid:32) where neffTE,TM is polarization dependent. Finally, Dq =(cid:80) In order to diagonalize H, the polariton annihilation operators pi,q = wi,qaq +(cid:80) −q +(cid:80) j,−q for i{LP, MP, UP} are introduced, where aq and a† † squared magnetic vector potential. 1 − sin2(θ) eff (TE,TM) Ω2 j,q 4ωj j zi,j,qb yi,qa n2 † j 2 (4) is the contribution of the j xi,j,qbj,q + q respectively an- nihilate and create a photon at frequency ωcavq , bj and b † j respectively annihilate and create a j-exciton at frequency ωj. The terms w, x, y and z label, respectively, the photon, exciton, anomalous photon and anomalous exciton Hopfield coefficients. The eigenvalues of Hq were fitted to the experimental results for each cavity, for both TE- and TM-polarization, using the R-minima in the 10 - 70◦ range. In order to minimize the number of fitting parameters and obtain meaningful results, only ωcavTE,TM(0), neffTE,TM & Ω01TE,TM were allowed to vary in fittings of the PFO and F6T2 24 cavities. Similarly, only ωcavTE,TM(0), neffTE,TM, Ω01TE,TM & Ω02TE,TM were allowed to vary in the fitting of the F8BT cavities. For each exciton, the value of ωj was set to be at the energy that corresponds to the mid-point of the integral oscillator strength for the corresponding optical transition using(cid:82) ωj (cid:82) Emax Emin (ω)dω = 1 2 Emin (ω)dω, where (ω) is the extinction coefficient for Xj in the Emin to Emax energy range. Acknowledgement The authors thank Prof. Moritz Riede for access to his facilities and Dr Richard Hamilton for fruitful discussions. They also acknowledge funding from the University of Oxford, from the UK Engineering and Physical Sciences Research Council and the Jiangsu Industrial Tech- nology Research Institute. F.L.R. further thanks Wolfson College and Dr Simon Harrison for the award of a Wolfson Harrison UK Research Council Physics Scholarship. References (1) Weisbuch, C.; Nishioka, M.; Ishikawa, A.; Arakawa, Y. Observation of the coupled exciton-photon mode splitting in a semiconductor quantum microcavity. Physical Re- view Letters 1992, 69, 3314 -- 3317. (2) Marks, R. N.; Halls, J. J. M.; Bradley, D. D. C.; Friend, R. H.; Holmes, A. B. The pho- tovoltaic response in poly(p-phenylene vinylene) thin-film devices. Journal of Physics: Condensed Matter 1994, 6, 1379 -- 1394. (3) Alvarado, S.; Seidler, P.; Lidzey, D. G.; Bradley, D. D. C. Direct Determination of the Exciton Binding Energy of Conjugated Polymers Using a Scanning Tunneling Micro- scope. Physical Review Letters 1998, 81, 1082 -- 1085. (4) Plumhof, J. D.; Stoferle, T.; Mai, L.; Scherf, U.; Mahrt, R. F. Room-temperature 25 BoseEinstein condensation of cavity excitonpolaritons in a polymer. Nature Materials 2014, 13, 247 -- 252. (5) Daskalakis, K. S.; Maier, S. A.; Murray, R.; K´ena-Cohen, S. Nonlinear interactions in an organic polariton condensate. Nature Materials 2014, 13, 271 -- 278. (6) Lerario, G.; Ballarini, D.; Fieramosca, A.; Cannavale, A.; Genco, A.; Mangione, F.; Gambino, S.; Dominici, L.; De Giorgi, M.; Gigli, G.; Sanvitto, D. High-speed flow of interacting organic polaritons. Light: Science & Applications 2017, 6, e16212 -- e16212. (7) Zasedatelev, A. V.; Baranikov, A. V.; Urbonas, D.; Scafirimuto, F.; Scherf, U.; Stoferle, T.; Mahrt, R. F.; Lagoudakis, P. G. A room-temperature organic polariton transistor. Nature Photonics 2019, 13, 378 -- 383. (8) Lidzey, D. G.; Bradley, D. D. C.; Skolnick, M. S.; Virgili, T.; Walker, S.; Whit- taker, D. M. Strong excitonphoton coupling in an organic semiconductor microcavity. Nature 1998, 395, 53 -- 55. (9) Schwartz, T.; Hutchison, J. A.; Genet, C.; Ebbesen, T. W. Reversible Switching of Ultrastrong Light-Molecule Coupling. Physical Review Letters 2011, 106, 196405. (10) K´ena-Cohen, S.; Maier, S. A.; Bradley, D. D. C. Ultrastrongly Coupled Exciton- Polaritons in Metal-Clad Organic Semiconductor Microcavities. Advanced Optical Ma- terials 2013, 1, 827 -- 833. (11) Gambino, S.; Mazzeo, M.; Genco, A.; Di Stefano, O.; Savasta, S.; Patan`e, S.; Bal- larini, D.; Mangione, F.; Lerario, G.; Sanvitto, D.; Gigli, G. Exploring LightMatter Interaction Phenomena under Ultrastrong Coupling Regime. ACS Photonics 2014, 1, 1042 -- 1048. (12) Mazzeo, M.; Genco, A.; Gambino, S.; Ballarini, D.; Mangione, F.; Di Stefano, O.; Patan`e, S.; Savasta, S.; Sanvitto, D.; Gigli, G. Ultrastrong light-matter coupling in 26 electrically doped microcavity organic light emitting diodes. Applied Physics Letters 2014, 104, 233303. (13) Suzuki, M.; Nishiyama, K.; Kani, N.; Yu, X.; Uzumi, K.; Funahashi, M.; Shimokawa, F.; Nakanishi, S.; Tsurumachi, N. Observation of ultrastrong-coupling regime in the Fab- ryP´erot microcavities made of metal mirrors containing Lemke dye. Applied Physics Letters 2019, 114, 191108. (14) Liu, B.; Rai, P.; Grezmak, J.; Twieg, R. J.; Singer, K. D. Coupling of exciton-polaritons in low Q coupled microcavities beyond the rotating wave approximation. Physical Review B 2015, 92, 155301. (15) Liu, B.; Crescimanno, M.; Twieg, R. J.; Singer, K. D. Dispersion of ThirdHarmonic Generation in Organic Cavity Polaritons. Advanced Optical Materials 2019, 1801682. (16) Frisk Kockum, A.; Miranowicz, A.; De Liberato, S.; Savasta, S.; Nori, F. Ultrastrong coupling between light and matter. Nature Reviews Physics 2019, 1, 19 -- 40. (17) Forn-D´ıaz, P.; Lamata, L.; Rico, E.; Kono, J.; Solano, E. Ultrastrong coupling regimes of light-matter interaction. Reviews of Modern Physics 2019, 91, 025005. (18) Askenazi, B.; Vasanelli, A.; Todorov, Y.; Sakat, E.; Greffet, J.-J.; Beaudoin, G.; Sagnes, I.; Sirtori, C. Midinfrared Ultrastrong LightMatter Coupling for THz Ther- mal Emission. ACS Photonics 2017, 4, 2550 -- 2555. (19) Yoshihara, F.; Fuse, T.; Ao, Z.; Ashhab, S.; Kakuyanagi, K.; Saito, S.; Aoki, T.; Koshino, K.; Semba, K. Inversion of Qubit Energy Levels in Qubit-Oscillator Circuits in the Deep-Strong-Coupling Regime. Physical Review Letters 2018, 120, 183601. (20) Bayer, A.; Pozimski, M.; Schambeck, S.; Schuh, D.; Huber, R.; Bougeard, D.; Lange, C. Terahertz LightMatter Interaction beyond Unity Coupling Strength. Nano Letters 2017, 17, 6340 -- 6344. 27 (21) Benz, F.; Schmidt, M. K.; Dreismann, A.; Chikkaraddy, R.; Zhang, Y.; Demetri- adou, A.; Carnegie, C.; Ohadi, H.; de Nijs, B.; Esteban, R.; Aizpurua, J.; Baum- berg, J. J. Single-molecule optomechanics in "picocavities". Science (New York, N.Y.) 2016, 354, 726 -- 729. (22) Ciuti, C.; Bastard, G.; Carusotto, I. Quantum vacuum properties of the intersubband cavity polariton field. Physical Review B 2005, 72, 115303. (23) George, J.; Wang, S.; Chervy, T.; Canaguier-Durand, A.; Schaeffer, G.; Lehn, J.-M.; Hutchison, J. A.; Genet, C.; Ebbesen, T. W. Ultra-strong coupling of molecular mate- rials: spectroscopy and dynamics. Faraday Discussions 2015, 178, 281 -- 294. (24) Tropf, L.; Gather, M. C. Investigating the Onset of the Strong Coupling Regime by Fine-Tuning the Rabi Splitting in Multilayer Organic Microcavities. Advanced Optical Materials 2018, 6, 1800203. (25) Barachati, F.; Simon, J.; Getmanenko, Y. A.; Barlow, S.; Marder, S. R.; K´ena- Cohen, S. Tunable Third-Harmonic Generation from Polaritons in the Ultrastrong Cou- pling Regime. ACS Photonics 2018, 5, 119 -- 125. (26) Campoy-Quiles, M.; Ferenczi, T.; Agostinelli, T.; Etchegoin, P. G.; Kim, Y.; Anthopou- los, T. D.; Stavrinou, P. N.; Bradley, D. D. C.; Nelson, J. Morphology evolution via self-organization and lateral and vertical diffusion in polymer:fullerene solar cell blends. Nature Materials 2008, 7, 158 -- 164. (27) Le Roux, F.; Bradley, D. D. C. Conformational control of exciton-polariton physics in metal-poly(9,9-dioctylfluorene)-metal cavities. Physical Review B 2018, 98, 195306. (28) Perevedentsev, A.; Chander, N.; Kim, J.-S.; Bradley, D. D. C. Spectroscopic properties of poly(9,9-dioctylfluorene) thin films possessing varied fractions of β-phase chain seg- ments: enhanced photoluminescence efficiency via conformation structuring. Journal of Polymer Science Part B: Polymer Physics 2016, 54, 1995 -- 2006. 28 (29) Virgili, T.; Lidzey, D.; Grell, M.; Walker, S.; Asimakis, A.; Bradley, D. Completely polarized photoluminescence emission from a microcavity containing an aligned conju- gated polymer. Chemical Physics Letters 2001, 341, 219 -- 224. (30) Campoy-Quiles, M.; Etchegoin, P. G.; Bradley, D. D. C. On the optical anisotropy of conjugated polymer thin films. Physical Review B 2005, 72, 045209. (31) Hertzog, M.; Rudquist, P.; Hutchison, J. A.; George, J.; Ebbesen, T. W.; Borjesson, K. Voltage-Controlled Switching of Strong Light-Matter Interactions using Liquid Crys- tals. Chemistry - A European Journal 2017, 23, 18166 -- 18170. (32) Gao, W.; Li, X.; Bamba, M.; Kono, J. Continuous transition between weak and ultra- strong coupling through exceptional points in carbon nanotube microcavity excitonpo- laritons. Nature Photonics 2018, 12, 362 -- 367. (33) Grell, M.; Bradley, D. D. C.; Inbasekaran, M.; Woo, E. P. A glass-forming conjugated main-chain liquid crystal polymer for polarized electroluminescence applications. Ad- vanced Materials 1997, 9, 798 -- 802. (34) Grell, M.; Redecker, M.; Whitehead, K. S.; Bradley, D. D. C.; Inbasekaran, M.; Woo, E. P. Monodomain alignment of thermotropic fluorene copolymers. Liquid Crys- tals 1999, 26, 1403 -- 1407. (35) Whitehead, K. S.; Grell, M.; Bradley, D. D. C.; Jandke, M.; Strohriegl, P. Highly polarized blue electroluminescence from homogeneously aligned films of poly(9,9- dioctylfluorene). Applied Physics Letters 2000, 76, 2946 -- 2948. (36) Dyreklev, P.; Berggren, M.; Inganas, O.; Andersson, M. R.; Wennerstrom, O.; Hjert- berg, T. Polarized electroluminescence from an oriented substituted polythiophene in a light emitting diode. Advanced Materials 1995, 7, 43 -- 45. 29 (37) Cimrov´a, V.; Remmers, M.; Neher, D.; Wegner, G. Polarized light emission from LEDs prepared by the Langmuir-Blodgett technique. Advanced Materials 1996, 8, 146 -- 149. (38) Hagler, T. W.; Pakbaz, K.; Voss, K. F.; Heeger, A. J. Enhanced order and electronic de- localization in conjugated polymers oriented by gel processing in polyethylene. Physical Review B 1991, 44, 8652 -- 8666. (39) Zheng, Z.; Yim, K.; Saifullah, M.; Welland, M. E.; Friend, R. H.; Kim, J.-S.; S., H. W. T. Uniaxial Alignment of Liquid-Crystalline Conjugated Polymers by Nanoconfinement. 2007, (40) Whitehead, K.; Grell, M.; Bradley, D.; Inbasekaran, M.; Woo, E. Polarized emission from liquid crystal polymers. Synthetic Metals 2000, 111-112, 181 -- 185. (41) Schmid, S. A.; Yim, K. H.; Chang, M. H.; Zheng, Z.; Huck, W. T. S.; Friend, R. H.; Kim, J. S.; Herz, L. M. Polarization anisotropy dynamics for thin films of a conjugated polymer aligned by nanoimprinting. Physical Review B 2008, 77, 115338. (42) Lee, D.-M.; Lee, Y.-J.; Kim, J.-H.; Yu, C.-J. Birefringence-dependent linearly-polarized emission in a liquid crystalline organic light emitting polymer. Optics Express 2017, 25, 3737. (43) Misaki, M.; Chikamatsu, M.; Yoshida, Y.; Azumi, R.; Tanigaki, N.; Yase, K.; Naga- matsu, S.; Ueda, Y. Highly efficient polarized polymer light-emitting diodes utilizing oriented films of β-phase poly(9,9-dioctylfluorene). Applied Physics Letters 2008, 93, 023304. (44) Sirringhaus, H.; Wilson, R. J.; Friend, R. H.; Inbasekaran, M.; Wu, W.; Woo, E. P.; Grell, M.; Bradley, D. D. C. Mobility enhancement in conjugated polymer field-effect transistors through chain alignment in a liquid-crystalline phase. Applied Physics Let- ters 2000, 77, 406 -- 408. 30 (45) Seki, T. New strategies and implications for the photoalignment of liquid crystalline polymers. Polymer Journal 2014, 46, 751 -- 768. (46) Li, X.; Kozenkov, V. M.; Yeung, F. S.-Y.; Xu, P.; Chigrinov, V. G.; Kwok, H.-S. Liquid- Crystal Photoalignment by Super Thin Azo Dye Layer. Japanese Journal of Applied Physics 2006, 45, 203 -- 205. (47) Ma, L.-L.; Li, S.-S.; Li, W.-S.; Ji, W.; Luo, B.; Zheng, Z.-G.; Cai, Z.-P.; Chigrinov, V.; Lu, Y.-Q.; Hu, W.; Chen, L.-J. Rationally Designed Dynamic Superstructures Enabled by Photoaligning Cholesteric Liquid Crystals. Advanced Optical Materials 2015, 3, 1691 -- 1696. (48) Zhang, H.; Ma, L.; Zhang, Q.; Shi, Y.; Fang, Y.; Bradley, D. D. C. In Preparation 2019, (49) Ma, L.-L.; Tang, M.-J.; Hu, W.; Cui, Z.-Q.; Ge, S.-J.; Chen, P.; Chen, L.-J.; Qian, H.; Chi, L.-F.; Lu, Y.-Q. Smectic Layer Origami via Preprogrammed Photoalignment. Ad- vanced Materials 2017, 29, 1606671. (50) Sanvitto, D.; K´ena-Cohen, S. The road towards polaritonic devices. Nature Materials 2016, 15, 1061 -- 1073. (51) Solnyshkov, D.; Bleu, O.; Malpuech, G. All optical controlled-NOT gate based on an excitonpolariton circuit. Superlattices and Microstructures 2015, 83, 466 -- 475. (52) Espinosa-Ortega, T.; Liew, T. C. H. Complete architecture of integrated photonic cir- cuits based on and and not logic gates of exciton polaritons in semiconductor micro- cavities. Physical Review B 2013, 87, 195305. (53) Campoy-Quiles, M.; Heliotis, G.; Xia, R.; Ariu, M.; Pintani, M.; Etchegoin, P.; Bradley, D. D. C. Ellipsometric Characterization of the Optical Constants of Polyfluo- rene Gain Media. Advanced Functional Materials 2005, 15, 925 -- 933. 31 (54) Campoy-Quiles, M.; Alonso, M. I.; Bradley, D. D. C.; Richter, L. J. Advanced Ellipso- metric Characterization of Conjugated Polymer Films. Advanced Functional Materials 2014, 24, 2116 -- 2134. (55) Campoy-Quiles, M.; Etchegoin, P. G.; Bradley, D. D. C. On the optical anisotropy of conjugated polymer thin films. Physical Review B 2005, 72, 045209. (56) Tropf, L.; Dietrich, C. P.; Herbst, S.; Kanibolotsky, A. L.; Skabara, P. J.; Wurthner, F.; Samuel, I. D. W.; Gather, M. C.; Hofling, S. Influence of optical material properties on strong coupling in organic semiconductor based microcavities. Applied Physics Letters 2017, 110, 153302. (57) Rothe, C.; Galbrecht, F.; Scherf, U.; Monkman, A. The β-Phase of Poly(9,9- dioctylfluorene) as a Potential System for Electrically Pumped Organic Lasing. Ad- vanced Materials 2006, 18, 2137 -- 2140. (58) Valyukh, I.; Arwin, H.; Chigrinov, V.; Valyukh, S. UV-induced in-plane anisotropy in layers of mixture of the azo-dyes SD-1/SDA-2 characterized by spectroscopic ellipsom- etry. physica status solidi (c) 2008, 5, 1274 -- 1277. (59) Agranovich, V. M. Opt Spektrosk. 1957, 2 . (60) Hopfield, J. J. Theory of the Contribution of Excitons to the Complex Dielectric Con- stant of Crystals. Physical Review 1958, 112, 1555 -- 1567. (61) Economou, E. N. Surface Plasmons in Thin Films. Physical Review 1969, 182, 539 -- 554. (62) Litinskaya, M.; Agranovich, V. M. Polariton trap in microcavities with metallic mirrors. Journal of Physics: Condensed Matter 2012, 24, 015302. (63) Perevedentsev, A.; Sonnefraud, Y.; Belton, C. R.; Sharma, S.; Cass, A. E. G.; Maier, S. A.; Kim, J.-S.; Stavrinou, P. N.; Bradley, D. D. C. Dip-pen patterning 32 of poly(9,9-dioctylfluorene) chain-conformation-based nano-photonic elements. Nature Communications 2015, 6, 5977. (64) Scafirimuto, F.; Urbonas, D.; Scherf, U.; Mahrt, R. F.; Stoferle, T. Room-Temperature Exciton-Polariton Condensation in a Tunable Zero-Dimensional Microcavity. ACS Pho- tonics 2018, 5, 85 -- 89. (65) Rajendran, S. K.; Wei, M.; Ohadi, H.; Ruseckas, A.; Turnbull, G. A.; Samuel, I. D. W. Low Threshold Polariton Lasing from a SolutionProcessed Organic Semiconductor in a Planar Microcavity. Advanced Optical Materials 2019, 7, 1801791. 33
1803.04655	1	1803	2018-03-13T06:49:13	About the Problem of Utilization the Low-Potential Heat and Recent Perspective Developments	[ "physics.app-ph" ]	Problem of utilization of the low-potential heat and its perspectives are considered and one example from the author works is presented in detail. Description of the project and the results obtained are presented by a development of the Liquid Metal Magnetohydrodynamic (LMMHD) Gravitational Mini Power Plant (GMPP) for utilization of the low-potential heat from any available low-exergy sources, which are the huge sources of the wasted energy around the globe. Project is based on the experience by the LMMHD energy transformation with gravitational vapour/gas-lift driving principle. An example of the GMPP was developed and designed for small local consumers (100kW-1MW) for the application in the geothermal low-temperature sources 150-250C, big ferries (unused hot waters from engine), hot waters and gases from metallurgical and chemical factories, and many other similar customers. The drawings for the construction have been made and the optimal parameters computed for some potential liquid metal working media and two variants of the unit modules assembling have been elaborated (parallel assembling for getting the desired voltage or consecutive assembling for obtaining the desired current in the electrical network). The optimal height of the liquid metal circulating loop was obtained in the range 10-15 m, and the voltage in a unit 1-2 Volt.	physics.app-ph	physics	About the Problem of Utilization the Low-Potential Heat and Recent Perspective Developments Ivan V. Kazachkov 1,2 1 Dept of Information technology and data analysis, Nizhyn Gogol state university, Grafs'ka, 2, Ukraine, 16600, Tel: +380973191343;Email: [email protected] 2 Dept of Energy Technology, Royal Institute of Technology, Sweden Abstract Problem of utilization of the low-potential heat and its perspectives are considered and one example from the author's works is presented in detail. Description of the project and the results obtained are presented by a development of the Liquid Metal Magnetohydrodynamic (LMMHD) Gravitational Mini Power Plant (GMPP) for utilization of the low-potential heat from any available low-exergy sources, which are the huge sources of the wasted energy around the globe. Project is based on the experience by the LMMHD energy transformation with gravitational vapour/gas-lift driving principle. An example of the GMPP was developed and designed for small local consumers (100kW-1MW) for the application in the geothermal low-temperature sources 150- 250C, big ferries (unused hot waters from engine), hot waters and gases from metallurgical and chemical factories, and many other similar customers. The drawings for the construction have been made and the optimal parameters computed for some potential liquid metal working media and two variants of the unit modules' assembling have been elaborated (parallel assembling for getting the desired voltage or consecutive assembling for obtaining the desired current in the electrical network). The optimal height of the liquid metal circulating loop was obtained in the range 10-15 m, and the voltage in a unit 1.2-1.5 Volt. Keywords: Liquid Metal; MHD; Circulating Loop; Gravitational Vapor Lift; Faraday Generator; Standard Unit; Voltage and Current; Parallel and Consecutive Assembling. 1. Accomplishments and challenges in utilization of low-potential heat sources The rapid development of modern earthly civilizations requires ever greater resources of energy that humanity is constantly looking for. You need large reserves of clean energy, and it is desirable as cheap as possible. So, recently, for example, the topic of extracting huge reserves of energy and fresh water due to the movement of atmospheric air is rapidly developing in many countries around the globe [l-10]. The Swedish company United Science and Capital (US&C) [9] has recently signed an agreement with the Arizona State University about common promotion of the energy tower proposed by company. Also US&C is now in an advanced stage of negotiations with Governments of many countries with issues in water and food security, as well as for profit and non-profit organizations. People have been looking for the sources of energy since ancient times when recognized their advantages from using the energy of the sun and fire, later on using the energy of waterfalls, wind and other sources of energy. Viktor Schauberger nearly hundred years ago voted to the mankind with his clear statement to use the implosion instead of explosion, burning the mineral resources of the planet (V. Schauberger: Unsere sinnlose Arbeitet - Die Quelle der Weltkrise, Der Aufbau durch Atomverwandlung, nicht Atomzertrümmerung (1933, Krystall-Verlag GmbH, 2001, Jörg Schauberger, released in English as "Our Senseless Toil - The Cause of the World Crisis - Progress Through Transformation of the Atom - Not its destruction!"). Nowadays there are known many attempts for creation and using the clean energy of the diverse renewable sources instead of burning the coal, oil, etc. destroying and exhausting our beautiful planet. One of the modern directions in a development and implementation of the new technologies and devices is cogeneration and a use of the low-potential energy sources, which have a low exergy and as such are not 1 suitable for a profitable intensive production of energy. This paper deals with one of such directions connected to utilization of the low-potential energy [l1-21], which is presently wasted in a global scale around the world due to the engineering and technical problems of its efficient use. Liquid Metal MagnetoHydroDynamic Energy Converters (LMMHD EC) of gravity type have been recently proposed for a variety of the heat sources for electrical power generation [l1]. In these systems, the vertical two- phase flows consisting of a steam and the high density liquid metals like a lead alloy take place in the riser pipe. The design of an optimum LMMHD EC and a scaling up of the system requires an accurate modeling of the two-phase flows. The most important parameter which governs the two-phase flow is a void fraction. Such approaches for a modeling of the multiphase flows including our works have been developed and proven in the experimental studies [22-35]. The accuracy and the range of applicability of these relate to be verified for liquid metal two- phase flows. The experiments have been conducted in a nitrogen-mercury simulation LMMHD EC facility. The time averaged void fraction was measured for various flow rates using gamma-ray attenuation method. Corrections for the cross-sectional variations and dynamic fluctuations in the void fraction as well as the errors due to finite beam size of the gamma ray have been applied to improve the accuracy of measurement. Measured void fraction and pressure profiles were compared with predicted values based on the well known empirical relations for a void fraction. In addition, the data was compared with the values based on bubble/slug flow models, which take into consideration the momentum equation of the gas explicitly. The results of the analysis have been presented in the paper [13]. The steam in the LMMHD ETGAR (gravitational) system was initially generated by a conventional boiler and then injected through the mixer to the hot lead [12]. In order to further reduce capital and operation costs of the ETGAR systems, it was proposed to inject the thermodynamic fluid in its liquid phase (instead of steam) into the hot liquid metal in the "riser" branch of the loop. The boiling of the volatile liquid in this case occurs under direct contact with the liquid metal, and thus avoids the need for an expensive external steam generator. It is also anticipated that (the boiling process leads to mixing of the two-phase flow and hence a decrease of the slip between vapor bubbles and liquid metal (higher loop efficiency). The applicability of this idea was tested in an integrated experimental system the "OFRA" system described in [12]. It simulated a real LMMHD ETGAR loop and was designed to operate with lead and eutectic lead/bismuth, at temperatures up to 480°C. The following goals of study were stated: direct contact boiling phenomena of a multi-droplet bed of water in lead; Influence of different additives (surfactants, foaming materials etc.), dissolved in the lead, on the two-phase flow characteristics, namely, void fraction distribution and phase velocity ratio, flow configuration and overall system performances; water reaction with molten lead over a wide range of temperatures, pressures and flow rates; lead reaction with the construction materials in the presence of liquid/vapor water. Our project [14] started from these results. The development of superconducting installations, such as mini-power plants (MPPs), and their justification is a new trend in the energy applications of superconductivity [18], which can enhance the advantages of MPP. For development, construction and testing, the author [18] developed a high-temperature superconducting (HTS) MPP model with a design capacity of 10 kW. Appropriate approaches and tools are needed to promote the industrial application and introduction of HTS power devices on the market, including HTS MPP equipment. Methods of solving problems of automated design effectively help qualitatively and quantitatively evaluate the technical and economic capabilities of HTS devices. The Qena Paper Industry Company [19] prepared analysis of exergy for various components of a 45 MW cogeneration thermal power plant. Each component was analyzed separately, and exergy losses were identified and quantified for each component of the cogeneration power plant. In addition, the effect of changing the ambient temperature on this analysis was investigated. The results showed that the rate of destruction of exergy increases with increasing ambient temperature in all components of the system. The main contribution to the destruction of exergy occurs from the boiler, and it increases with increasing ambient temperature. With a full load and an ambient temperature of 18 to 45 °C, the percentage of destruction for the boiler, deaerator and condenser was 78.9% to 80.5% and from 3.6% to 3.3% and 3.1% up to 0,6% 2 respectively. The cogeneration power plant is an efficient, clean and reliable approach to the generation of electricity, as well as the supply of thermal energy through a single fuel. In particular, cogeneration, which sometimes uses the waste heat through its processing, achieves a significant improvement in system efficiency. The main advantages of a cogeneration are relatively high efficiency, reducing emissions to the atmosphere, such as nitric oxide (N2O), sulfur dioxide (SO2) and carbon dioxide (CO2). Today, about 80% of the world's electricity is produced approximately from fossil fuels (coal, oil, mazut, natural gas) from thermal power plants, while 20% of electricity is compensated from various sources such as hydraulic, nuclear, wind, solar, geothermal and biogas. Energy consumption is one of the most important indicators, demonstrates the stages of development of countries and the living standards of communities. Population growth, urbanization, industrialization and technological development lead to increased energy consumption, which leads to decisive environmental problems, such as pollution and the greenhouse effect. In recent decades, the exergy index based on the second law of thermodynamics proved to be a useful method in the development, evaluation, optimization and improvement of thermal power plants. Kamate and Gangavati conducted an energy and exergy analysis in a cogeneration plant based at Belgaum in India [19]. ORMAT [17] developed the geothermal recovery mini-power plants based on the organic Rankine cycle for the moderate temperatures (85–150 °C) and some of them were done for the higher temperatures. The power of these MPPs was in a range of 300–1300 kW. These factory-integrated and tested modules of the two or more units can be combined for applications, where the geothermal or industrial waste heat source is economically sufficient to install the larger power plants. Experience has been acquired in operation of the low-enthalpy geothermal power plants in the USA, as well as in other countries. A few typical power plants of the power 800 kW, 3.2 and 30 MW have been presented in [17]. 2. Liquid Metal Magnetohydrodynamic Gravitational Mini Power Plant Our paper is devoted to a description of the project and the results obtained as concern to a development of the LMMHD GMPP for utilization of the low-potential heat from any available low-exergy sources, which are the huge sources of the wasted energy around the globe. The project was based on the gravitational vapour/gas-lift driving principle and gained experience by the LMMHD energy transformation with. An example of the GMPP was developed and designed for small local consumers (100kW-1MW) for the application in the geothermal low-temperature sources 150-250 °C, big ferries (unused hot waters from engine), hot waters and gases from metallurgical and chemical factories, and many other similar customers. The drawings for the construction have been made and the optimal parameters computed for some potential liquid metal working media and two variants of the unit modules' assembling have been elaborated (parallel assembling for getting the desired voltage or consecutive assembling for obtaining the desired current in the electrical network). The optimal height of the liquid metal circulating loop was obtained in the range 10-15 m, and the voltage in a unit 1.2-1.5 Volt. The project was based on the experience by LMMHD energy transformation with gravitational vapour/gas-lift driving principle and the authors' experience in the development of MPP for utilization of low- potential heat performed in Ukraine. The MPP for utilization of low-potential heat was developed with the goal to reduce an energy production cost at the power plants and to utilize the low-temperature heat sources in any available industrial application, e.g. geothermal heat sources, hot waters and gases from metallurgical and chemical factories, big ferries, and so on, where the heat sources with temperature above around 150 0C are available for use. The structure and functionality of the basic components of MPP can be explained according to the simple schematic representation given in Fig.1. The MPP consists of a circulatory loop made by channels (tubes), two vertical and two horizontal. The diameter of the channels is about 10-20 cm depending on the power and other parameters. The two vertical channels of the circulatory loop are the lifting and falling ones (the left and right one, correspondingly), where the liquid metal (working fluid) is circulating. The bottom horizontal channel of 3 the circulatory loop contains the heater, which accepts the utilized low-potential heat from the heat source (may be liquid or gas, which passes through the heater). The left vertical lifting channel is driving part of the loop. A two-phase (vapour/gas-liquid) flow is going up, while in falling channel the separated pure working fluid is moving down through the Faraday MHD-generator producing the direct current due to electroconductive fluid (liquid metal) passing through the MHD generator. The two vertical channels are connected at the top of an installation with a separator. At the bottom they are connected with a heater utilizing the low-potential heat. Fig.1. The simplified structural scheme of the MPP. Circulatory liquid metal (LM) loop is created for a continuous circulation of the working LM due to a difference of the densities in the left and right vertical channels and due to a vaporization of the injected water from the bottom of the left channel. The mixer installed at the bottom of the left channel serves for injection of the thermodynamic fluid (saturated water) into a working fluid in the loop. The heater delivers a heat to the working fluid. Upper part of the channel is connected to a separator where the vapor is separated and supplied to the condensation and cleaning system. Separated working fluid is going to the right vertical channel. Channel of LMMHD-generator contains the confuser, straight forward channel and the diffuser (analogy of the Laval nozzle) for speeding-up the working fluid in the MHD-generator. The Faraday-type induction LMMHD-generator is used, with electric commutation (parallel or consecutive) of the separate autonomous unified modules into electrical energy system. The last one is very useful for practice as far as it allows assembling of the MPP of the desired power and output electric energy from the unified modules. 4 The facility can work autonomously as well as together with another similar devices having consecutive or parallel electrical commutation of their MHD-generators. The electric facility can be equipped with the transformer of a direct current to an alternating current. The estimated parameters of MPP are as follows:  The initial temperature of a heating fluid (water, gas, vapour) in a circulatory loop is 120-180 °C  The upper temperature of the cycle – 150-250 °C  Lower temperature of the cycle – 65 °C  Vapor pressure – 5 bar  Power of the Plant – 0.1-1.0 MW  Height of the channels, 7-20 m  Distance between the channels of MPP, more than 0.5 m By preliminary estimations from our construction and calculation work, the LMMHD-generator has to have the following parameters:  Length of the electrodes – 0.5 m  The magnetic field – 0.1-1.0 T  The voltage – 50-100 V (can be regulated by a stated number of the assembled unified modules and the type of an electric commutation applied: parallel or consecutive)  The capacity, current and voltage are regulated by a desire of the consumer under assembling of the MPP through a number of the assembled unified modules and the type of their commutation chosen. 3. Modeling of Thermal Hydraulic Processes in MPP The project aimed on increasing the energy efficiency through utilization of the low-potential heat sources, which are presently in a wide amount not in use or are used partially and in an inefficient way. The MPP was be developed and designed for small local consumers (100kW-1MW) in the areas, where any low- potential heat sources are available, for example, geothermal hot water with temperatures about 120-180 0C, big ferries (unused hot waters from engine), hot waters and gases from metallurgical and chemical factories, etc. First of all, the investigation of the processes was performed, with the calculation and optimization of the MPP' parameters, e.g. geometry (the height and diameter of the channels, size of the circulatory loop, the thermal hydraulic parameters of a liquid metal flow in the loop, heat transfer processes in the heater and in the loop, vaporization of injected water, etc. The parameters of the LMMHD-generator were computer modeled and optimized. Altogether parameters of the MPP have been simulated and optimized afterwards. One of the most important sections of MPP is the driving vapor/gas-lift flow in the left vertical channel. Preliminary investigation has shown that this part may play a key role as the MPP's efficiency is changing depending on the evaporation regime and the liquid-bubble lifting flow regime. Injected water is heated in the hot liquid metal inside the channel and evaporates. This leads to a vapor expansion and a vapor lifting force in the left channel, which strongly depends on the size and distribution of the vapor bubbles in the flow. In different conditions the efficiency of MPP may change dramatically, e.g. ten times and more. To optimize investigation of the processes I a heterogeneous two-phase flow of the liquid metal with vapor bubbles the methods developed and elaborated in [22, 28-32] seem to be the most suitable. According to the method of heterogeneous multiphase flows developed by Prof. A.I. Nakorchevski [22] the characteristics of a mixture (mass, velocity, impulse, etc.) of the corresponding characteristics of different phases in a multiphase flow are expressed in next form , (1) where are co-called function-indicators of the phases in multiphase mixture determined as 5 ()lat()liat1()()()mlliiiatBtat()iBt. (2) With this approach, the analog of the Navier-Stokes equations in a boundary layer approximation was derived: (3) where the sum by mute index i is taken by all phases i. In the stationary equations of incompressible liquids (3) written in a cylindrical coordinate system are: p- pressure, ρ- density, u,v- the longitudinal and transversal velocity components, - turbulent stress for a phase i. Index m belongs to the values at the axis of the flow (symmetry axis). All values are averaged on the given interval by time. The schematic representation of the flow of two immiscible liquids is given in Fig. 2 for example of turbulent two-phase jet of two immiscible liquids: Fig. 2 The structural scheme of the two-phase jet of immiscible liquids One liquid is going from the nozzle of the radius r0 with velocity u01 (the velocity profile is supposed simple uniform) in a surroundings occupied by other liquid (phase 2) being in a rest. The structure of a jet's mixing with surrounding liquid according to the Fig. 2 is simplified according to a traditional scheme [23]. First, the initial part of the length xi with the approximately linear boundaries for the conical surface (in cylindrical coordinates) of the internal potential core of a first phase and the external interface (conical too) are considered as the boundaries of the mixing area and the central potential core, correspondingly. The mixing turbulent zone between the above surfaces contains drops and fragments of the phases as far as 6 1,(cid:160) phase occupies the elementary volume V()0,(cid:160)if phase is outside the elementary volume ViifiBti1()()0,1,miiiiiiiiyBuyBvBxy1(),iiiiiiiimuudpBuvyBxydxyyiimmiscible liquids have behaviors like the separate phases, with their interfacial multiple surfaces interacting in all such locations (exchange of mass, impulse, energy among the phases). Close to the external interface there is a flow of a surrounding liquid with some drops of the first phase, while, in turn, close to the boundary of the potential core there is a flow of the first phase with the drops of a second phase. As far as we have mutually immiscible liquids, the mixing zone contains two phase mixture in a turbulent flow. After an initial part of the mixing zone when all first phase in a potential core is spent, the short transit area is preparing the ground part of the turbulent two-phase jet, where both phases are well mixed across all layer. Normally for description of multiphase flows the spatial averaging of the differential equations of mass, impulse and energy conservation is performed using the concept of volumetric phase content [24-26], which does not fit so well to experimental study of the separate phase movement in a mixture as the approach proposed in [22], where the special experimental technology and micro sensor for a measurements in two-phase flows has been developed as well. In [25, 26] fundamentals and analysis of the different methods for modeling of the multiphase systems can be found. Actually the known methods of multiphase methods are well connected including the one in [22], and the parameters averaged by time can be easily transformed to the ones by [24-27]. The external interface of the mixing zone is determined zero longitudinal velocity of the second phase and transversal velocity of the first phase (the second phase is sucked from immovable surrounding into the mixing zone). The function-indicator of the first phase B1(t) is zero at the external interface because the first phase is absent in surroundings. Similar, the function-indicator B2(t) is zero on the interface of the potential core, boundary of the first phase going from the nozzle. In a first approach, an influence of the mass, viscous and capillary forces is neglected. With account of the above-mentioned, the boundary conditions are stated as follows [22]: y=y0, ui= u0i, vi=0, τi=0, B1=1; y=y0+δ, ui= 0, vi=0, τi=0, B1=0. (4) The turbulent stress in the phase is stated by the "new" Prandtl's formula τi=ρi δumi , (5) where is the coefficient of turbulent mixing for i-th phase, δ is the width of the mixing layer. The polynomial approximations for the velocity profiles and other functions in the turbulent mixing zone have been obtained based on the boundary conditions (4) in the form: (6) , , , (7) , , , , 7 i/iuyi34101/143,uu234202/1683,uu(0)321110.5(1)(),6,0BBhxh(1)3422111430.5(1)(),12,6BBhxh(2)34523111101560.5(1)(),20,12BBhxh(3)345624111204536100.5(1)(),30,20BBhxh(4)34567251113510512670150.5(1)(),42,30BBhxh(5)345678261156210336280120210.5(1)(),56,42Bhxh(6)3456789271184378756840540189290.5(1)(),72,56Bhxhwhere h(x)= is an interesting function, which determines a transition of the piecewise continuous function-indicator to its next approximation, determined from the condition that the derivative by with respect to a point be equal to zero up to (n+1)-th and including order. Based on these approximations the integral correlations have been derived for the two-phase turbulent jet [22]: , , , . (8) The first equation in (8) was got integrating by y the mass conservation equation, the second and the third ones – integrating the impulse conservation for the total flow of a two-phase mixture for y=y0+δ and y=y, respectively. The polynomial approximations for the functions u2, B1 on a ground part of the jet keep the same but for the function u1 approximation and the integral correlations for the ground part of a jet are [22]: , (9) where the first is equation of the mass conservation for the first phase, the second and the third – the momentum conservation equations for the total and for the part of the cross section, respectively, according to the methodology [23]. And the momentum equation on the jet's axis (y=0) is used too: . (10) The mathematical model including the ordinary differential equations (8)-(10) by longitudinal coordinate x are implemented for analysis and numerical simulation on computer the basic features of the stationary turbulent two-phase jet of two immiscible liquids. The function-indicator B1 shows how much is a presence of the first phase in a selected point of mixing zone, which can be directly compared to an experimental data by two-phase sensor. Therefore a solution of the task may give both parameters of the flow together with their belonging to a particular phase. Examples of computer modeling and comparison of the results obtained against the experimental data for turbulent heterogeneous jets are given in Fig. 3 and Fig. 4 for different density ratio of the mixing liquids and available slip of phases. As shows the Fig. 3 this method reveals good correlation with experimental data. The intensity of the phase mixing is seen from the Fig. 4. The density ratio 8.0 can approximately correspond to vapor/liquid melt of the MPP accounting for available slip of the phases. After optimization of thermal hydraulic parameters, the geometry of MPP was done and then the parameters of the LMMHD-generator have been determined. This served as a basis for a design of the construction of MPP according to the specified needs of consumer. The construction was performed in such a way that the MPP was proposed to be always assembled from the unified modules, which allows assembling the required plant with account of the local consumer's needs in each specific case. Thus, the MPP supposed to be mobile-based and easily transported to a location of the consumer and then assembled locally [36]. 8 2210/B()1nB1122010011002uryBuyd12222210100111222002uryBuBuyd22'210101100000011jjjiiiijjdduuuyyBuyduBuyddxdx2*001jjjjjjuyBu121BB2311/132,muu21101002,Buydyur22221010012jjjjBudyuyr22112mjjmjmjjjjjmduBudxyBFig. 3 Parameters of turbulent two-phase jet against experimental data for oil-water immiscible liquids Fig. 4 The stream lines in a turbulent two-phase jet of the immiscible liquids 9 The research was carried out through the fundamental analysis and the numerical experiments in a wide range of the parameters of interest. The design has been done in a way to achieve the movable MPP construction, which is easily assembled from the unified modules in any required place according to the needs of consumer and specific low-potential heat source available there for utilization to produce electricity. The cost for the project was estimated at $1 mln. Each of the 8 stages of the project was estimated approximately at 3 months duration. Cost estimate was made after discussion of the proposal with the contractor based on the specific task stated. 4. The exergy and the new molecular thermodynamic theory by Eroshenko V.A. With regard to the exergy of heat available at a given temperature, we must indicate that the maximum possible conversion of heat to work or the content of exergy, depending on the temperature, depends on the temperature at which the heat is and the temperature at which heat can be used, the latter is ambient temperature. Carnot (1824) found the upper limit for the transformation of heat into work [37], which is known as Carnot's efficiency: η=1- TC / TH (absolute temperatures), where TH is a higher temperature, and TC is the lower temperature. Exergy exchange: B=Q(1-T0 /Ts),, where Ts is the temperature of the heat source, T0 is the ambient temperature. It is obvious that the cycle's efficiency is maximized by the maximum of TH and the minimum TC. In the 1870s, one of the most prominent scientists, J.W. Gibbs (1873) united Thermochemistry into a single theory [38], incorporating a new concept of chemical potential to evoke a shift away from chemical equilibrium into Carnot's work in describing thermal and mechanical equilibrium and their potential for change. The unified Gibbs theory has led to the functions of the state of thermodynamic potential, which describes the differences from the thermodynamic equilibrium. Gibbs received the mathematics of the available energy of the heat source in the form above. Since then, the physics that described exergy has not changed much. The actual amount of available energy depends on conditions that cannot be simulated. But for a real system, energy is saved, and exergy is energy available for useful work. As we mentioned above, there is a lot of energy around the world (solar, wind, hydropower, geothermal, industrial heat loss, etc.), but for various reasons it is inaccessible, just some of it. Thus, energy consists of two parts: exergy (an available part of energy) plus anergy (inaccessible part). For more details on the concepts of energy and exergy and the differences about them, shown in several cases considered with calculations, there are, for example, references to present the concept of exergy and its applications [39, 40]. Understanding the concept of exergy and its potential is very useful for conducting an exergy analysis of the real systems or processes. One of the most exciting theories on the end of XXth century is molecular thermodynamics founded by Professor of the NTUU "KPI" Valentyn Andriyovych Eroshenko [41-49] who revealed the huge potential of a surface energy as a new working medium instead of used for the centuries volumetric energy of a working bodies. Gas/vapor has been used as a working body of all heat engines, turbines, heat and power facilities, mechanical power batteries and other devices and thermo mechanical systems for two centuries in many branches of industry (energy, engineering, space and defense industry, etc.). All thermodynamic transformations are based on kinetic energy of a chaotic thermal movement of the first time in the world it was proposed gas/vapor molecules. Intermolecular potential energy is zero for an ideal gas, and negligible for real gases. For the Interfacial surfaces in heterogeneous lyophobic systems as the new working bodies in the thermo-mechanical devices and systems for energy accumulation, dissipation and conversion. In the new working bodies named repulsive clathrates (RK), the potential energy of Intermolecular interaction becomes predominant, while the kinetic energy of the molecules is less interesting as the lower one. Because the forces and Intermolecular energy potential depend on the temperature, the new direction of technical thermodynamics has been called "Thermomolecular energy". to use In terms of the technology repulsive clathrates are considered the condensed systems "liquid + capillary- porous matrix unmoisten to this liquid" (contact angle  >> 90 deg). Extensive parameter of the system 10 is interfacial surface , and intensive parameter - the surface tension of the liquid σ. The matrices are characterized by specific surface (from 200 to 1000 m2/g) and by radiuses of the capillaries and pores r (from 0.3 to 10 nm). As the porous matrices Silica gels, Silchromes Al2O3, porous glasses, zeolites and other carriers of a well developed specific surface are used. Porosity of is from 0.2 to 0.7 cm3/cm3. The fluids used are as follows: water and aqueous solutions, electrolytes, salts, low temperature eutectics and metal alloys. This new direction in the thermodynamics is the most promising in the reveling the huge potential of the surface energy and low-exergy sources of energy totally. the matrices applied All thermodynamic transformations are based on the thermomechanical processes of the interfacial surface  "extension-reduction" with the work W and heat q: W=σ cosd - the work by isothermal surface development (the Gibbs work); q=T(d/dT)cosd - heat of the isothermal surface (T- temperature of a process, dσ/dT is the temperature coefficient of a surface tension of a fluid, the physical constant). The specifics of the RK are that the pressure is a subject to the hydraulic Pascal law and is defined by the hydraulic capillary Laplace pressure. Changing the volume of the condensed heterogeneous system V is determined by the rate of filling the pore space, which is the rate of a development of the interfacial surface , and it is not determined by a deformation process. The total benefit from the creation and use of thermomolecular devices, for example motors invented by V.A. Eroshenko should be connected to a radical saving of reducing the heat and chemical pollution: the thermomolecular engine (TME) works from the external heat supply. This makes the TME similar to the Stirling engine, which one with the humankind's last hope is connected as for the solution of the environmental problems: "Stirling will save the world!". But the three versions of the TME patented by Eroshenko are substantially more effective and they are available for producing nowadays. fuel and construction materials, as well as the to a 5. The conclusions The Liquid Metal Magnetohydrodynamic (LMMHD) Gravitational Mini Power Plant (GMPP) for utilization of the low-potential heat from any available low-exergy sources presented in the paper may be recommended for practical implementation after its technological finalization for producers of the ones. The main its advantage is possibility for utilization of any sources of the low/potential heat and mobile module principle of the assembling. It may be applied around the world for use of the geothermal low-temperature sources (150-250 Celsius degree), big ferries (unused hot waters from engine), hot waters and gases from metallurgical and chemical factories, etc. The optimal liquid metal applied as a working media in the MPP requires some more investigation for optimization from a few points of view. Two variants of assembling of the unit modules are proposed for the customer (parallel assembling for getting the desired voltage or consecutive assembling for obtaining the desired current in the electrical network). The optimal height of the liquid metal circulating loop was obtained in the range 10-15 m, and the voltage in a unit 1.2-1.5 Volt (the total voltage can be got any desired by consecutive assembling of the units). References 1. US Patent 6,510,687 B1, Zaslavsky D., et al., Jan. 28, 2003. Renewable resource hydro/aero- power generation plant and method of generating hydro/aero- power. 2. US Patent 3,894,393 (1975-07-15) Carlson; Phillip R., Power generation through controlled convection (aeroelectric power generation). 3. Zaslavsky, Dan; Rami Guetta et al. (December 2001). Energy Towers for Producing Electricity and Desalinated Water without a Collector (435 KiB). Technion Israel, Israel - India Steering Committee. Retrieved on 2007-03-15. 4. Altman, Talia; Dan Zaslavsky, Rami Guetta and Gregor Czisch (May 2006). "Evaluation of the potential of electricity and desalinated water supply by using technology of "Energy Towers" for Australia, America and Africa". http://www.ecmwf.int/about/special_projects/czisch_enrgy-towers-global-potential/report_2006_extended.pdf. Retrieved on 2007-03-18. 11 5. Altmann, T.; Y. Carmel, R. Guetta, D. Zaslavsky, Y. Doytsher (June 2005). "Assessment of an "Energy Tower" potential in Australia using a mathematical model and GIS" (PDF). Solar Energy (Elsevier Ltd.) 78 (6): 799–808. doi:10.1016/j.solener.2004.08.025. http://envgis.technion.ac.il/publications/energy%20tower%20potential.pdf. Retrieved on 2007-03-12. 6. Czisch, Gregor (June 2005). "Evaluation of the global potential of energy towers". http://www.ecmwf.int/about/special_projects/czisch_enrgy-towers-global-potential/. Retrieved on 2007-03-13. 7. Czisch, Gregor (September 2001). "Aeroelectric Oasis System". Global Renewable Energy Potential, Approaches to its Use. http://www.iset.uni-kassel.de/abt/w3-w/folien/magdeb030901/folie_26.html. Retrieved on 2007-03-13. 8. Gutman, Per-Olof; Eran Horesh, Rami Guetta, Michael Borshchevsky (2003-04-29). "Control of the Aero-Electric Power Station - an exciting QFT application for the 21st century". International Journal of Robust and Nonlinear Control (John Wiley & Sons, Ltd.) 13 (7): 619–636. doi:10.1002/rnc.828. 9. Aqua Tower. http://exicon.website/uploads/editor/arwadex2017/speeches_and_presentations/002%20_Workshop- Kujtim-Hyseni.pdf, Kujtim Hyseni, et al. http://uniscap.se/power-aqua-tower/ . 10. Mills D (2004). "Advances in solar thermal electricity technology". Solar Energy 76 (1-3): 19–31. doi:10.1016/S0038-092X(03)00102-6. 11. Branover H., Liquid metal MHD research and development in Israel// Progress in Astronautics and Aeronautics, Vol. 148, pp. 209- 221,1993. 12. Lesin S., Breido A., Zilberman I., Branover H. Integrated experimental system for testing the ETGAR concept using direct contact boiling in liquid lead/ Proc. Int. Conf. MAHYD: Riga, Latvia, Jan 1995.- P. 134. 13. Satyamurthy P., Thiyagarajan T.K., Dixit N.S., Venkatramani N. Two-phase flow studies of mercury-nitrogen in a simulated liquid metal magnetohydrodynamic energy converters of gravity type/ Proc. Int. Conf. MAHYD: Riga, Latvia, Jan 1995.- P. 135. 14. Beloded V.D., Kazachkov I.V., Shilovitch I.L., Jakovlev V.S. The locally liquid-metal MHD geo-power station development design/ Proc. Int. Conf. MAHYD: Riga, Latvia, Jan 1995.- P. 131. 15. Kapooria R.K. Possible developments in energy conversion using liquid metal magnetohydrodynamics// Journal of Energy in Southern Africa.- 2009, Vol. 20, No 2, P. 17-25. 16. Chandra A., Pandey G.K. Conceptual design and application of liquid metal MHD system for hydrogen production/ The 15th Riga and 6th PAMIR Conference on Fundamental and Applied MHD Liquid metal technologies.- June 27 – July 1, 2005. - P. 295-298. 17. Bronicki Lucien. Electrical power from moderated temperature geothermal sources with modular mini-power plants.- December 1988.- Geothermics 17(1):83-92 18. Vajda Istvan. Conceptual design of an all superconducting mini power plant model/ Proc. Conf. Electronic Design, Test and Applications, 2002. 19. Hefny Ramadan, Harby Khaled, Maghrabie Hussein M, Attalla M.. Exergy Analysis of Cogeneration Power Plant in Paper Industry/ Proceedings of 4th Int. Conference on Energy Engineering.- Egypt, December 26-28, 2017. 20. Combined Heat and Power (CHP) Technical Potential in the United States: US Department of Energy, March 2016. 21. Patent US20140260218 A1: Combined heat and power (CHP) system by Jan Hubertus Deckers, Dejatech, 18 September 2014 (with detailed technical drawings of a modern CHP system: engine, heat exchanger, generator). Nakorchevski А.I. Heterogeneous turbulent jets. Кyiv: Naukova Dumka, 1980, 142 p. 22. 23. Ginevskii A.S. The theory of turbulent jets and traces. Integral calculation methods. Мoscow: Mashinostroenie, 1969, 400 p. 24. Nakoryakov V.E., Pokusaev B.G., Shreiber I.R. Spreading of the waves in gas- and vapor-liquid media. Novosibirsk: ITF, 1983, 238 p. (In Russian). 25. Nigmatulin R.I. Fundamentals of mechanics of heterogeneous media. Moscow: Nauka, 1978. - 336 с. (In Russian). 26. R.I. Nigmatulin and John C. Friedly. Dynamics of Multiphase Media. CRC Press: Volumes 1 & 2. Revised and Augmented Edition (v. 1 & v. 2), Oct 1, 1990, 878 p. 27. Nikolay Ivanov Kolev, Multiphase Flow Dynamics 1. 2015. Springer International Publishing. 840 p. (totally 5 volumes by different applications including thermo hydraulics in nuclear energy processes). 28. Nakorchevski А.I., Kazachkov I.V. Calculation of heterogeneous turbulent jet. In book: Systems of automation of continuous technological processes, Institute of Cybernetics of NASU, 1979, P. 68-79. 29. Kazachkov I.V. Der turbulente versenkte Strahl von zweier unmischbaren Fluessigkeiten. Beitrag. Inst. fuer Kybernetik der Akademie der Wissenschaften der Ukraine. Kiev, 1980. 20 p. 30. Kazachkov I.V. Mathematical modeling of heterogeneous turbulent jets in cylindrical chamber// Soviet automatic control, 1980, vol.13, jan.-feb., p. 1-6. 31. Kazachkov I.V. Investigation of a turbulent mixing and wall protecting by garnissage in the jet devices working on two immiscible fluids. Candidate of Physics and Mathematics Dissertation (PHD). Kiev T.G. Shevchenko State Univ. 1981. 138 p. (in Russian). 32. Kazachkov I.V., Paladino D. and Sehgal B.R. Ex-vessel coolability of a molten pool by coolant injection from submerged nozzles/ 9th Int. Conf. Nucl. Energy Devel. April 8-12, 2001. Nice, France. 33. Kazachkov I.V. and Konovalikhin M.J. A Model of a Steam Flow through the Volumetrically Heated Particle Bed// 12 Int. J. of Thermal Sciences.- 2002.- Vol.41, 1077-1087. 34. Nakorchevski А.I., Basok B.I. Hydrodynamics and heat transfer in heterogeneous systems and devices of pulsating type. Kyiv: Naukova Dumka, 2001, 348 p. 35. Computational Methods for Multiphase Flow. Andrea Prosperetti. Cambridge university press. 2011. 742 p. 36. Beloded V.D., Zinoviev P.P., Zabarny G.N., Kazachkov I.V., Shurchkov A.V. Prospects for the development of Kamchatka before 2000/ Vses. Conf. by Renewable Energy Sources: Yerevan.- 1985.- P. 59. 37. Carnot S. (1824). Réflexions sur la puissance motrice du feu sur les machines propres a developper cette puissance. (Reflections on the Motive Power of Fire and on Machines Fitted to Develop That Power. Translated and edited by R.H. Thurston 1890). Paris: Bachelier. Archived from the original on 2012-02-04. 38. Gibbs J.W. (1873). A method of geometrical representation of thermodynamic properties of substances by means of surfaces: reprinted in Gibbs, Collected Works, ed. W. R. Longley and R. G. Van Name (New York: Longmans, Green, 1931)// Transactions of the Connecticut Academy of Arts and Sciences.- 2: 382–404. 39. Gundersen T. An introduction to the concept of exergy and energy quality by Department of Energy and Process Engineering Norwegian University of Science and Technology Trondheim, Norway Version 4, March 2011. 40. Wall G. Exergetics.- Bucaramanga, 2009, 151 pp. 41. Eroshenko V.А. Maximum efficiency or maximum thermodynamic compactness of thermal engines?// Izv. AN SSSR. Energy and transport. – 1987. – № 2. – С. 125–133 (In Russian). 42. Patent 1,254,811 А USSR (1981–DSP) – RF (1993), F 03 G 7/00. The heterogeneous thermodynamic system, the method of Eroshenko transforming thermal energy into a mechanical one and the device for its implementation/ Presented 24.07.1981 (DSP); Published 30.09.1996, Bulletin № 3 (In Russian). 43. Patent 1380357 А2 USSR (1983– DSP) – FR (1993), F 03 G 7/00. the method of Eroshenko transforming thermal energy into a mechanical one/ Presented 30.11.1983 (DSP); Published 30.09.1996, Bulletin № 3 (In Russian). 44. Eroshenko V.А. Unusual properties of one complex thermodynamic system// Dokl. Academy of Sciences of Ukraine. Ser. А. Fiz.-mat. And eng. sciences. – 1990. – № 10. – P. 79–82 (In Russian). 45. Eroshenko V.А. Thermomolecular energy// Prom. teplotehnika. – 1992. – 14, № 1-3. – P. 22–25 (In Russian). 46. V.A. Eroshenko, "Heterogeneous structure for accumulation or dissipation of energy, process to use it and associated devices", Pat. WO 96/18040 France, F 15 B 1/04, F 16 F 9/00, 5/00, F 15 B 21/06, June 13, 1996. 47. V.A. Eroshenko et al., "A new paradigm of mechanical energy dissipation. Part. 2. Experimental investigation and effectiveness of a novel damper", Proc. Mech. Engrs. Pt D. J. Automobile Engng., vol. 221, no. 3, P. 301–312, 2007. 48. V.A. Eroshenko et al., "Les systèmes hétérogènes 'eau-zéolithe hydrophobe': de nouveaux ressorts moléculaires", vol. 3, P. 111–119, 2002. 49. Patent 1452262 СССР (1986–DSP) – RF (1993), А1 F03 G 7/00. Rotary heat engine Eroshenko/ В.А. Ерошенко; Presented 16.06.1986 (ДСП); Published 30.09.1996, Bulletin № 3 (In Russian). 13
1810.00942	2	1810	2019-05-09T04:54:48	Gate-Controlled VO2 Phase Transition for High-Performance Smart Window	[ "physics.app-ph", "cond-mat.str-el" ]	VO2 material is promising for developing energy-saving "smart window", owing to its thermochromic property induced by metal-insulator transition (MIT). However, its practical application is greatly limited by the relatively high critical transition temperature (~68oC), low luminous transmittance (<60%) and poor solar energy regulation ability (<15%). Here we developed a reversible and non-volatile electric-field control on the MIT of monoclinic VO2 film. With a solid electrolyte layer assisted gating treatment, we modulated the insertion/extraction of hydrogens into/from VO2 lattice at room temperature, causing tri-state phase transitions accompanied with controllable transmission adjustment. The dramatic increase of visible/infrared transmittance during the phase transition from the metallic (lightly H-doping) to insulating (heavily H-doping) phase leads to an increased solar energy regulation ability up to 26.5%, while keep 70.8% visible-luminous transmittance. These results beat all previous records and even exceeded the theoretical limit for traditional VO2 smart window, removing intrinsic disadvantages of VO2 for energy-saving utilizations. Our findings not only demonstrated an electric-field controlled phase modulation strategy, but also open the door for high-performance VO2-based smart window applications.	physics.app-ph	physics	S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E A P P L I E D P H Y S I C S Gate-controlled VO2 phase transition for high-performance smart windows Shi Chen1, Zhaowu Wang2,3, Hui Ren1, Yuliang Chen1, Wensheng Yan1, Chengming Wang2, Bowen Li1, Jun Jiang2†, Chongwen Zou1† Vanadium dioxide (VO2) is a promising material for developing energy-saving "smart windows," owing to its in- frared thermochromism induced by metal-insulator transition (MIT). However, its practical application is greatly limited by its relatively high critical temperature (~68°C), low luminous transmittance (<60%), and poor solar energy regulation ability (<15%). Here, we developed a reversible and nonvolatile electric field control of the MIT of a monoclinic VO2 film. With a solid electrolyte layer assisting gating treatment, we modulated the insertion/extraction of hydrogen into/from the VO2 lattice at room temperature, causing tristate phase transitions that enable control of light transmittance. The dramatic increase in visible/infrared transmittance due to the phase transition from the metallic (lightly H-doped) to the insulating (heavily H-doped) phase results in an increased solar energy regulation ability up to 26.5%, while maintaining 70.8% visible luminous transmittance. These results break all previous records and exceed the theoretical limit for traditional VO2 smart windows, making them ready for energy-saving utilization. Copyright © 2019 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC). INTRODUCTION Designing improved energy-saving materials is an active research topic in materials science because of the increasing concerns related to energy consumption and sustainable development. It is known that residential buildings are responsible for almost 30 to 40% of the primary energy consumption in the world (1). Thus, the use of chromogenic materials on building fenestration and the applications of building-integrated photovoltaic technologies such as solar cells have been demonstrated to be effective ways to increase the energy efficiency of buildings and reduce the energy consumption (2 -- 7). Notably, smart coatings based on thermochromic materials, the so-called "smart windows," can mod- ulate the transmittance of solar radiation in response to the change of temperature, making them very promising for improving the energy utilization efficiency of buildings (8). As a typical thermochromic material, vanadium dioxide (VO2) has been widely investigated because of its pronounced metal-insulator transition (MIT) behavior at the critical temperature of 68°C (9 -- 11). During the MIT process, VO2 shows a sharp change of resistance and a pronounced optical switching behavior especially in the infrared region, which makes it promising for energy-efficient coating applica- tions (12 -- 21). However, obstacles including high transition tempera- ture at ~68°C, low luminous transmittance (Tlum < 60%), and weak solar energy regulation ability (DTsol, usually <15%), make the VO2- based smart window still far from practical (22 -- 24). In this study, we developed an electrochromic smart window system based on a VO2 film with gate voltage control via a solid electrolyte. By adding gating voltage to control the H-doping level, we tuned non- destructive and reversible phase transformations between the insulating pristine monoclinic VO2, metallic HxVO2 (0 < x < 1), and insulating 1National Synchrotron Radiation Laboratory, University of Science and Technology of China, Hefei 230029, China. 2Hefei National Laboratory for Physical Sciences at the Microscale, CAS Center for Excellence in Nanoscience, CAS Key Laboratory of Mechanical Behavior and Design of Materials, School of Chemistry and Materials Science, University of Science and Technology of China, Hefei, Anhui 230026, China. 3School of Physics and Engineering, Henan University of Science and Technology, Henan Key Laboratory of Photoelectric Energy Storage Materials and Applications, Luoyang, Henan 471023, China. These authors contributed equally to this work. †Corresponding author. Email: [email protected] (J.J.); [email protected] (C.Z.) HVO2 phases at room temperature. Synchrotron-based spectroscopies were used to identify the crystal structures, electronic states, and the orbital occupancies during these field-controlled phase transitions. By analyzing the optical transmission properties of this system, we have demonstrated that the solar energy regulation ability is greatly augmented (DTsol ~ 26.5%), via the transition between the metallic HxVO2 (0 < x < 1) phase, which can block most infrared transmis- sions, and the insulating HVO2 substantially, which has a high visible luminous transmittance (Tlum up to ~70.8%). This not only substantial- ly surpasses the performance of systems reported previously but also exceeds the theoretical limit of DTsol ~ 23% for a pure VO2 material. Thus, the current results will provide a promising basis for the develop- ment of practical energy-saving devices in the future. RESULTS Gating-induced reversible three-phase transitions for VO2 films An electric field applied with ion liquid (IL) gating is effective in mod- ulating the properties of metal oxide materials. Recent reports showed that gating voltage with water-involved IL could induce the insertion/ extraction of either oxygen anions or hydrogen cations into/from some oxides, implying phase modulation through electric field control (25 -- 30). Here, we adopted a gel-like solid electrolyte [NaClO4 dissolved in poly- ethylene oxide (PEO) matrix; see Materials and Methods] as the gating layer (Fig. 1, A and B). The solid electrolyte covered both the prepared 30-nm VO2/Al2O3(0001) thin film and the three gold electrodes. In addi- tion, the source and drain electrodes had good contact with the VO2 thin film, realizing a typical field-effect transistor (FET) configuration. Positive or negative voltages applied through the gating electrode could drive the insertion/extraction of hydrogen ions into/from the VO2 lattice (29 -- 31), resulting in the stabilization of the metallic VO2 phase at room temperature due to hydrogen insertion. Hydrogen doping in the VO2 crystal was frequently studied un- der the high-temperature annealing treatment assisted by a noble metal catalyst in H2 gas (32 -- 34), which could also lead to the for- mation of a metallic H-doped VO2 state. Conversely, as reported recently by Yoon et al. (35) and our group (36), catalyst-assisted Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 1 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E Fig. 1. Experimental schemes. (A) The gating diagram of the VO2 device with source, drain, and gate electrodes. (B) Hydrogen ion movement under gating control. (C) The reversible insulator-metal-insulator tristate phase transitions of VO2 by tuning hydrogenating level with positive (blue) or negative (red) gating voltages. high-temperature annealing can produce another insulating HVO2 phase. Thus, is it possible to realize the tristate phase transitions as shown in Fig. 1C by the gating treatment at room temperature? As expected, phase transitions are found to be strongly dependent on the gating voltage. Figure 2A shows the changes in electrical resistance under gating voltages of 1.0 and 2.0 V, respectively, as a monitor of the phase transitions between insulator and metallic states (see the laboratory-scale gating device in fig. S1A). At a gating voltage of 1.0 V, the film resistance gradually decreases from ~106 to ~103 ohms even after 400 min (red curve), indicating a transition from an insulator to a metallic state. Upon the application of a gating voltage of 2.0 V (green curve), the film resistance first decreases from ~106 to ~3 × 103 ohms after about 25 min and then gradually increases up to ~2 × 105 ohms, indicating a transition from an insulator to a metallic state, and then back to an insulator state. This gating process is also sensitive to sample temperature (see fig. S1B): The higher the temperature, the quicker the gating-induced phase changes appear. For example, at 60°C, the gating-induced phase transition can be finished within several minutes, which demonstrates the potential for fast switching during practical applications. Figure 2B shows the x-ray diffraction (XRD) measurements for the pristine VO2 film and the samples gated by 1.0- or 2.0-V gating for 300 min, respectively. In contrast to the 39.8° XRD peak of the VO2 (020) diffraction for the nongated pristine sample, peaks at 39.4° and 36.8° are assigned to the gating-induced metallic film and the insulating sample by 1.0 and 2.0 V, respectively (36). One can also monitor the varying of XRD signals (Fig. 2C) to understand the gradual evolution of crystal structures during the gat- ing cycles. It is observed that the reversible phase modulations are made by adjusting voltages as a function of gating time (Fig. 2D). Starting from the pristine VO2 structure, the original insulating sam- ple is first transformed to the metallic phase by a gating voltage of 1.0 or 1.5 V. A new insulating phase is formed when the gating voltage is increased to 2.0 V. Then, upon the addition of a negative voltage of −1.0 V, the insulating VO2 is gradually converted to the metallic phase. The initial monoclinic VO2 phase is recovered by further increasing the negative voltage to −2.0 V. Loop tests were conducted by keeping the source-drain voltage of 0.3 V while sweeping the gating voltage at either a fast or a slow rate. Under the fast sweeping mode with a rate of 0.4 mV/s (Fig. 2E), the film resistance gradually decreased from ~106 to ~103 ohms as the gating voltage increased from 0.0 to 2.0 V (starting from point A), indicating a transition from an insulator to a metallic state. If the voltage decreased from 2.0 to −2.0 V, the metallic state (~103 ohms) was maintained, but once the negative gating voltage reached a thresh- old value, it completely changed to an insulator state (~106 ohms). Thus, during the fast-rate loop test, a simple and reversible transition between the insulator M-VO2 and the metallic HxVO2 state was observed. Con- versely, when applying the gating voltage at a slower sweeping rate of 0.1 mV/s, a quite different variation contour involving tristate phase transi- tions was obtained (Fig. 2F). Along the positive gating direction, the film was transformed from the initial M-VO2 state (point A) to the metallic HxVO2 state (point B), and then higher positive voltage led to another insulating HVO2 state (point C). Thereafter, as the gating voltage de- creased toward −2.0 V, the insulating HVO2 phase was first changed back to the metallic state and then eventually reverted to the initial insulating M-VO2 state, finishing a full cycle. This transition loop was ascribed to the sequential insulator-metal-insulator tristate phase tran- sitions. This was consistent with the resistance varying as a function of gating time in Fig. 2A, as the 2.0-V gating treatment gradually turned the pristine structure to the metallic state and later to the insulating phase. Accordingly, for gating treatment, both the gating voltage and gating time were key factors for phase modulation as they are the primary determinants of the degree of hydrogen insertion and diffusion into the VO2 lattice. Gating-induced hydrogen doping in VO2 films We then performed additional experiments to identify the varying of the VO2 crystal structure along with gating treatment. By gating with positive voltage, the cations will be driven into the VO2 lattice. Since the radius of Na+ (~0.102 nm) is a bit larger than that of V4+ (~0.058 nm), it is quite difficult to insert Na+ in the electrolyte into the VO2 lattice or for it to replace other cations there. The detailed x-ray photoelectron spectroscopy (XPS) test (see fig. S2) validates the situation, while sec- ondary ion mass spectrometry (SIMS) tests (see fig. S3) reveal that the metallic and insulating phases induced by positive gating hold increased hydrogen concentrations. The mechanism of hydrogen doping into the VO2 lattice by gating treatment basically involves the migration of hydrogen driven by the electric field, which is analogous to the charging and discharging pro- cess in a Li+ battery or the catalyst-free hydrogenation of VO2 in acid solution (37). The chemical reaction for the hydrogenation can be de- scribed as VO2þ xðH þÞ þ ðx eÞ → HxVO2ð0 < x ≤ 1; e is an electronÞ ð1Þ In this process, (H+) migrates into VO2 from the electrolyte driven by the positive gating voltage, and the electron is provided through the gating circuit. For the restoration process, the chemical reaction is HxVO2 → VO2 þ xðH þÞ þ ðx eÞ ð2Þ Protons will migrate into the electrolyte from the VO2 lattice be- cause of the negative gating voltage. For this gating treatment, the source of protons in the electrolyte should be considered (25 -- 27, 30). Our gating experiment shows that 1.0-V gating voltage can induce the metallic H-doped VO2 film. As Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 2 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E 106 ) s m h o ( e c n a t s s e R i 105 104 103 0 VG = 1.0 V VG = 2.0 V Al2O3 (0001) VO2 (020) 39.8o Pristine VO2 film 1.0 V to 300 min 2.0 V to 300 min 39.4o 36.8o . ) . u a ( y t i s n e n t I 50 100 150 200 250 300 350 400 Gating time (min) 36 37 38 39 2 (o) 40 41 42 Al2O3 (0001) VO2 (020) 0.0 V 1.0 V 1.5 V 2.0 V -- 1.0 V -- 2.0 V . ) . u a ( y t i s n e t n HVO2I VO2 HxVO2 HxVO2 VO2 36 38 40 2 (o) 42 44 106 ) s m h o ( e c n a 105 t s i s e R 104 103 -- 3 -- 3 -- 2 A (M-VO2) B (Metallic HxVO2) -- 1 0 Gating voltage (V) 1 2 ) s m h o ( e c n a i t s s e R 107 106 105 104 103 106 105 104 ) s m h o ( e c n a i t s s e R Hydrogenation VO2 HVO2 Dehydrogenation VO2 +1.0 V +2.0 V +1.5 V -- 2.0 V -- 1.0 V HxVO2 0 600 300 Gating time (min) 900 HxVO2 1200 1500 A (M-VO2) C (Insulating HVO2) A (M-VO2) l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h B (Metallic HxVO2) B (Metallic HxVO2) -- 2.0 -- 1.5 -- 2.0 -- 1.5 -- 1.0 -- 0.5 0.0 0.5 1.0 1.5 2.0 Gating voltage (V) 1 5 , 2 0 1 9 Fig. 2. Phase modulation by the gating treatment. (A) The resistance changes as a function of gating time at gating voltages of 1.0 V (red) or 2.0 V (green). (B) XRD curves for the pristine VO2 film (black) and for VO2 films subjected to gating voltages of 1.0 V (red) or 2.0 V (green) for 300 min. a.u., arbitrary units. (C) XRD curves for VO2 films treated with different gating voltages. To monitor the intermediate states, the gating time was set to 60 min for voltages of 1.0 and −2.0 V, while it was set to 60 and 120 min for voltages of 1.5, 2.0, and −1.0 V. (D) The reversible phase modulations by different voltages as a function of gating time. (E) A cycling resistance test plot produced by sweeping the gating voltage from −2.0 to 2.0 V at a sweep rate of 0.4 mV/s and a source-drain voltage of 0.3 V, indicating a series of reversible phase transitions between the insulating monoclinic M-VO2 state (starting from point A) and the metallic HxVO2 state. (F) A cycling resistance test plot at a slower sweep rate of 0.1 mV/s and a source-drain voltage of 0.3 V, indicating a series of reversible phase transitions between three states: the initial monoclinic insulator M-VO2 state (point A), the metallic HxVO2 state (point B), and the insulating HVO2 state (point C). Arrows indicate the direction of resistance changes during voltage sweeping. this voltage is less than the 1.23 V required for electrolysis of water, the adsorbed water in the electrolyte layer is unlikely to be the source of protons. Residual methanol in the electrolyte (see fig. S4) might be the possible source of protons because of its lower electrolysis voltage, although further chemical or electrochemical studies are required to confirm it. In addition, it was also observed that these H-doped VO2 films were quite stable in air (see fig. S5), which demonstrated the non- volatile property of the electric field control phase transitions under ambient conditions. Raman tests (see fig. S6) on gated samples sug- gested that the hydrogenated VO2 film showed a rutile-like phase structure since the characteristic peaks at 192, 223, and 617 cm−1 in pristine monoclinic VO2 were completely absent. These phase structures were sensitive to the H-doping concentration, providing further validation that hydrogen ions are being inserted into the VO2 lattice. Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 3 of 8 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E Detailed synchrotron-based XPS and x-ray absorption near-edge spectroscopy (XANES) in Fig. 3 show that pronounced peaks due to the presence of -- OH emerge in the metallic phase after gating treatment at 1.0 to 1.5 V. In addition, the peak becomes much stronger with higher voltage and longer treatment time (Fig. 3A), reflecting a higher H-doping concentration in the newly formed insulating state. Specifically, the V2p3/2 peaks in the XPS curves become broader and slightly shift toward the low binding energy direction after gating treatment, indicating the appearance of a small amount of V3+ state due to the charge transfer induced by H atoms (38). Synchrotron-based XANES for V L-edge in Fig. 3B confirms the electronic structure variations induced by H-doping during the gating treatment. The V LIII-edge undergoes a pronounced shift to lower energy after gating, indicating that the V atoms are in a lower valence state (39, 40). The O K-edges in Fig. 3C show the transitions from O1s to O2p states, in which two distinct features of the t2g and eg peaks reflect the unoccupied states. The intensity ratio of t2g/eg decreased substantially with the gating treatment, which is caused by the increased H-doping concentration and reflects the fact that t2g levels including the d// and p* orbitals are gradually filled by electrons. Theoretical simulations for the gating effects First-principles calculations at the density functional theory (DFT) level were carried out to explore the underlying microscopic mechanism of the gating effects. The formation energies of H-doping and H-diffusion energy barrier were computed using supercells containing different H concentrations: V4O8, HV4O8, and H2V4O8 (Fig. 4, A to C). The for- mation energy of H-doping impurity is near −3.6 eV, which is in- sensitive to H concentrations (see table S1). This is because all H impurities are stabilized at the center of lattice vacancy, forming rela- tively strong H -- -- O bonds, as shown in Fig. 4 (B and C) (see also fig. S7). In contrast, the H-diffusion energy barrier becomes increasingly larger with the increase in H-doping concentration. In VO2, the channels for H diffusion are mainly along the [100] direction because there are no V atoms or O atoms to hinder the diffusion (Fig. 4A). The calculated H-diffusion energy barrier value is 0.8 eV along the [100] direction (Fig. 4D). Conversely, the diffusion barriers in HV4O8 and H2V4O8 were increased to 1.15 eV (Fig. 4E) and 1.38 eV (Fig. 4F), respectively. This increase is due to blockage of the channels for H diffusion by the intercalation of H atoms. Moreover, the geometric variations induced V-O -- OH V-LIII V-L II V4+ V3+ ) . u . a ( y t i s n e t n I ) . u . a ( y t i s n e t n I Pure VO2 1.0 V at 60 min 1.5 V at 60 min 2.0 V at 60 min 2.0 V at 240 min Pure VO2 2.0 V to 30 min 2.0 V to 60 min 2.0 V to 240 min 2.0 V to 360 min ) . u . a ( y t i s n e t n I O K-edge t2g eg Pure VO2 2.0 V to 30 min 2.0 V to 60 min 2.0 V to 240 min 2.0 V to 360 min 510 515 525 520 Binding energy (eV) 530 535 540 507 510 516 513 Photon energy (eV) 519 522 525 526 528 530 532 534 Photon energy (eV) Fig. 3. Chemical and electronic states of gating-treated VO2 films. (A) XPS curves of VO2 films with different gating voltages. (B) V L-edge XANES curves and (C) O K-edge curves for the samples gated at 2.0 V with different gating times. A D B E C F Fig. 4. Theoretical calculation for H insertion and diffusion processes. (A to C) Optimized atomic structures of V4O8 (pure VO2), HV4O8, and H2V4O8, respectively. Purple beads with arrows represent H-diffusion channels along the [100] direction. (D to F) Calculated energy barriers for diffusion of H atoms into V4O8, HV4O8, and H2V4O8, respectively. Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 4 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E by initial H-doping may also hinder the diffusion of successive H atoms. This H-dependent barrier increase explains our experimental finding that a high gating voltage of 2.0 V was required to induce the large H-doping concentration required to generate the insulation phase of the H-doped VO2 film. Gate-controlled VO2-based smart window with high performance The smart window concept was proposed many years ago, and many materials showed promise for use as smart windows, including hydro- gels, liquid crystals, transition metal oxides, and perovskite (41, 42), be- cause of their special optical response in the visible region upon external stimulus/excitation. The VO2 material exhibited a special switching ef- fect in the near-infrared (NIR) range across the phase transition, which demonstrated a unique advantage in energy-saving applications. Since the phase transformations of VO2 films can be controlled by gating treatment at room temperature, we were able to apply it to a voltage- controlled smart window, as shown schematically in Fig. 5A. For the VO2 films with three different phase structures, obvious color changes were observed. The pure VO2 film and the gating-produced metallic HxVO2 film showed a similar yellowish color, while the heavily H-doped VO2 film was almost transparent. In Fig. 5B, the metallic HxVO2 film showed transmittance spectra similar to those of the rutile VO2 film (VO2 film at 90°C), demonstrating a pronounced switching-off effect in the infrared region. Conversely, the insulating HVO2 film maintained high transparency in both visible light and infrared regions. For exam- ple, the transmittance at 650 nm for the pristine VO2 film and metallic HxVO2 was about 56%, while that of the insulating HVO2 film was increased up to 72%. In the infrared region, the transmittance difference at 2000 nm for the pure VO2 film across the MIT process (25° and 90°C) was about 40.3%, while the change from the metallic HxVO2 film to the insulated HVO2 film reached 49.1%. On the basis of these spectral measurements, the integrated lumi- nous transmittances (380 to 780 nm) and solar transmittances (350 to 2500 nm) can be calculated by the following equations (18) Tlum ¼ ∫FlumðlÞTðlÞdl=∫FlumðlÞdl Tsol ¼ ∫FsolðlÞTðlÞdl=∫FsolðlÞdl ð3Þ ð4Þ where T(l) is the transmittance function, Flum is equal to the standard luminous efficiency function for photopic vision, and Fsol represents the solar irradiance spectrum for an air mass of 1.5. The solar modula- tion ability (DTsol) of the metallic HxVO2 and insulating HVO2 films can be calculated by DTsol ¼ Tsol;insulatingHVO2 Tsol;metallicH1xVO2 ð5Þ According to the above formula, the transparent, heavily insulating HVO2 film exhibits a high Tlum value of about 70.8%, while the transition between the metallic and insulating phases of the H-doped VO2 film is accompanied by a good DTsol value of 26.5% (Fig. 5C). ) % ( e c n a t t i m s n a r T ) % ( y t i l i b a n o i t l a u d o m l r a o S 72% 56% 650 nm 2000 nm HVO2 VO2 at 25oC 49.1% 40.3% HxVO2 VO2 at 90oC 500 1000 1500 (nm) 2000 2500 Current work 90 80 70 60 50 40 30 20 25 20 15 10 5 20 30 60 Luminous transmittance (%) 40 50 70 Fig. 5. Optical properties in relation to possible smart window applications. (A) The schematic diagram of an electrochromic smart window. Properties of three films with different phases are compared. Both the pure VO2 film and the gating-produced metallic HxVO2 film show a similar yellowish color, while the HVO2 film is almost transparent. USTC, University of Science and Technology of China. (B) Optical transmittance spectra of VO2, HxVO2, and HVO2 thin films. The transmittance at 650 nm for the pristine VO2 film and metallic HxVO2 is about 56%, while the insulating HVO2 film has an increased transmittance of 72%. In the infrared region, the transmittance difference at 2000 nm for the pure VO2 film across the MIT process (transmittance of film at 25°C compared to 90°C) is about 40.3%, while the change in transmittance from the metallic HxVO2 film to the insulating HVO2 film is higher at 49.1%. (C) A plot of luminous transmittance Tlum versus solar modulation ability DTsol that highlights the results of this work (★) compared to previously reported data. Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 5 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E The data from previous reports for traditional VO2 smart windows (see table S2) are also plotted. The performance of a smart window based on the transition between the metallic and insulating phases of an H-doped VO2 film appreciably surpasses previous records. Furthermore, it even surpasses the theoretical limitation of the DTsol of ~23% for a traditional VO2 smart window, which can only be achieved by texture surface de- sign or complex multiyear film fabrication (43). For architectural smart coatings, the excellent luminous transmit- tance in the visible range and high modulation efficiency in the NIR regions are both important and greatly desirable. There are two long- lasting issues hindering the practical application of VO2 in smart windows: low luminous transmittance (Tlum) and poor solar energy reg- ulation ability (DTsol). For a traditional smart window constructed with the pure VO2 film, increasing the VO2 film thickness is necessary for better infrared switching performace, and as a result, the visible light transmittance will be greatly decreased. This problem is caused by the intrinsic band structures and optical properties of VO2, making it very difficult to be solved for the theromochromic VO2 smart window. In addition, its relatively high critical temperature (~68°C) is another big obstacle for practical applications (14, 16). Thus, in this work, the voltage-controlled phase transitions between the metallic HxVO2 and insulating HVO2 films not only overcome the temperature issue by operating at room temperature but also demonstrate high Tlum and large DTsol values in Fig. 5C, showing the unique advantages for prac- tical smart window applications. DISCUSSION In this study, we have achieved a reversible tristate phase transformation in a VO2 epitaxial film by using solid electrolyte -- assisted electric field control. These gating voltage -- controlled phase transitions originate from the insertion/extraction of hydrogens into/from the VO2 lattice. By changing the voltage, we can reversibly modulate the phase transforma- tions among the insulating monoclinic VO2, the metallic H1−xVO2 (0 < x < 1), and the insulating HVO2 phase structures. The transition between the metallic HxVO2 phase and the insulating HVO2 phase shows pronounced electrochromic effects at room temperature with greatly enhanced luminous transmittance in the visible range and the highest modulation efficiency in the NIR regions. This system overcomes the primary weaknesses of the traditional VO2 smart window. Our findings thus demonstrate a facile route using phase modulation by a solid elec- trolyte for creating realistic energy-saving devices in the future. MATERIALS AND METHODS Thin-film growth High-quality VO2 (020) epitaxial films were grown on an Al2O3 (0001) crystal substrate by radio frequency (rf) plasma -- assisted oxide molecu- lar beam epitaxy equipment; more details on film preparation were re- ported elsewhere (44). Solid electrolyte preparation NaClO4 (Sigma-Aldrich) dissolved in a PEO [molecular weight (Mw) = 100,000; Sigma-Aldrich] matrix was used as the solid electrolyte. NaClO4 and PEO powders (0.3 and 1 g, respectively) were mixed with 15 ml of anhydrous methanol (Alfa Aesar). The mixture was then stirred over- night at 50°C until a gel-like solid electrolyte was formed (45). A 10 mm by 10 mm electrolyte-covered VO2 thin-film FET structure with three terminals was fabricated. Characterizations Electrical properties were conducted by Keithley 2400 sourcemeters at room temperature. For the gating test, the voltage sweep rates were set to 1.0 V/2500 s for the fast mode and 1.0 V/10,000 s for the slow mode. Raman spectra (JY LabRAM HR, Ar+ laser at 514.5 nm, 0.5 mW) were recorded for different samples at room temperature. The laser spot size was 1 mm by 2 mm, and five points were tested for each film sample, which showed consistent results. The XPS (Thermo ESCALAB 250; Al Ka, 1486.7 eV) was used to examine the chemical states of the VO2 film samples before and after gat- ing treatment. To detect the hydrogen concentration in the gated VO2 film sample, SIMS measurements (Quad PHI 6600) were conducted. XANES was conducted at the x-ray magnetic circular dichroism (XMCD) beamline (BL12B) at the National Synchrotron Radiation Laboratory (NSRL), Hefei. The NSRL is a second-generation accelerator with a 0.8-GeV storage ring, which is suitable for soft x-ray experiments. Total electron yield mode was applied to collect the sample drain cur- rent under a vacuum higher than 3.0 × 10−10 torr. The energy range is 100 to 1000 eV with an energy resolution (DE/E) of 1.0 × 10−3 at 1.0 keV, and the beam size is about 1.0 mm by 3.0 mm. High-resolution synchro- tron XRD curves were characterized at the BL14B beamline of the Shanghai Synchrotron Radiation Facility (SSRF). The SSRF is a third- generation accelerator with a 3.5-GeV storage ring. The BL14B beam- line shows an energy resolution (DE/E) of 1.5 × 10−4 at 10 keV and a beam size of 0.3 mm by 0.35 mm with a photon flux of up to 2 × 1012 phs/s at 10 keV. Considering the photon flux distribution and the resolution, the 0.124-nm x-ray was chosen for the experiment. First-principles calculations All calculations were based on DFT, using the Vienna ab initio Sim- ulation Package code (46). Exchange and correlation terms were de- scribed using general gradient approximation (GGA) in the scheme of Perdew-Burke-Ernzerhof (47). Core electrons were described by pseudopotentials generated from the projector-augmented wave method (48), and valence electrons were expanded in a plane-wave basis set with an energy cutoff of 480 eV. The GGA + U method was used to optimize the structure; U and J were chosen to be 4 and 0.68 eV, respec- tively. The relaxation was carried out until all forces on the free ions converged to 0.01 eV/Å. The climbing image nudged elastic band (49) method was used to find the minimum energy paths and the transition states for diffusion of H into VO2 and H-doped VO2. SUPPLEMENTARY MATERIALS Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/5/3/eaav6815/DC1 Fig. S1. Temperature-dependent gating-induced phase transitions. Fig. S2. XPS spectra for the pure and gated VO2 samples. Fig. S3. SIMS test for the gated VO2 film samples. Fig. S4. Fourier transform infrared test for the residual methanol in PEO electrolyte. Fig. S5. Stability test for the H-doped VO2 films. Fig. S6. Raman spectra for the films. Fig. S7. Atomic structures of H4V4O8. Table S1. Formation energies calculations. Table S2. Comparison of the solar energy regulation ability (DTsol) and luminous transmittance (Tlum) values for various VO2-based smart-windows. References (50 -- 58) REFERENCES AND NOTES 1. T. Ramesh, R. Prakash, K. K. Shukla, Life cycle energy analysis of buildings: An overview. Energy Build. 42, 1592 -- 1600 (2010). 2. D. Bi, W. Tress, M. I. Dar, P. Gao, J. Luo, C. Renevier, K. Schenk, A. Abate, F. Giordano, J.-P. C. Baena, J.-D. Decoppet, S. M. Zakeeruddin, M. K. Nazeeruddin, M. Grätzel, Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 6 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E A. Hagfeldt, Efficient luminescent solar cells based on tailored mixed-cation perovskites. Sci. Adv. 2, e1501170 (2016). 3. C. Costa, I. Mesquita, L. Andrade, A. Mendes, Photoelectrochromic devices: Influence of device architecture and electrolyte composition. Electrochim. Acta 219, 99 -- 106 (2016). 4. F. Bella, J. Popovic, A. Lamberti, E. Tresso, C. Gerbaldi, J. Maier, Interfacial effects in solid-liquid electrolytes for improved stability and performance of dye-sensitized solar cells. ACS Appl. Mater. Interfaces 9, 37797 -- 37803 (2017). 5. M. Imperiyka, A. Ahmad, S. A. Hanifah, F. Bella, A UV-prepared linear polymer electrolyte membrane for dye-sensitized solar cells. Physica. B Condens. Matter 450, 151 -- 154 (2014). 6. D. Pintossi, G. Iannaccone, A. Colombo, F. Bella, M. Välimäki, K.-L. Väisänen, J. Hast, M. Levi, C. Gerbaldi, C. Dragonetti, S. Turri, G. Griffini, Luminescent downshifting by photo-induced sol-gel hybrid coatings: Accessing multifunctionality on flexible organic photovoltaics via ambient temperature material processing. Adv. Electron. Mater. 2, 1600288 (2016). 7. S. Galliano, F. Bella, G. Piana, G. Giacona, G. Viscardi, C. Gerbaldi, M. Grätzel, C. Barolo, Finely tuning electrolytes and photoanodes in aqueous solar cells by experimental design. Sol. Energy 163, 251 -- 255 (2018). 26. K. Liu, D. Fu, J. Cao, J. Suh, K. X. Wang, C. Cheng, D. F. Ogletree, H. Guo, S. Sengupta, A. Khan, C. W. Yeung, S. Salahuddin, M. M. Deshmukh, J. Wu, Dense electron system from gate-controlled surface metal-insulator transition. Nano Lett. 12, 6272 -- 6277 (2012). 27. H. Ji, J. Wei, D. Natelson, Modulation of the electrical properties of VO2 nanobeams using an ionic liquid as a gating medium. Nano Lett. 12, 2988 -- 2992 (2012). 28. J. Jeong, N. Aetukuri, T. Graf, T. D. Schladt, M. G. Samant, S. S. P. Parkin, Suppression of metal-insulator transition in VO2 by electric field-induced oxygen vacancy formation. Science 339, 1402 -- 1405 (2013). 29. Y. Zhou, J. Park, J. Shi, M. Chhowalla, H. Park, D. A. Weitz, S. Ramanathan, Control of emergent properties at a correlated oxide interface with graphene. Nano Lett. 15, 1627 -- 1634 (2015). 30. K. Shibuya, A. Sawa, Modulation of metal -- insulator transition in VO2 by electrolyte gating-induced protonation. Adv. Electron. Mater. 2, 1500131 (2016). 31. M. Nakano, D. Okuyama, K. Shibuya, M. Mizumaki, H. Ohsumi, M. Yoshida, M. Takata, M. Kawasaki, Y. Tokura, T. Arima, Y. Iwasa, Distinct substrate effect on the reversibility of the metal -- insulator transitions in electrolyte-gated VO2 thin films. Adv. Electron. Mater. 1, 1500093 (2015). 8. R. Baetens, B. P. Jelle, A. Gustavsen, Properties, requirements and possibilities of smart windows for dynamic daylight and solar energy control in buildings: A state-of-the-art review. Sol. Energy Mater. Sol. Cells 94, 87 -- 105 (2010). 32. K. Sato, Y. Itoh, T. Aida, Columnarly assembled liquid-crystalline peptidic macrocycles unidirectionally orientable over a large area by an electric field. J. Am. Chem. Soc. 133, 13767 -- 13769 (2011). 9. R. M. Wentzcovitch, W. W. Schulz, P. B. Allen, VO2: Peierls or Mott-Hubbard? A view from 33. Y. Filinchuk, N. A. Tumanov, V. Ban, H. Ji, J. Wei, M. W. Swift, A. H. Nevidomskyy, band theory. Phys. Rev. Lett. 72, 3389 -- 3392 (1994). 10. M. M. Qazilbash, M. Brehm, B.-G. Chae, P.-C. Ho, G. O. Andreev, B.-J. Kim, S. J. Yun, A. V. Balatsky, M. B. Maple, F. Keilmann, H.-T. Kim, D. N. Basov, Mott transition in VO2 revealed by infrared spectroscopy and nano-imaging. Science 318, 1750 -- 1753 (2007). 11. N. B. Aetukuri, A. X. Gray, M. Drouard, M. Cossale, L. Gao, A. H. Reid, R. Kukreja, H. Ohldag, C. A. Jenkins, E. Arenholz, K. P. Roche, H. A. Dürr, M. G. Samant, S. S. P. Parkin, Control of the metal-insulator transition in vanadium dioxide by modifying orbital occupancy. Nat. Phys. 9, 661 -- 666 (2013). D. Natelson, In situ diffraction study of catalytic hydrogenation of VO2: Stable phases and origins of metallicity. J. Am. Chem. Soc. 136, 8100 -- 8109 (2014). 34. J. Lin, H. Ji, M. W. Swift, W. J. Hardy, Z. Peng, X. Fan, A. H. Nevidomskyy, J. M. Tour, D. Natelson, Hydrogen diffusion and stabilization in single-crystal VO2 micro/nanobeams by direct atomic hydrogenation. Nano Lett. 14, 5445 -- 5451 (2014). 35. H. Yoon, M. Choi, T.-W. Lim, H. Kwon, K. Ihm, J. K. Kim, S.-Y. Choi, J. Son, Reversible phase modulation and hydrogen storage in multivalent VO2 epitaxial thin films. Nat. Mater. 15, 1113 -- 1119 (2016). 12. H. Kim, Y. Kim, K. S. Kim, H. Y. Jeong, A.-R. Jang, S. H. Han, D. H. Yoon, K. S. Suh, H. S. Shin, 36. S. Chen, Z. Wang, L. Fan, Y. Chen, H. Ren, H. Ji, D. Natelson, Y. Huang, J. Jiang, C. Zou, T. Kim, W. S. Yang, Flexible thermochromic window based on hybridized VO2/graphene. ACS Nano 7, 5769 -- 5776 (2013). Sequential insulator-metal-insulator phase transitions of VO2 triggered by hydrogen doping. Phys. Rev. B 96, 125130 (2017). 13. M. Li, S. Ji, J. Pan, H. Wu, L. Zhong, Q. Wang, F. Li, G. Li, Infrared response of self-heating VO2 nanoparticles film based on Ag nanowires heater. J. Mater. Chem. A 2, 20470 -- 20473 (2014). 14. M. Li, X. Wu, L. Li, Y. Wang, D. Li, J. Pan, S. Li, L. Sun, G. Li, Defect-mediated phase transition temperature of VO2 (M) nanoparticles with excellent thermochromic performance and low threshold voltage. J. Mater. Chem. A 2, 4520 -- 4523 (2014). 37. Y. Chen, Z. Wang, S. Chen, H. Ren, L. Wang, G. Zhang, Y. Lu, J. Jiang, C. Zou, Y. Luo, Non-catalytic hydrogenation of VO2 in acid solution. Nat. Commun. 9, 818 (2018). 38. J. Wei, H. Ji, W. Guo, A. H. Nevidomskyy, D. Natelson, Hydrogen stabilization of metallic vanadium dioxide in single-crystal nanobeams. Nat. Nanotechnol. 7, 357 -- 362 (2012). 39. M. Abbate, The O 1s and sV 2p X-ray absorption spectra of vanadium oxides. Brazilian J. Phys. 24, 785 -- 795 (1994). 15. J. Zheng, S. Bao, P. Jin, TiO2(R)/VO2(M)/TiO2(A) multilayer film as smart window: 40. R. Zimmermann, R. Claessen, F. Reinert, P. Steiner, S. Hüfner, Strong hybridization in Combination of energy-saving, antifogging and self-cleaning functions. Nano Energy 11, 136 -- 145 (2015). vanadium oxides: Evidence from photoemission and absorption spectroscopy. J. Phys. Condens. Matter 10, 5697 (1998). 16. J. Zhu, A. Huang, H. Ma, Y. Ma, K. Tong, S. Ji, S. Bao, X. Cao, P. Jin, Composite film of 41. Y. Ke, C. Zhou, Y. Zhou, S. Wang, S. H. Chan, Y. Long, Emerging thermal-responsive vanadium dioxide nanoparticles and ionic liquid-nickel-chlorine complexes with excellent visible thermochromic performance. ACS Appl. Mater. Interfaces 8, 29742 -- 29748 (2016). 17. M. Liu, B. Su, Y. V. Kaneti, Z. Chen, Y. Tang, Y. Yuan, Y. Gao, L. Jiang, X. Jiang, A. Yu, Dual-phase transformation: Spontaneous self-template surface-patterning strategy for ultra-transparent VO2 solar modulating coatings. ACS Nano 11, 407 -- 415 (2017). 18. T. Chang, X. Cao, N. Li, S. Long, X. Gao, L. R. Dedon, G. Sun, H. Luo, P. Jin, Facile and low-temperature fabrication of thermochromic Cr2O3/VO2 smart coatings: Enhanced solar modulation ability, high luminous transmittance and UV-shielding function. ACS Appl. Mater. Interfaces 9, 26029 -- 26037 (2017). 19. Y. Chen, L. Fan, Q. Fang, W. Xu, S. Chen, G. Zan, H. Ren, L. Song, C. Zou, Free-standing SWNTs/VO2/Mica hierarchical films for high-performance thermochromic devices. Nano Energy 31, 144 -- 151 (2017). 20. T. Chang, X. Cao, L. R. Dedon, S. Long, A. Huang, Z. Shao, N. Li, H. Luo, P. Jin, Optical design and stability study for ultrahigh-performance and long-lived vanadium dioxide-based thermochromic coatings. Nano Energy 44, 256 -- 264 (2018). materials and integrated techniques targeting the energy-efficient smart window application. Adv. Funct. Mater. 28, 1800113 (2018). 42. J. Lin, M. Lai, L. Dou, C. S. Kley, H. Chen, F. Peng, J. Sun, D. Lu, S. A. Hawks, C. Xie, F. Cui, A. P. Alivisatos, D. T. Limmer, P. Yang, Thermochromic halide perovskite solar cells. Nat. Mater. 17, 261 -- 267 (2018). 43. A. Taylor, I. Parkin, N. Noor, C. Tummeltshammer, M. S. Brown, I. Papakonstantinou, A bioinspired solution for spectrally selective thermochromic VO2 coated intelligent glazing. Opt. Express 21, A750 -- A764 (2013). 44. L. L. Fan, S. Chen, Y. F. Wu, F. H. Chen, W. S. Chu, X. Chen, C. W. Zou, Z. Y. Wu, Growth and phase transition characteristics of pure M-phase VO2 epitaxial film prepared by oxide molecular beam epitaxy. Appl. Phys. Lett. 103, 131914 (2013). 45. Y. Yu, F. Yang, X. F. Lu, Y. J. Yan, Y.-H. Cho, L. Ma, X. Niu, S. Kim, Y.-W. Son, D. Feng, S. Li, S.-W. Cheong, X. H. Chen, Y. Zhang, Gate-tunable phase transitions in thin flakes of 1T-TaS2. Nat. Nanotechnol. 10, 270 -- 276 (2015). 46. G. Kresse, D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758 -- 1775 (1999). 21. Y. Wang, X. Sun, Z. Chen, Z. Cai, H. Zhou, T.-M. Lu, J. Shi, Defect-engineered epitaxial VO2±d 47. J. P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple. in strain engineering of heterogeneous soft crystals. Sci. Adv. 4, eaar3679 (2018). Phys. Rev. Lett. 77, 3865 -- 3868 (1996). 22. S.-Y. Li, G. A. Niklasson, C. G. Granqvist, Thermochromic fenestration with VO2-based materials: Three challenges and how they can be met. Thin Solid Films 520, 3823 -- 3828 (2012). 23. M. Li, S. Magdassi, Y. Gao, Y. Long, Hydrothermal synthesis of VO2 polymorphs: Advantages, challenges and prospects for the application of energy efficient smart windows. Small 13, 1701147 (2017). 48. G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169 -- 11186 (1996). 49. D. Sheppard, P. Xiao, W. Chemelewski, D. D. Johnson, G. Henkelman, A generalized solid-state nudged elastic band method. J. Chem. Phys. 136, 074103 (2012). 50. Y. Choi, Y. Jung, H. Kim, Low-temperature deposition of thermochromic VO2 thin films on glass substrates. Thin Solid Films 615, 437 -- 445 (2016). 24. S. Wang, M. Liu, L. Kong, Y. Long, X. Jiang, A. Yu, Recent progress in VO2 smart coatings: Strategies to improve the thermochromic properties. Prog. Mater. Sci. 81, 1 -- 54 (2016). 25. N. Lu, P. Zhang, Q. Zhang, R. Qiao, Q. He, H.-B. Li, Y. Wang, J. Guo, D. Zhang, Z. Duan, Z. Li, M. Wang, S. Yang, M. Yan, E. Arenholz, S. Zhou, W. Yang, L. Gu, C.-W. Nan, J. Wu, Y. Tokura, P. Yu, Electric-field control of tri-state phase transformation with a selective dual-ion switch. Nature 546, 124 -- 128 (2017). 51. B. Zhu, H. Tao, X. Zhao, Effect of buffer layer on thermochromic performances of VO2 films fabricated by magnetron sputtering. Infrared Phys. Technol. 75, 22 -- 25 (2016). 52. M. Panagopoulou, E. Gagaoudakis, N. Boukos, E. Aperathitis, G. Kiriakidis, D. Tsoukalas, Y. S. Raptis, Thermochromic performance of Mg-doped VO2 thin films on functional substrates for glazing applications. Sol. Energy Mater. Sol. Cells 157, 1004 -- 1010 (2016). Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 7 of 8 l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 S C I E N C E A D V A N C E S R E S E A R C H A R T I C L E 53. S. Long, H. Zhou, S. Bao, Y. Xin, X. Cao, P. Jin, Thermochromic multilayer films of WO3/VO2/WO3 sandwich structure with enhanced luminous transmittance and durability. RSC Adv. 6, 106435 -- 106442 (2016). 54. G. Sun, X. Cao, X. Gao, S. Long, M. Liang, P. Jin, Structure and enhanced thermochromic performance of low-temperature fabricated VO2/V2O3 thin film. Appl. Phys. Lett. 109, 143903 (2016). 55. H. Liu, D. Wan, A. Ishaq, L. Chen, B. Guo, S. Shi, H. Luo, Y. Gao, Sputtering deposition of sandwich-structured V2O5/metal (V, W)/V2O5 multilayers for the preparation of high- performance thermally sensitive VO2 thin films with selectivity of VO2 (B) and VO2 (M) polymorph. ACS Appl. Mater. Interfaces 8, 7884 -- 7890 (2016). 56. X. Wu, Z. Wu, H. Zhang, R. Niu, Q. He, C. Ji, J. Wang, Y. Jiang, Enhancement of VO2 thermochromic properties by Si doping. Surf. Coatings Technol. 276, 248 -- 253 (2015). 57. G. Sun, X. Cao, H. Zhou, S. Bao, P. Jin, A novel multifunctional thermochromic structure with skin comfort design for smart window application. Sol. Energy Mater. Sol. Cells 159, 553 -- 559 (2017). 58. E. Gagaoudakis, I. Kortidis, G. Michail, K. Tsagaraki, V. Binas, G. Kiriakidis, E. Aperathitis, Study of low temperature rf-sputtered Mg-doped vanadium dioxide thermochromic films deposited on low-emissivity substrates. Thin Solid Films 601, 99 -- 105 (2016). Research Funds for the Central Universities, the Youth Innovation Promotion Association CAS, the Major/Innovative Program of Development Foundation of Hefei Center for Physical This Science and Technology, and the China Postdoctoral Science Foundation (2017M622002). This work was partially carried out at the USTC Center for Micro- and Nanoscale Research and work was partially carried out at the USTC Center for Micro- and Nanoscale Research and Fabrication. Fabrication. We also acknowledge support from the XMCD beamline (BL12B), the photoelectron spectroscopy beamline (BL10B) at the NSRL (Hefei), and the x-ray diffraction beamline (BL14B1) at the SSRF. Author contributions: J.J. and C.Z. conceived the study. S.C. and C.Z. designed the experiments and performed the initial tests. Z.W. and J.J. conducted the theoretical calculations. S.C., H.R., Y.C., W.Y., C.W., and B.L. conducted the sample characterizations and synchrotron-based measurements. S.C., J.J., and C.Z. wrote the manuscript. All authors discussed the results and commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. Submitted 10 October 2018 Accepted 30 January 2019 Published 15 March 2019 10.1126/sciadv.aav6815 Acknowledgments Funding: This work was partially supported by the National Key Research and Development Program of China (2018YFA0208603, 2016YFA0401004), the National Natural Science Foundation of China (U1432249, 11404095, 11574279, and 11704362), the Fundamental Citation: S. Chen, Z. Wang, H. Ren, Y. Chen, W. Yan, C. Wang, B. Li, J. Jiang, C. Zou, Gate- controlled VO2 phase transition for high-performance smart windows. Sci. Adv. 5, eaav6815 (2019). l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g i . o r g / o n M a r c h 1 5 , 2 0 1 9 Chen et al., Sci. Adv. 2019; 5 : eaav6815 15 March 2019 8 of 8 Gate-controlled VO 2 Shi Chen, Zhaowu Wang, Hui Ren, Yuliang Chen, Wensheng Yan, Chengming Wang, Bowen Li, Jun Jiang and Chongwen Zou phase transition for high-performance smart windows 5 Sci Adv DOI: 10.1126/sciadv.aav6815 (3), eaav6815. ARTICLE TOOLS http://advances.sciencemag.org/content/5/3/eaav6815 SUPPLEMENTARY MATERIALS REFERENCES http://advances.sciencemag.org/content/suppl/2019/03/11/5.3.eaav6815.DC1 This article cites 58 articles, 4 of which you can access for free http://advances.sciencemag.org/content/5/3/eaav6815#BIBL PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions Use of this article is subject to the Terms of Service l D o w n o a d e d f r o m h t t p : / / a d v a n c e s . s c e n c e m a g . o r g / o n M a r c h 1 5 , 2 0 1 9 i (ISSN 2375-2548) is published by the American Association for the Advancement of Science, 1200 New Science Advances York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title registered trademark of AAAS. Science Advances is a
1902.10392	1	1902	2019-02-27T08:51:17	Enhancing thermophotovoltaic performance using graphene-BN-InSb near-field heterostructures	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Graphene---hexagonal-boron-nitride---InSb near-field structures are designed and optimized to enhance the output power and energy efficiency of the thermophotovoltaic systems working in the temperature range of common industrial waste heat, $400~\rm K \sim 800~\rm K$, which is also the working temperature range for conventional thermoelectric devices. We show that the optimal output electric power can reach $3.5\times10^{4} \rm\ W/\rm m^2$ for the system with a graphene---hexagonal-boron-nitride heterostructure emitter and a graphene-covered InSb cell, whereas the best efficiency is achieved by the system with the heterostructure emitter and an uncovered InSb cell (reaching to $27\%$ of the Carnot efficiency). These results show that the performances of near-field thermophotovoltaic systems can be comparable with or even superior than the state-of-art thermoelectric devices. The underlying physics for the significant enhancement of the thermophotovoltaic performance is understood as due to the resonant coupling between the emitter and the cell, where the surface plasmons in graphene and surface phonon-polaritons in boron-nitride play important roles. Our study provides a stepping stone for future high-performance thermophotovoltaic systems.	physics.app-ph	physics	Enhancing thermophotovoltaic performance using graphene-BN-InSb near-field heterostructures Rongqian Wang,1 Jincheng Lu,1 and Jian-Hua Jiang1, 2, ∗ 1School of Physical Science and Technology & Collaborative Innovation Center of Suzhou Nano Science and Technology, Soochow University, Suzhou 215006, China 2Department of Physics, University of Toronto, 60 St. George Street, Toronto, Ontario, Canada M5S 1A7 (Dated: February 28, 2019) Graphene -- hexagonal-boron-nitride -- InSb near-field structures are designed and optimized to enhance the output power and energy efficiency of the thermophotovoltaic systems working in the temperature range of common industrial waste heat, 400 K ∼ 800 K, which is also the working tem- perature range for conventional thermoelectric devices. We show that the optimal output electric power can reach 3.5 × 104 W/m2 for the system with a graphene -- hexagonal-boron-nitride het- erostructure emitter and a graphene-covered InSb cell, whereas the best efficiency is achieved by the system with the heterostructure emitter and an uncovered InSb cell (reaching to 27% of the Carnot efficiency). These results show that the performances of near-field thermophotovoltaic systems can be comparable with or even superior than the state-of-art thermoelectric devices. The underlying physics for the significant enhancement of the thermophotovoltaic performance is understood as due to the resonant coupling between the emitter and the cell, where the surface plasmons in graphene and surface phonon-polaritons in boron-nitride play important roles. Our study provides a stepping stone for future high-performance thermophotovoltaic systems. I. INTRODUCTION Thermophotovoltaic (TPV) cells, as an emergent breed of clean energy resource, have attracted a wide range of research interest due to their potential high- performances1 -- 14. In general, a TPV system consists of a thermal radiation emitter and an infrared photo- voltaic cell which converts thermal radiations into elec- tric power. The emitter can be heated either by the sun or by industrial waste heat, giving rise to different ap- plications. Far-field thermal radiation is limited by the Stefan-Boltzmann law, results in reduced output power, particularly when the temperature of the emitter is below 1000 K. Most of the existing studies are focused on the regime where the emitter has very high temperature, e.g., Temit > 1000 K, which corresponds to solar radiations or heat radiations from a secondary thermal emitter which receives the solar radiation energy. In this work, we focus on the situations where the tem- perature of the thermal emitter is in the range of com- mon industrial waste heat, 400 K < Temit < 800 K. At such a temperature range, thermal radiation is strongly limited by the Stefan-Boltzmann law, leading to sup- pressed output power and energy efficiency for the TPV systems. One can turn to the near-field effect15 -- 22 to enhance thermal radiation flux, which may enable ap- pealing energy efficiency and output power. However, the near-field effect becomes ineffective if the tunneling through the vacuum gap is inefficient for photons that carry most of the heat energy, i.e., photons of angular frequencies ω ∼ kBTemit/. One way to overcome this ob- stacle is to exploit surface-polariton enhanced near-field radiations23 -- 27. Graphene and hexagonal-boron-nitride (h-BN) provide surface plasmon polaritons (SPPs) and surface phonon polaritons (SPhPs)28 -- 31 due to strong light-matter interactions32. These surface polaritons are right in the mid infrared frequency range that fits the peak thermal radiation frequencies of emitters in the tem- perature range of 400 K < Temit < 800 K. The narrow band gap indium antimonide (InSb) p-n junctions can be used as the TPV cells to convert such mid infrared thermal radiation into electricity. However, there is still considerable mismatch between the optical property of InSb and that of graphene or h-BN, which significantly reduces the near-field radiation. It has been shown that a sheet of graphene covered on the InSb p-n junction can strongly improve the coupling between the emitter and the TPV cell and thus enhance the near-field ther- mal radiation24. Here, we propose to use hBN-graphene heterostructures33 -- 38 as the emitter and the graphene- covered InSb p-n junction as the TPV cell. We find that such a design leads to significantly improved perfor- mance as compared to the existing studies24,39,40. The main difference between our near-field TPV (NTPV) sys- tems and that in Ref. [24] is that the thermal emitter in our system is made of h-BN -- graphene heterostructures, making our NTPV systems more powerful and efficient than the NTPV systems considered in Ref. [24] where the emitter is made of bulk h-BN. In this paper, we examine the performances of four different NTPV configurations: (i) h-BN-InSb cell (de- noted as hBN-InSb, with the mono-structure bulk h- BN being the emitter and the uncovered InSb junc- tion being the cell), (ii) h-BN-graphene/InSb cell (de- noted as hBN-G/InSb, with the bulk h-BN being the emitter and the graphene-covered InSb junction as the cell), (iii) h-BN/graphene-InSb cell (denoted as fhBN/G- InSb, with the h-BN/graphene heterostructure film be- ing the emitter and the uncovered InSb junction as the cell), and (iv) h-BN/graphene-graphene/InSb (de- noted as fhBN/G-G/InSb, with the h-BN/graphene het- erostructure film being the emitter and the graphene- covered InSb junction as the cell). We study and com- pare their performances for various conditions to opti- mize the performance of the NTPV system and to re- veal the underlying physical mechanisms toward high- performance NTPV systems working at the temperature range of 400 K < Temit < 800 K. Although it has been shown that a single graphene sheet can act as an excellent thermal emitter23,41 which gives high power density for the NTPV system. However, such a structure is techno- logically challenging and the induced energy efficiency is considerably lower than the NTPV systems considered in this work, despite that the power density is comparable with our NTPV systems. We remark that our findings are consistent with the study in Ref. [42] where the synergy between near-field thermal radiation and inelastic thermoelectricity43 -- 48 is shown to have considerably improved performance even in the linear-transport regime. On the other hand, our work is also based on the previous studies where the near-field effects are shown to improve the heat radiation flux by orders-of-magnitude by using infrared hyperbolic metamaterials33 -- 38. This paper is structured as follows. In Sec. II, we de- scribe our near-field TPV system, introduce the optical properties of the emitter and the InSb cell. We clar- ify the radiative heat flux exchanged between the emit- ter and the cell. We also recall the basic formulations of the photo-induced current, electric power and energy efficiency of the NTPV cell. In Sec. III, we study the performances of the four different NTPV systems at vari- 2 FIG. 1: (Color online) Schematic representation of the NTPV system. A thermal emitter of temperature Temit made of h- BN/graphene heterostructure is placed in the proximity of a thermophotovoltaic cell of temperature Tcell made of InSb. The emitter-cell distance is kept at d. The red arrows repre- sent the heat flux radiated from the emitter to the cell. The coordinate axes on the left side shows the in-plane (parallel to the x − y plane) and out-of plane (perpendicular to the x − y plane) directions. ous conditions to search for high-performance NTPV sys- tems. Finally, we summarize and conclude in Sec. IV. II. SYSTEM AND MODEL A. Near-field thermal radiation The proposed NTPV system is presented in Fig. 1. The emitter is a graphene covered h-BN film of thickness h, kept at temperature Temit. The TPV cell is made of an InSb p-n junction, kept at temperature Tcell, which is also covered by a layer of graphene. The thermal radiation from the emitter is absorbed by the cell and converted into electricity via infrared photoelectric conversion. In this system, plasmons in graphene can interact strongly with infrared photons and give rise to SPPs. Mean- while, phonons in h-BN can also couple strongly with photons and leads to SPhPs28,29,31,33 -- 35,49. These emer- gent quasi-particles due to strong light-matter interaction can dramatically enhance thermal radiation, particularly via the near-field effect33 -- 38. Because of their coincident frequency ranges, the SPPs and SPhPs hybrid with each other when graphene is placed together with h-BN, lead- ing to hybrid polaritons called surface plasmon-phonon polaritons (SPPPs)33,34. SPPPs have been shown to be useful in improving near-field heat transfer between two h-BN/graphene heterostructures36 -- 38. The radiative 𝑇cell𝑇emitd𝑦𝑥𝑧(cid:33) (cid:32) heat transfer is optimized when the emitter and absorber are made of the same material, which leads to a resonant energy exchange between the emitter and the absorber50. In order to enable such resonant energy exchange in our NTPV cell, we add another layer of graphene onto the InSb cell, as shown in Fig. 1. The propagation of electromagnetic waves in these ma- terials is described by the Maxwell equations, in which the dielectric functions of these materials are the key fac- tors that determine the propagation of the electromag- netic waves. The dielectric function of h-BN is described by a Drude-Lorentz model, which is given by34 εm = ε∞,m 1 + LO,m − ω2 ω2 TO,m − iγmω − ω2 ω2 TO,m , (1) where m = (cid:107),⊥ denotes the out-of-plane and the in-plane directions, respectively. The in-(out-of-)plane direction is determined by whether the electric field is perpendicular (parallel) to the optical axis of the h-BN film (h-BN is a uniaxial crystal, and the optical axis is in the z direction, as defined in Fig. 1). ε∞,m is the high-frequency relative permittivity, ωTO and ωLO are the transverse and lon- gitudinal optical phonon frequencies, respectively. γm is the damping constant of the optical phonon modes. The values of these parameters are chosen as those deter- mined by experiments30,51, ε∞,⊥ = 4.87, ωTO,⊥ = 1370 cm−1, ωLO,⊥ = 1610 cm−1 and γ⊥ = 5 cm−1 for in- plane phonon modes, ε∞,(cid:107) = 2.95, ωTO,(cid:107) = 780 cm−1, ωLO,(cid:107) = 830 cm−1 and γ(cid:107) = 4 cm−1 for out-of-plane phonon modes. There are certain frequency ranges where the in-plane and out-of-plane dielectric functions of h-BN have opposite signs, leading to the hyperbolicity of isofre- quency contour, which makes h-BN a natural hyperbolic metamaterial34. The optical property of graphene sheet is described here by a modeled conductivity, σg, which sums over both the intraband σD and interband σI contributions, i.e., σg = σD + σI, given respectively by52 (cid:18) µg (cid:19)(cid:21) 2kBT ln 2 cosh , (2) σD = and σI = e2 4 [G i ω + i τ (cid:18)ω (cid:19) 2 where G(ξ) = cosh (cid:20) (cid:90) ∞ (cid:17) (cid:16) µg 0 kB T 2e2kBT π2 4ω (cid:16) ξ π (cid:17) kB T +cosh + i (cid:16) ξ sinh kB T ω 2 ) G(ξ) − G( (ω)2 − 4ξ2 dξ], (cid:17) is a dimensionless (3) function. µg is the chemical potential of the graphene sheet, which can be readily tuned via a gate-voltage53. 3 τ is the electron scattering time and chosen as 100 fs in our calculations35. T is the temperature of the graphene sheet. e, kB and are the electron charge, Boltzmann constant and the reduced Planck constant, separately. The dielectric function of the InSb cell is given by24 (cid:18) (cid:19)2 icα(ω) 2ω ε = n + , (4) (cid:90) (cid:88) ωgap (cid:113) ω where n = 4.12 is the refractive index and c is the speed of light in vacuum. α(ω) is a step-like function which describes the absorption probability of the incident pho- tons, and it is given by α(ω) = 0 for ω < ωgap and − 1 for ω > ωgap with α0 = 0.7 µm−1 α(ω) = α0 being the absorption coefficient and ωgap = Egap being the angular frequency corresponding to the band gap of InSb. The gap energy Egap of InSb is temperature depen- dent, which is given by Egap = 0.24 − 6 × 10−4 T 2 54. For instance, for Tcell = 320 K, the gap energy Egap is nearly 0.17 eV and the corresponding frequency is about 2.5 × 1014 Hz. T +500 The near-field radiative heat flux exchanged between the emitter and the cell is given by 55,56, δΘ(Temit, Tcell, ω, ∆µ) Prad = 4π2 j δΘ (Temit, Tcell, ω, ∆µ) where Θ2 (Tcell, ω, ∆µ), with Θ1 (Temit, ω) = = and Θ2 (Tcell, ω, ∆µ) = exp (cid:16) ω−∆µ ω kB Tcell (cid:17)−1 kdkζj (ω, k) , (5) Θ1 (Temit, ω) − (cid:17)−1 ω ω (cid:16) exp kB Temit being the Planck mean oscillator energy of blackbody at temperature Temit and Tcell, respectively. ∆µ is the electrochemical potential difference across the p-n junction in the TPV cell. k denotes the magnitude of the in-plane wavevector of thermal radiation waves and ζj(ω, k) is called the photon transmission coefficient for the j-polarization (j = s, p), which represents the probability for a photon with angular frequency ω and wavevector k to tunnel through the vacuum gap from the emitter to the cell. The transmission coefficient for each polarization sums over the contribution from both propagating and evanescent waves21, where kz = wavevector in vacuum. rj c2 − k2 is the normal component of the cell (j = s, p) are the emit and rj ζj(ω, k) =  (1−remit2)(1−rcell2) (cid:113) ω2 1−rj emitrj 4Im(rj emit)Im(rj 1−rj cell exp(2ikzd) , cell exp(2ikzd) emitrj cell) exp(2ikzd) k < ω c , k > ω c (6) reflection coefficients of the emitter and the cell, respec- tively, which measure the amplitude ratio between the re- flected and incident waves at the interface of the emitter and the cell, respectively. These coefficients can be calcu- lated by the scattering matrix approach where the optical interference in the heterostructure film is also taken into account57, 4 3 = ε (cid:107) 1 = ε⊥ tensor. For isotropic medium like vacuum and InSb, (cid:107) (cid:107) ε⊥ 3 = 1, and ε⊥ 1 = ε cell, where the dielectric function of the InSb cell is defined before. ε0 and µ0 are the permittivity and permeability for vacuum, respectively. cell = ε rs emit = 12 + (1 + rs rs 1 − rs 12 + rs 21rs 23 exp(2iks k,2h) 21) rs 23 exp(2iks z,2h) rs cell = z − ks ks z + ks ks z,cell − µ0σgω z,cell + µ0σgω for s-polarization. And rp emit = 12 + (1 − rp rp 1 − rp 12 − rp 21rp 21) rp 23 exp(2ikp z,2h) 23 exp(2ikp k,2h) rp cell = ε⊥ cellkp ε⊥ cellkp z − kp z + kp z,cell + z,cell + σgkp z,cell σgkp z,cell z kp ε0ω z kp ε0ω , (7) (8) , (9) (10) B. Characterizations of TPV cells When the TPV cell is located at a distance d which is on the order of or smaller than the thermal wavelength λth = 2πc/kBTemit from the emitter, the radiative en- ergy transfer will be enhanced due to the presence of evanescent waves23, which exists only in the vicinity of the emitter, because of the exponential decay from the interface. The enhanced radiation with energy greater than the band gap Egap will be absorbed by the TPV cell, creating electron-hole pairs and leading to electric current flow and output power generation. The induced electric current density can be written as1,58 Ie = Iph − I0[exp(V /Vcell) − 1], (13) for p-polarization. Where 1, 2, and 3 are the indexes for the vacuum region above h-BN film, the h-BN film region, and the vacuum region below h-BN film,respectively, as defined in Ref. [36]. rs,p is the reflection coefficient be- ab tween media a and b (a, b = 1, 2, 3) for either s or p polarization, which is given by rs ab = rp ab = z,a − kz,b − σgωµ0 ks ks z,a + kz,b + σgωµ0 z,aε⊥ a + kp kp z,aε⊥ a + kp kp b − kz,bε⊥ b + kz,bε⊥ , z,akz,b z,akz,b (11) σg ωε0 σg ωε0 , when there is a graphene layer in between media a and b, rs ab = rp ab = , z,a − kz,b ks ks z,a + kz,b z,aε⊥ kp z,aε⊥ kp , a b − kz,bε⊥ b + kz,bε⊥ (cid:113) in a (12) (cid:114) dia a and b. when there is no Here ks graphene z,i = c2 − k2 and k2 (i = a, b, cell) are the z-component between me- ε⊥ ω2 i ε⊥ c2 − ε⊥ i ω2 i (cid:107) ε i kp z,i = wavevectors for media i for s and p polarization, re- (cid:107) ε⊥ spectively. i and ε i (i = a, b, cell) are the in-plane and out-of plane components of the relative dielectric e where V = ∆µ/e is the voltage bias across the TPV cell, Vcell = kBTcell is a voltage which measures the tempera- ture of the cell1. Iph and I0 are called the photo-induced current density and reverse saturation current density, re- spectively. The reverse saturation current density (also named the dark current density, the electric current with- out light) is determined by the diffusion of minority car- riers in the InSb p-n junction, which is given by (cid:32) (cid:114) De τe 1 NA (cid:33) (cid:114) Dh τh + 1 ND I0 = en2 i , (14) √ 3 2 2kBTcell NcNv exp(− Egap ) (cm−3) is the intrinsic where ni = carrier concentration, which is dependent on the temper- ature of the TPV cell54. Nc and Nv are temperature- dependent effective density of states in the conduction and valance bands, respectively given by Nc = 8× 1012× cell (cm−3) and Nv = 1.4 × 1015 × T cell (cm−3). NA T and ND are the p-region and n-region impurity con- centrations, respectively. Numerical values are set as NA = ND = 1019 (cm−3). De and Dh are the diffusion coefficients of the electrons and holes, taken as De = 186 cm2/s and Dh = 5.21 cm2/s41. τe and τh are the relax- ation times of the electron-hole pairs in the n-region and p-region, correspondingly. Values for relaxation times are calculated by τe = τh = (5×10−26 cm−6s−1)n2 54. 3 2 1 i Photo-induced current is the electric current induced by the motion of photo-carriers. The photo-induced cur- rent density Iph is given by59 Iph = e Prad (Temit, Tcell, ω, ∆µ) ω dω, (15) (cid:90) ∞ ωgap where the radiative spectral heat flux is defined in Sec. II A. Here we assume that all incident photons are ab- sorbed, and each photon with an energy greater than the band gap creates one electron-hole pair, i.e., we assume 100% quantum efficiency24. Under this assumption, we only consider the energy efficiency, which measures how much incident radiative heat can be transferred into elec- tricity. We keep this assumption which is widely adopted in the literature. The output electric power density Pe is defined as the product of the net electric current density and the voltage bias, Pe = −IeV, (16) and the energy efficiency η is given by the ratio between the output electric power density Pe and incident radia- tive heat flux Qinc, η = Pe Qinc , (17) where the incident radiative heat flux is given by Qinc = dω 4π2 Θ1 (Temit, ω) kdkζj(ω, k) (cid:90) ∞ (cid:90) ∞ (cid:90) ωgap ωgap 0 − − 0 (cid:90) (cid:88) j (cid:90) (cid:88) (cid:90) (cid:88) j dω 4π2 Θ2 (Tcell, ω, ∆µ) dω 4π2 Θ1 (Tcell, ω) kdkζj(ω, k) kdkζj(ω, k), j (18) III. RESULTS AND DISCUSSIONS A. Performance of the designed NTPV systems We first examine the electric power and the normal- ized energy efficiency in units of the Carnot efficiency (ηc) for our proposed near-field TPV cells. Fig. 2 shows the performances for the four different configura- tions, respectively denoted as hBN-InSb, hBN-G/InSb, fhBN/G-InSb, and fhBN/G-G/InSb. The electric power density and efficiency are calculated through Eq. (16) 5 FIG. 2: (Color online) Performances of four NTPV cells. (a) Electric power density Pe, (b) energy efficiency in units of Carnot efficiency (η/ηc), (c) electric current density Ie and (d) incident radiative heat flux Qinc. The temperatures of the emitter and the cell are kept at Temit = 450 K and Tcell = 320 K, respectively. The vacuum gap is d = 22 nm, the thickness of h-BN is set as hbulk = 10000 nm for bulk one, and hfilm = 20 nm for h-BN thin film. The chemical potential of graphene is µg = 0.37 eV. All parameters are same for the four configurations. The Carnot efficiency is given by ηc = 1 − Tcell . Temit and (17), respectively. It is noted that by adding a single layer of graphene either on the emitter or the TPV cell, the output electric power density is consid- erably enhanced. Specifically, the enhancement factors of maximum electric power for fhBN/G-InSb cell (blue solid line) and hBN-G/InSb cell (green solid line) are re- spectively given by P hBN−G/InSb and P fhBN/G−InSb . This enhancement in- duced by graphene (due to graphene SPPs) has been found in Ref. [24]. Notably, when the emitter and cell are both covered by graphene, the maximum electric power density is significantly enhanced, with the enhancement factor P fhBN/G−G/InSb = 3.2P hBN−InSb = 1.2P hBN−InSb = 5.1P hBN−InSb max max max max . max max As indicated in Fig. 2(b), the energy efficiency also shows strong enhancement for hBN-G/InSb and fhBN/G-InSb cells, with the enhancement of the max- imum efficiency ηhBN−G/InSb and ηfhBN/G−InSb = 2.7ηhBN−InSb , respectively. While for max fhBN/G-G/InSb cell, contrary to the considerable en- hancement in output electric power, the energy efficiency for fhBN/G-G/InSb cell is substantially reduced. = 1.1ηhBN−InSb max max max (a)(b)(c)(d)𝐼sh𝑉oc6 FIG. 3: (Color online) (a) Dielectric function ε of h-BN, (b) conductivity of graphene σg, (c) photo-induced current spectrums Iph (ω) and (d) incident radiative heat spectrums Qinc (ω) for the four configurations. The parameters are same with those in Fig.2. The two shaded areas represent type-I and type-II hyperbolic regions of h-BN, respectively. sh sh sh sh sh sh = 2.0I hBN−InSb and I fhBN/G−G/InSb As shown in Fig. 2(c), the enhancement factors of short-circuit current densities Ish (when the voltage bias is zero) for hBN-G/InSb, fhBN/G-InSb and fhBN/G- G/InSb cells are about I hBN−G/InSb = 1.1I hBN−InSb , I fhBN/G−InSb = 2.8I hBN−InSb , respectively. Besides, the enhancement factors of open-circuit voltage Voc (when the electric current is zero) for the three covered cells are given by V hBN−G/InSb = 1.6V hBN−InSb , re- spectively. The product of the short-circuit current Ish and the open-circuit voltage gives a good estimation of the output power. Therefore, the enhancements of the short-circuit current and the open-circuit voltage lead to the significant improvement of the output power density shown in Fig. 2(a). = 1.1V hBN−InSb and V fhBN/G−G/InSb , V fhBN/G−InSb = 1.8V hBN−InSb oc oc oc oc oc oc The different manifestation of energy efficiency for graphene-covered near-field systems can be interpreted by the competition between the electric power out- inc inc inc inc = 1.1QhBN−InSb = 1.3QhBN−InSb put and the radiative heat consumption. As shown in Fig. 2(d), the incident radiative heat flux (given by Eq. (18)) is increased QhBN−G/InSb , and QfhBN/G−InSb for the hBN-G/InSb cell and fhBN/G-InSb cell, respectively. The enhancement of the output electric power exceeds the increase of the radiative heat consumption, leading to the improvement of energy efficiency. While for the fhBN/G-G/InSb cell, the radiative heat consumption is significantly increased QfhBN/G−G/InSb , much higher than the improvement of the electric power output, leading to sup- pressed energy efficiency. = 18QhBN−InSb inc inc To figure out the microscopic mechanisms for the en- hancement of the output power, we study the spectral distributions of the photo-induced current Iph (ω) with the help of the dielectric function of h-BN and the op- tical conductivity of graphene. As shown in Fig. 3(a), two hyperbolic regions (labeled with type-I and type-II) are marked with gray shades. For type-I hyperbolicity, (a)(b)Real(𝜎g)Imag(𝜎g)(c)(d)21127 FIG. 4: (Color online) Radiative emission spectrums Prad (ω) for (a) hBN-InSb cell, (b) hBN-G/InSb cell, (c) fhBN/G-InSb cell and (d) fhBN/G-G/InSb cell. The parameters are same with those in Fig.2. the corresponding frequency range is from 1.5 × 1014 Hz (corresponding to the frequency of transverse phonon vi- brations, ωTO,⊥ = 780 cm−1) to 1.6 × 1014 Hz (corre- sponding to the frequency of longitudinal phonon vibra- tions, ωLO,⊥ = 830 cm−1). For type-II hyperbolicity, the corresponding frequency range is from ωTO,⊥ = 1370 cm−1) to 3.1 × 1014 Hz (corresponding to the frequency of longitudinal phonon vibrations, ωLO,⊥ = 1610 cm−1). The sharp peaks shown in these two regions indicate the existence of SPhPs51,60. While in Fig. 3(b), the optical conductivities of graphene at various chemical potentials show the broadband nature of the graphene SPPs61. The photo-induced current spectrum Iph (ω) is given by Eq. (15) but integrated over k only. As shown in Fig. 3(c), for hBN-InSb cell (black solid curve) the photo- induced current spectrum with a broad resonant peak is shown in the type-II hyperbolic regime (marked by the gray shade), which comes from the SPhPs. In hBN- G/InSb cell, the graphene sheet on InSb cell induces a resonant absorption between the emitter and the cell, de- picted by the higher peak (blue solid curve), with about 20% increase compared with the hBN-InSb cell. Since near-field radiation is dominated by evanescent wave cou- plings between the emitter and the absorber, the resonant near-field absorption peak can be understood as due to the resonant coupling between the h-BN emitter and the InSb absorber. The resonant near-field absorption peak gives the major contribution to the photo-induced cur- rent and the output electric power. Interestingly, there are two resonant peaks in the photo-induced current spectrum of the fhBN/G-InSb cell, separately labeled as 1 and 2 in the Fig. 3(c). And the corresponding frequencies are given by ω1 = 2.6 × 1014 Hz and ω2 = 3.1 × 1014 Hz, respectively. The low-frequency resonance and the broad high-absorption band in the type-II hyperbolic regime, are due to the hy- brid polaritons named as the hyperbolic plasmon-phonon polaritons (HPPPs), which are resulted from the coupling between the SPPs in graphene and the SPhPs in h-BN in the type-II hyperbolic region28. The high-frequency (a)(b)(c)(d)absorption resonance outside the type-II region is due to the surface plasmon-phonon polaritons (SPPPs), which come from the strong coupling between the broadband graphene plasmons and the optical phonons outside the hyperbolic regions (i.e., regions without SPhPs)34,62. As a consequence, the photo-induced current is much larger and the bandwidth is much broader, which give rise to the major contribution in the photo-induced carrier genera- tion and the electric power output. Further enhancement of the photo-induced carrier generation can be realized by adding a monolayer graphene to the InSb cell, as shown in Fig. 2(c). The incident radiative heat flux Qinc (ω) is examined in Fig. 3(d). The overall trends of Qinc (ω) are the same with the photo-induced current spectrums shown in Fig. 3(c). However, the incident radiative heat flux also includes the contributions from the below-band-gap near field heat transfer, which is quite different for the four configurations. In order to understand the below-band-gap near field heat transfer, we present the near-field heat flux in a very large frequency range for both the emitter and the ab- sorber in Fig. 4. The difference between the heat fluxes flowing out of the emitter and the absorber gives the ra- diative heat consumption, which consists of the useful part (above the InSb band gap) and the wasted part (be- low the InSb band gap). The band gap frequency ωgap of the InSb cell is marked by the black dashed line. Comparing between Fig. 4(a) and Fig. 4(b), one can see that adding a graphene layer to the InSb cell increases both the useful and the wasted parts of the radiative heat flux, leading to slightly improved output power and energy efficiency. The best energy efficiency improve- ment comes from the configuration with a graphene layer added to the h-BN emitter. As shown in Fig. 4(c), for this configuration the useful part of the radiative heat flux is significantly increased, while the wasted heat flux is re- duced. Therefore, the fhBN/G-InSb cell yields a signifi- cantly improved energy efficiency and output power. For the configuration with fhBN/G emitter and graphene- covered InSb junction, as shown in Fig. 4(d), the wasted radiative heat flux is dramatically increased. Although this configuration has the largest output power, the en- ergy efficiency is the lowest among those four configura- tions. We further elaborate on the tunnel coupling between the emitter and the absorber by studying the photon tun- neling probabilities ζ (ω, k) (given by Eq. (6)) for the four 8 FIG. 5: (Color online) Photon tunneling probabilities ζ (ω, k) for (a) hBN-InSb cell, (b) hBN-G/InSb cell, (c) fhBN/G-InSb cell and (d) fhBN/G-G/InSb cell. The parameters are same with those in Fig. 2. Note that the horizontal axis ranges are different for these figures. configurations as shown in Fig. 5 for photons with fre- quency higher than the InSb band gap frequency, ωgap. The bright bands shown in Fig. 5 represent near 100% photon tunneling probability, which originates from the strong coupling modes due to various surface polaritons. The angular frequency and wavevector of these strong coupling modes satisfy the efficient tunneling condition, i.e., the denominator of Eq. (6) is minimized. It is seen from Figs. 5(a) and 5(b) that with the bulk h-BN being the emitter, the high photon tunneling band is mainly provided by the Reststrahlen band of type-II hyperbolic phonon polaritons (the region between the two white dot- ted lines) with low wavevector. After integration over the wavevector k, this narrow light band can only con- tribute to a small photo-induced current and limited output power. Adding a monolayer of graphene onto the InSb cell does not considerably improve the photo- induced current and the TPV performance. In contrast, when the h-BN/graphene heterostructure film serves as the emitter, the high photon tunneling band is no longer limited to the small wavevector region in the Reststrahlen band of type-II hyperbolic phonon polaritons. High pho- ton tunneling extends to very large wavevector (the deep- subwavelength evanescent wave regime) and goes beyond the Reststrahlen band, as shown in Figs. 5(c) and 5(d), which leads to significantly increased phase space for high photon tunneling and hence considerably improved out- put power. (a) hBN-InSb(c) fhBN/G−InSb(b) hBN-G/InSb(d) fhBN/G−G/InSb9 FIG. 6: (Color online) (a) Band gap frequency ωgap of the InSb cell, (b) reverse saturation current density I0, (c) open- circuit voltage Voc and (d) absorption fractions of incident radiation P ω>ωgap /Pinc as functions of the cell temperature Tcell. The temperature of the emitter is kept at Temit = 450 K. The vacuum gap is d = 22 nm, the thickness of h-BN is set as hbulk = 10000 nm for bulk one, and hfilm = 20 nm for h-BN thin film. The chemical potential of graphene is µg = 0.37 eV. inc FIG. 7: (Color online) Effect of the chemical potential of graphene on the photo-induced current spectrums for hBN- G/InSb cell, fhBN/G-InSb cell and fhBN/G-G/InSb cell at (a) µg = 0.2 eV, (b) µg = 0.4 eV, (c) µg = 0.6 eV and (d) µg = 1.0 eV. The temperatures of the emitter and the cell are set at Temit = 450 K and Tcell = 320 K, respectively. The vacuum gap is d = 22 nm, the thickness of h-BN is set as hbulk = 10000 nm for bulk one, and hfilm = 20 nm for h-BN thin film. To understand the performances of the four NTPV configurations under different working temperatures, we study the band gap frequency, the reverse saturation cur- rent, the open-circuit voltage, and the absorption fraction of the incident thermal radiation as functions of the InSb junction temperature Tcell. As presented in Fig. 6(a), the InSb band gap frequency ωgap (orange solid curve) decreases from 2.6 × 1014 Hz to 2.0 × 1014 Hz when Tcell increases from 300 K to 400 K. Naively, one would ex- pect that smaller band gap gives larger photon absorp- tion flux and leads to better performances. Indeed, the photo-induced current increases considerably. However, the reverse saturation current I0 increases exponentially with the cell temperature Tcell because of the reduction of the InSb band gap. As a consequence, the open-circuit voltages (Voc) are subsequently decreased, since Voc is given by Voc = Vcell log (Iph/I0 + 1). The total outcome does not necessarily yield better performance. Another important factor that dominates the energy efficiency is the absorption fraction, which is defined as the ratio of the radiative heat flux carried by photons with frequency higher than ωgap to the total absorbed heat flux, i.e., P ω>ωgap /Pinc. The absorption fraction inc measures how much fraction of the absorbed heat flux is useful. It is seen from Fig. 6(d) that the absorption frac- tions for the four configurations are all increased with the temperature Tcell. Higher absorption fraction often leads to higher energy efficiency. It is reasonable to set the InSb cell temperature to 320 K which balances many different aspects. The corresponding band gap is 0.17 eV and the gap frequency is 2.5×1014 Hz. At such a working temper- ature, the reverse saturation current density remains low, I0 (cid:39) 1.6 × 104 A. Meanwhile, one can make good use of the incident photons, with the absorption fractions 0.80, 0.32, 0.31 and 0.043 for the hBN-InSb cell, hBN-G/InSb cell, fhBN/G-InSb cell, and fhBN/G-G/InSb cell, respec- tively. Under these circumstances, the TPV cells can make the most of the incident radiative heat and realize good performances, as shown in Fig. 2. The best NTPV system should have resonant photon absorption right above the band gap42, which needs a fine matching between the emitter and the absorber. To the best of our knowledge, the fhBN/G-InSb configura- tion has the best thermophotovoltaic performances due to such matching. (a)(b)(c)(d)(a)SPPPsHPPPs(b)SPPPsHPPPs(c)SPPPsHPPPs(d)SPPPsHPPPs10 FIG. 8: (Color online) Effect of the chemical potential of graphene on the performances for hBN-G/InSb cell, fhBN/G- InSb cell and fhBN/G-G/InSb cell at (a) µg = 0.2 eV, (b) µg = 0.4 eV, (c) µg = 0.6 eV and (d) µg = 1.0 eV. Other parameters are same with those in Fig. 7. The near-field heat transfer can be strongly modi- fied by the chemical potential of the graphene layer, µg. In Fig. 7, we plot the photo-induced current spectrums of the four different NTPV configurations with differ- ent µg. For the hBN-G/InSb cell, with large chemi- cal potentials, the resonant peak in the HPPPs region of the photo-induced current spectrum is enhanced and split into small side peaks due to the resonant coupling with graphene plasmons. For the large chemical poten- tial µg = 1.0 eV, the near field coupling between the SPhPs and the graphene plasmons is significantly en- hanced. These behaviors are due to the dependence of the graphene plasmon frequencies on the chemical po- tential, as demonstrated in Refs. 28, 31 and 24. The enhancement of photo-induced current spectrum by tun- ing µg provides an effective way to further enhance the performances of the graphene-based NTPV systems. B. Optimization The output power and energy efficiency for the three configurations with graphene at four different chemical potentials are shown in Fig. 8. The output powers and energy efficiencies for these three configurations are all FIG. 9: (Color online) Effect of the vacuum gap d on the performances for (a) hBN-InSb cell, (b) hBN-G/InSb cell, (c) fhBN/G-InSb cell and (d) fhBN/G-G/InSb cell. The temper- atures of the emitter and the cell are set at Temit = 450 K and Tcell = 320 K. The thickness of h-BN is set as hbulk = 10000 nm for bulk one, and hfilm = 20 nm for h-BN thin film. The chemical potential of graphene is µg = 0.37 eV. improved when the chemical potential of graphene is in- creased. Especially for the fhBN/G-G/InSb cell, the maximum output power density at µg = 1.0 eV is in- creased to nearly 6 times of that at µg = 0.2 eV. Mean- while, the corresponding maximum efficiency is enhanced to 8.3 times of that at µg = 0.2 eV. For the hBN-G/InSb cell, the maximum output power density at µg = 1.0 eV is enhanced to about 12 times of that at µg = 0.2 eV, while the the corresponding maximum efficiency is enhanced to 1.6 times of that at µg = 0.2 eV. Also for the fhBN/G- InSb cell, the maximum output power density and maxi- mum efficiency at higher chemical potential are enhanced to 2.5 and 1.4 times of those at µg = 0.2 eV, respectively. Such significant improvements of the TPV performances of the fhBN/G-InSb, hBN-G/InSb and fhBN/G-G/InSb cells are in accordance with the strongly enhanced photo- induced current spectrums shown in Fig. 7. We also study the influence of the vacuum gap d and the temperature of the emitter Temit on the performances of four configurations, as shown in Figs. 9 and 10, re- spectively. The electric power density and the normal- ized energy efficiency are remarkably enhanced when the vacuum gap is reduced to 20 nm or when the emitter (a) (b) (c) (d) (a) hBN−InSb(b) hBN−G/InSb(c) fhBN/G−InSb(d) fhBN/G−G/InSb11 fhBN/G-G/InSb cell, where the SPhPs in h-BN and graphene plasmons as well as their coupling play vi- able roles in enhancing the energy efficiency and power of these NTPV systems. It is found that the optimal output electric power with 3.5 × 104 W/m2 is achieved in hBN/graphene-graphene/InSb cell. While the opti- mal efficiency, 27% of the Carnot efficiency, is reached in h-BN/graphene-InSb cell. The performance of the h-BN/graphene based cell can be further improved by tuning the chemical potential of graphene. Combining with the fact of the experimental availability of the h- BN/graphene heterostructure and the state-of-art dop- ing of graphene63,64, h-BN/graphene based cell can be useful for future thermophotovoltaic systems with high performances. Remarkably, using such a graphene-BN-InSb near-field heterostructure design, we show that the performance of our NTPV systems can be comparable with the state- of-art thermoelectric systems working in the same tem- perature range. For instance, the state-of-art output power density of thermoelectric generators is realized in Ref.[ 65] with a power factor (PF) P F = 2.5× 105 W/m2 for a device working between two baths with tempera- tures Th = 873 K and Tc = 330 K, whereas for the same conditions our NTPV device gives P F = 4.0×104 W/m2. The state-of-art device efficiency of thermoelectric gen- erators is given in Ref.[ 66] where η = 27%ηc for a ther- moelectric generator working between two thermal reser- voirs with temperatures Th = 773 K and Tc = 300 K. The corresponding device ZT is ZdevT = 1.34. In com- parison, for the same temperatures, our NTPV device can also give η = 27%ηc. These comparisons indicate that the NTPV systems have a high potential for future thermal to electric energy conversion. V. ACKNOWLEDGMENTS R.W., J.L., and J.-H.J. acknowledge support from the National Natural Science Foundation of China (NSFC Grant No. 11675116), the Jiangsu distinguished profes- sor funding and a Project Funded by the Priority Aca- demic Program Development of Jiangsu Higher Educa- tion Institutions (PAPD). R.W. thanks Professor Chen Wang for discussions. FIG. 10: (Color online) Effect of the emitter temperature Temit on the performances for (a) hBN-InSb cell, (b) hBN- G/InSb cell (c) fhBN/G-InSb cell and (d) fhBN/G-G/InSb cell. The temperature of the cell is kept at Tcell = 320 K. The thickness of h-BN is set as hbulk = 10000 nm for bulk one, and hfilm = 20 nm for h-BN thin film. The vacuum gap is d = 20 nm, and the chemical potential of graphene is µg = 0.37 eV. temperature is increased to 800 K. These results demon- strate that the near-field heat transfer is crucial for the enhancement of both the energy efficiency and the output power. The optimal output power is achieved in the fhBN/G- G/InSb cell, with a power density of 3.5×104 W/m2. The optimal efficiency, 27% of the Carnot efficiency, is realized in the fhBN/G-InSb cell, while a moderate output power density 2.0×104 W/m2 is realized as well. Therefore, the best balance between the energy efficiency and the output power comes from the configuration with fhBN/G-InSb cell. The overall trend is that narrow vacuum gap, proper chemical potential of graphene, and high temperature of the emitter are favorable for high TPV performances. IV. CONCLUSIONS We investigate the energy efficiency and output power of four different NTPV systems, denoted as the hBN- InSb cell, hBN-G/InSb cell, fhBN/G-InSb cell, and the (a) hBN−InSb(b) hBN−G/InSb(c) fhBN/G−InSb(d) fhBN/G−G/InSb(b) hBN−G/InSb12 ∗ Electronic address: [email protected] [email protected] ,joe- 24 R. Messina and P. Ben-Abdallah, Sci. Rep. 3, 1383 (2013). 25 B. Wang, X. Zhang, X. Yuan, and J. Teng, Appl. Phys. 1 W. Shockley and H. J. Queisser, J. Appl. Phys. 32, 510 Lett. 100, 131111 (2012). (1961). 26 V. Svetovoy, P. Van Zwol, and J. Chevrier, Phys. Rev. B 2 R. M. Swanson, Thermophotovoltaic converter and cell for 85, 155418 (2012). use therein (1980), uS Patent 4,234,352. 27 V. Svetovoy and G. Palasantzas, Phys. Rev. Appl. 2, 3 L. M. Fraas, J. E. Samaras, P. F. Baldasaro, and E. J. Brown, Spectral control for thermophotovoltaic generators (1995), uS Patent 5,403,405. 4 S. Wojtczuk, E. Gagnon, L. Geoffroy, and T. Parodos (1995), vol. 321, pp. 177 -- 187. 5 R. P. Hamlen, Thermophotovoltaic generator (1996), uS Patent 5,560,783. 6 T. J. Coutts and M. C. Fitzgerald, Sci. Am. 279, 90 (1998). 7 D. Mart´ın and C. Algora, Semicond. Sci. Tech. 19, 1040 (2004). 8 T. Nagashima, K. Okumura, and M. Yamaguchi (2007), vol. 890, pp. 174 -- 181. 9 L. M. Fraas and L. G. Ferguson, Three-layer solid infrared emitter with spectral output matched to low bandgap ther- mophotovoltaic cells (2000), uS Patent 6,091,018. 10 O. Sulima and A. Bett, Sol. Energ. Mat. Sol. C. 66, 533 (2001). 11 C. Wu, B. Neuner III, J. John, A. Milder, B. Zollars, S. Savoy, and G. Shvets, J. Optics 14, 024005 (2012). 12 W. R. Chan, P. Bermel, R. C. Pilawa-Podgurski, C. H. Marton, K. F. Jensen, J. J. Senkevich, J. D. Joannopoulos, M. Soljaci´c, and I. Celanovic, P. Nat. Acad. Sci. 110, 5309 (2013). 13 T. Liao, L. Cai, Y. Zhao, and J. Chen, J. Power Sources 306, 666 (2016). 034006 (2014). 28 S. Dai, Z. Fei, Q. Ma, A. S. Rodin, M. Wagner, A. S. McLeod, M. K. Liu, W. Gannett, W. Regan, K. Watanabe, et al., Science 343, 1125 (2014). 29 X. G. Xu, B. G. Ghamsari, J.-H. Jiang, L. Gilburd, G. O. Andreev, C. Zhi, Y. Bando, D. Golberg, P. Berini, and G. C. Walker, Nature Commun. 5, 4782 (2014). 30 J. D. Caldwell, A. V. Kretinin, Y. Chen, V. Giannini, M. M. Fogler, Y. Francescato, C. T. Ellis, J. G. Tischler, C. R. Woods, A. J. Giles, et al., Nature Commun. 5, 5221 (2014). 31 X. G. Xu, J.-H. Jiang, L. Gilburd, R. G. Rensing, K. S. Burch, C. Zhi, Y. Bando, D. Golberg, and G. C. Walker, Acs Nano 8, 11305 (2014). 32 J.-H. Jiang and S. John, Phys. Rev. X 4, 031025 (2014). 33 V. W. Brar, M. S. Jang, M. Sherrott, S. Kim, J. J. Lopez, L. B. Kim, M. Choi, and H. Atwater, Nano Lett. 14, 3876 (2014). 34 A. Kumar, T. Low, K. H. Fung, P. Avouris, and N. X. Fang, Nano Lett. 15, 3172 (2015). 35 A. Woessner, M. B. Lundeberg, Y. Gao, A. Prin- cipi, P. Alonso-Gonz´alez, M. Carrega, K. Watanabe, T. Taniguchi, G. Vignale, M. Polini, et al., Nat. Materials 14, 421 (2015). 36 B. Zhao and Z. M. Zhang, J. Heat Transfer 139, 022701 14 E. Tervo, E. Bagherisereshki, and Z. M. Zhang, Front. En- (2015). ergy 12, 5 (2018). 37 B. Zhao, B. Guizal, Z. M. Zhang, S. Fan, and M. Antezza, 15 E. G. Cravalho, C. L. Tien, and R. Caren, J. Heat Transfer Phys. Rev. B 95, 245437 (2017). 89, 351 (1967). 16 J. B. Xu, K. Lauger, K. Dransfeld, and I. H. Wilson, Rev. 38 K. Shi, F. Bao, and S. He, ACS Photonics 4, 971 (2017). 39 O. S. Heavens, Optical properties of thin solid films Sci. Instrum. 65, 2262 (1994). (Courier Corporation, 1991). 17 J. Xu, B. Koslowski, R. Moller, K. Lauger, K. Dransfeld, and I. H. Wilson, J. Vac. Sci. Technol. B 12, 2156 (1994). 18 J. B. Xu, K. Lauger, R. Moller, K. Dransfeld, and I. H. 40 Z. Knittl, Optics of thin films: an optical multilayer theory, vol. 1 (Wiley London:, 1976). 41 M. Lim, S. Jin, S. S. Lee, and B. J. Lee, Opt. Express 23, Wilson, Appl. Phys. A 59, 155 (1994). A240 (2015). 19 J. B. Xu, K. Lauger, R. Moller, K. Dransfeld, and I. H. Wilson, J. Appl. Phys. 76, 7209 (1994). 42 J.-H. Jiang and Y. Imry, Phys. Rev. B 97 (2018). 43 J.-H. Jiang, O. Entin-Wohlman, and Y. Imry, Phys. Rev. 20 A. Volokitin and B. Persson, Phys. Rev. B 63, 205404 B 85, 075412 (2012). (2001). 44 J.-H. Jiang, O. Entin-Wohlman, and Y. Imry, Phys. Rev. 21 J. P. Mulet, K. Joulain, R. Carminati, and J. J. Greffet, B 87, 205420 (2013). Microscale Thermophys. Eng. 6, 209 (2002). 22 L. Hu, A. Narayanaswamy, X. Chen, and G. Chen, Appl. 45 J.-H. Jiang, J. Appl. Phys. 116, 194303 (2014). 46 O. Entin-Wohlman, J.-H. Jiang, and Y. Imry, Phys. Rev. Phys. Lett. 92, 133106 (2008). E 89, 012123 (2014). 23 O. Ilic, M. Jablan, J. D. Joannopoulos, I. Celanovic, and M. Soljaci´c, Opt. Express 20, A366 (2012). 47 J.-H. Jiang, New J. Phys. 15, 075021 (2013). 48 J.-H. Jiang and Y. Imry, Phys. Rev. Applied 7, 064001 13 (2017). (2010). 49 J.-H. Jiang, X. G. Xu, L. Gilburd, and G. C. Walker, Opt. 59 M. Laroche, R. Carminati, and J.-J. Greffet, J. Appl. Phys. Express 25, 25059 (2017). 100, 063704 (2006). 50 K. Joulain, J.-P. Mulet, F. Marquier, R. Carminati, and J.-J. Greffet, Surf. Sci. Rep. 57, 59 (2005). 60 Z. Jacob, Nat. Mater. 13, 1081 (2014). 61 G. Yin, J. Yang, and Y. Ma, Appl. Phys. Express 9, 122001 51 R. Geick, C. Perry, and G. Rupprecht, Phys. Rev. 146, (2016). 543 (1966). 62 H. Hajian, A. Ghobadi, S. A. Dereshgi, B. Butun, and 52 L. A. Falkovsky, J. Phys.: Conference Series 129, 012004 E. Ozbay, JOSA B 34, D29 (2017). (2008). 53 H. Yan, T. Low, W. Zhu, Y. Wu, M. Freitag, X. Li, F. Guinea, P. Avouris, and F. Xia, Nat. Photonics 7, 394 (2013). 54 M. S. Shur, Handbook series on semiconductor parameters, vol. 1 (World Scientific, 1996). 55 D. Polder and M. Van Hove, Phys. Rev. B 4, 3303 (1971). 56 J. Pendry, J. Phys.: Condens. Matter 11, 6621 (1999). 57 Z. M. Zhang, Nano/microscale heat transfer (McGraw- Hill, 2007). 58 N. W. Ashcroft and N. D. Mermin, Google Scholar p. 461 63 G. Lu, T. Wu, P. Yang, Y. Yang, Z. Jin, W. Chen, S. Jia, H. Wang, G. Zhang, J. Sun, et al., Adv. Sci. 4 (2017). 64 S. Zhang, J. Li, H. Wu, X. Li, and W. Guo, Adv. Mater. Interfaces 5, 1800208 (2018). 65 R. He, D. Kraemer, J. Mao, L. Zeng, Q. Jie, Y. Lan, C. Li, J. Shuai, H. S. Kim, Y. Liu, et al., P. Natl. Acad. Sci. 113, 13576 (2016). 66 L.-D. Zhao, G. Tan, S. Hao, J. He, Y. Pei, H. Chi, H. Wang, S. Gong, H. Xu, V. P. Dravid, et al., Science 351, 141 (2016).
1910.12285	1	1910	2019-10-27T15:35:45	Dipolar cation accumulation at interfaces of perovskite light emitting solar cells	[ "physics.app-ph" ]	Ionic migration in organo-halide perovskites plays an important role in operation of perovskite based solar cells and light emitting diodes. Despite the ionic migration being a reversible process, it often leads to worsening of perovskite based device performance, hysteresis in current-voltage characteristics, and phase segregation in mixed halide perovskites being as the most harmful effect. The reason is in dynamical band structure changes, which controllable engineering would solve one of the biggest challenges for development of light-emitting solar cells. Here we demonstrate controllable band bending due to migration of both cation and anion ions in mixed halide perovskite devices. The band structure rearrangement is demonstrated in light emitting solar cells based on the perovskite with organic cations methylammonium (MA+) and formamidinium (FA+), possessing non-zero dipole momentum of 2.29 and 0.21 Debye, respectively, and with PEDOT:PSS and C60 transport layers having a high barrier of 0.8 eV for charge injection. Under applied external voltage MA+ and FA+ cations move towards the electron transport layer and form a dipole layer at the perovskite/electron transport interface, which lowers threshold voltage for electroluminescence down to 1.7 V for MAPbBr2I and 2.6 V for FAPbBr2I, whereas monohalide perovskite MAPbBr3 does not demonstrate such behavior. This ability to in-situ change the device band structure paves the way developing of dual-functional devices based on simple design. It also makes mixed halide perovskites more flexible than mono halides ones for developing different optoelectronic devices without the use of special types of work function modifying transport materials.	physics.app-ph	physics	ToC picture Dipolar cation accumulation at interfaces of perovskite light emitting solar cells D.S. Gets1, G.A. Verkhogliadov1, E.Y. Danilovskiy1, A. I. Baranov2, S.V. Makarov1, A.A. Zakhidov1,3. 1 ITMO University, Department of Nanophotonics and Metamaterials, Lomonosov Str. 9, Saint-Petersburg 191002, Russia 2 St Petersburg Academic University, ul. Khlopina 8/3a, St Petersburg 194021, Russia 3 Alan G. MacDiarmid NanoTech Institute, Department of Physics, University of Texas at Dallas, Richardson, TX 75083, USA E-mail addresses: [email protected] (D.S. Gets), [email protected] (A.A. Zakhidov). 1 Abstract Ionic migration in organo-halide perovskites plays an important role in operation of perovskite based solar cells and light emitting diodes. Despite the ionic migration being a reversible process, it often leads to worsening of perovskite based device performance, hysteresis in current-voltage characteristics, and phase segregation in mixed halide perovskites being as the most harmful effect. The reason is in dynamical band structure changes, which controllable engineering would solve one of the biggest challenges for development of light-emitting solar cells. Here we demonstrate controllable band bending due to migration of both cation and anion ions in mixed halide perovskite devices. The band structure rearrangement is demonstrated in light emitting solar cells based on the perovskite with organic cations methylammonium (MA+) and formamidinium (FA+), possessing non-zero dipole momentum of 2.29 and 0.21 Debye, respectively, and with PEDOT:PSS and C60 transport layers having a high barrier of 0.8 eV for charge injection. Under applied external voltage MA+ and FA+ cations move towards the electron transport layer and form a dipole layer at the perovskite/electron transport interface, which lowers threshold voltage for electroluminescence down to 1.7 V for MAPbBr2I and 2.6 V for FAPbBr2I, whereas monohalide perovskite MAPbBr3 does not demonstrate such behavior. This ability to in-situ change the device band structure paves the way developing of dual-functional devices based on simple design. It also makes mixed halide perovskites more flexible than mono halides ones for developing different optoelectronic devices without the use of special types of work function modifying transport materials. Introduction Organo-halide perovskites (OHP) is an encouraging family of organic-inorganic materials for developing highly efficient solar cells (SCs) and light emitting diodes (LEDs) [1, 2]. OHP have a high absorption coefficient, high exciton binding energy, wide band gap tunability, and solution processing and many more [3-7]. SCs based on perovskite reached efficiencies of SCs produced by well-established technologies like silicon [8], and perovskite LEDs demonstrate high efficiencies, narrow luminescence line width and high color rendering indexes [9]. But OHPs demonstrate some unwanted effects like ionic migration which tremendously affects the device performance [10, 11]. Ionic migration is induced either by illumination [12] or by passing the high current [13] through the device and most clearly manifested in mixed halide perovskites where it results in anion segregation [10, 12]. Perovskite ions and their vacancies have low activation energies, including mobility even at room temperature and for the long time, mnaking the ionic migration one of the biggest problems for perovskite devices [14-19]. The ionic migration strongly affects the device performance and results in hysteresis of PV parameters [20]. It clearly seen in current- voltage characteristic (J-V) of perovskite based solar cells where efficiency depends on the direction of voltage sweep. This unwanted behavior can be slow down by different ways like using multiple cation composition, producing perovskite nanocrystals, increasing perovskite grain size and interface stabilization [21-25]. Despite the negative effects of ionic migration, it can also demonstrate some positive effects. For example, under light illumination or application of voltage perovskite ions move towards electron and hole transport layers (ETL and HTL) and thus forming the p-i-n structure inside the perovskite layer [18, 26-29]. This p-i-n structure is responsible for temporal performance enhance of solar cells, which is known as light induced self-poling effect [26], and also it makes possible creation of dual-functional devices -- light emitting solar cells (LESC) [30-32]. 2 Latest experimental investigations of ionic migration in MAPbI3 shows that the anion ion has greatest diffusion coefficient and cation ion stays almost intact [33, 34], which means that p-i-n structure inside the perovskite layer is formed mostly due to halogen ion migration. Possibility of p-i-n structure formation due to ion migration inside the perovskite layer aids to the development of the LESC. The main problem of LESC is the absence of a mechanism to in-situ change the device band structure for certain working regime, LED or SC. Therefore, dual functional device will have additional losses due to the poor band alignment between the transport and emission layers when working in a reciprocal regime. Hence, creating LESC with high efficiencies in both working regimes is a challenge. Usually to overcome this potential barrier there is a need to insert additional layers to adjust work function of electrode [31, 32]. There are examples of differnet LESCs [35-37] the perovskite LESC based on standard design with PEDOT:PSS as HTL and C60 as ETL can work as a LESC despite high potential barrier (≈0.8eV) between the perovskite layer and ETL [30]. And the device within simple design can demonstrate low threshold voltage (Vth) and relatively good device performance with electroluminescence efficiency up to 0.04% (EQEEL) and photoconversion efficiency (PCE) up to 3-4% [30]. While in MAPbBr3-based LESC, which have high potential barrier between electrode and perovskite polyelyctrolyte ETL was used and thus LESC has the PCE ≈ 1% and EQEEL ≈ 0.12% [31], efficiencies. And in the case of MAPbI3-based LESC low work function electrode was used and device have PCE ≈ 12% and EQEEL ≈ 0.04% [32]. In this work, we demonstrate controllable real time band bending of the perovskite LESC based on mixed halide perovskites due to dipole layer formation from perovskite cation ions. Mixed halide perovskites demonstrate more pronounced ionic migration than monohalide ones due to the movement of both cation and anion ions. The difference in ionic migration is demonstrated by a high change in the LED threshold voltage values after prebiasing the device at different biasing voltages (Vb). The prebiasing results in migration of Br- and I- ions towards the PEDOT:PSS/perovskite interface and MA+ and FA+ molecules migrate towards the perovskite/C60 interface form the accumulation layers. The presence of these accumulation layers at the perovskite interfaces leads to p-i-n structure formation inside the perovskite layer, and, since, molecules of perovskite organic cations (MA+ and FA+) have non-zero dipole momentums (2.29 and 0.21 Debye) [38], the cation accumulation layer at the interface will bend device band structure more efficiently. This leads to improvement of charge injection with consequent Vth lowering down to 1.7V. Device Fabrication For the device fabrication the following scheme of functional layers ITO/PEDOT:PSS/FAxMA1- xPbBr2I/C60/LiF/Ag was chosen. The device functional layers were subsequently deposited onto ITO covered glass substrates and glass substrates with ITO were cleaned in an ultrasound bath in deionized water, acetone and isopropyl alcohol consequently. Dried substrates were exposed to UV irradiation (189, 254nm) for 900s. Water dispersion of PEDOT:PSS 4083 was used as a HTL, it was filtered through PTFE 0.45 syringe filter and deposited by spin coating. After spincoating the film was annealed on a heating plate for 10min at 150C. A photoactive layer based on FAxMA1- xPbBr2I was prepared by the consequent dissolving methylammonium iodide (MAI, DyeSol) and lead (II) bromide (PbBr2, Alfa Aesar "Puratronic" 99,999%) in a mixture of dimethylformamide (DMF) and dimethyl sulfoxide (DMSO) DMF:DMSO (7:3) respectively. The solutions were stirred 3 Figure 1. Ion migration. a J-V curves of MA- and FA-based perovskite devices after previasing at 1.5 Volts. Molecules of MA (d = 2.29 Debye) have greater dipole momentum than FA (d = 0.21 Debye) and therefore MA-based devices have stronger band bending and, consequently, lower Vth than FA-based devices. b dependence of Vth on cation composition of mixed halide perovskite. Values of Vth were obtained from linear approximation of J-V characteristics after several biasings of the devices at 1.5V for 1 minute in dark. The inset demonstrates probable band bending for devices based on MAPbBr2I (x = 0) and FAPbBr2I (x = 1). c and d Cation ion moves towards C60 layer and iodine and bromide ions moves towards PEDOT:PSS layer. Rearranged device band structure due to dipole layer formation of LESC based on MAPbBr2I and FAPbBr2I. overnight at room temperature. The acquired perovskite inks were deposited by the single step solvent engineering technique on top of HTL in two step spin-cycle inside the glovebox system with nitrogen atmosphere. Diethyl ether was used as an antisolvent and was slowly dripped on the rotating substrate in 10 s past the increase from 1000 to 3000 rpm. The acquired perovskite films were subjected to vacuum annealing for 1 minute with consequent annealing on the heating plate at 100C for 10 minutes. C60 layer was used as ETL. C60 as well as LiF and Ag layers were deposited by thermal evaporation. Results and discussion To investigate beneficial effect of ionic migration and p-i-n structure development inside the perovskite layer for band bending two perovskite organic cations FA+ and MA+ was chosen. The MAPbBr2I perovskite has a little bigger band gap than FAPbBr2I and FA cation offers greater stability against segregation than MA cation. For precise investigation of dual-functional devices performance of upon cation composition in mixed halide perovskite devices with cation FAxMA1- xPbBr2I were created (x = 0, 0.2, 0.4, 0.6, 0.8, 1). Obtained perovskite devices were tested for solar cell performance and they all demonstrated relatively good PV characteristics (VOC ≈ 0.85V, JSC ≈ 9mA/cm2, FF ≈ 50%, PCE ≈ 3%) regardless cation composition (Fig. S1). Study of the LED device regime performance and p-i-n structure formation was conducted by a slightly modified procedure suggested earlier [30]. LESC is a device capable of working in two reciprocal regimes as a light emitting diode (LED) and a solar cell (SC). Generally, LEDs and SCs perform reciprocal functions of converting electrical power to the light and vice-versa. Although these devices share similar designs, their designs are tuned in a specific way to maximize performance of the primary function, and therefore, the 4 reciprocal function is greatly suppressed by design due to additional losses because of high band mismatch. Therefore to create an efficient LESC a way to adjust band diagram for specific working regime should be found. However, dual-functionality in OHP devices was realized by using of low work function electrode material [32], a special transport material with polyelectrolyte properties [31], or ion movement [30]. The first two approaches are based on modulation of transport material properties. In the first case, the potential barrier for charge injection was overcome due to initially tuned device structure by using of low work function electrode material barium [32]. In the second case, the potential barrier was lowered by using polyethylenimine or pre-doped polyethylenimine (PEI or PEIBim4) as an ETL [31]. It is important to note that PEI demonstrates polyelectrolyte properties, which offers internal dipole formation, and it is usually used as a universal buffer layer to lower electrode work function [39]. The presence of dipoles at the interface MAPbBr3/ETL bends the band structure and, in turn, the potential barrier for charge injection also lowers, allowing for relatively good PV and LED performance in the same device. The third approach utilizes ion movement inside the perovskite layer [30]. Design used in these devices suits SC operation, but for LED operation it is highly unoptimized. The main problem is the high potential barrier for electron injection (≈0.8eV) between perovskite and ETL that makes this device design highly undesirable for LED realization. In this case under applied external voltage perovskite ions move towards perovskite-transport layer interfaces and form p-i-n structure inside the perovskite layer. This p-i-n structure aids in dual functionality. Recent investigations of halide diffusion in MAPbI3 perovskite demonstrated high mobility of the I- ion (D ≈ 10-9 cm2s-1) and extremely low mobility of the MA+ ion (D ≈ 10-12 cm2s-1) [19, 20, 27, 28]. Therefore, the p-i-n structure in perovskites forms only due to I- and Br- ions and its vacancies and it seems, which it is not enough to achieve high band structure bending in order to overcome potential barriers and enhance charge injection efficiency in the reciprocal working regime without using of any special electrode work function modifying layers. Each LESC device was subjected to a sequence of biasing cycles at a certain biasing voltage (Vb = 1, 1.5, 2 V etc.) for 1 minute in the dark and was immediately followed by J-V characteristic measurement from Vb up to 3 -- 4V (Fig. 1a). Applying the external electric field to the device leads to two processes: first, movement and accumulation of perovskite ions at the interfaces perovskite/ETL and perovskite/HTL, and second, current induced halide segregation [13]. An excess of perovskite ions at interfaces leads to p-i-n structure formation inside the perovskite layer since additional ions at interfaces serve as dopants [26, 18, 41]. Higher Vb should result in higher ion density at the interface, which in turn results in strongly pronounced p-i-n structure. However, ionic migration as well as p-i-n structure formation and segregation are temporal effects, and removing the external electric field leads to reverse redistribution of ions inside the perovskite layer and p-i-n junction disappearance. Therefore, to achieve low Vth the device must always be under applied external voltage. Along with J-V, we also measured the electroluminescence (EL) spectra of the LED (Fig. S2), which amplitude behavior supports J-V characteristics. Vth values were deduced from zero crossing of J-V curve linear approximation (Fig. S2). Fig. 1a demonstrates J-V characteristics of MA- and FA-based perovskite devices after several pre-biasings at 1.5V. The J-V characteristics demonstrate high difference in position of "knee". Obtained Vth values for MA-based (Vth ≈ 1.68V) and FA based (Vth ≈ 2.62V) perovskites demonstrate high difference (ΔVth ≈ 1V), which reflects the difference in charge injection into perovskite layer for these cations. For clarification the origin of such high difference between Vth of MA- and FA-based perovskite devices we conducted the same sequence for different FAxMA1-x 5 Figure 2. Biasing the mixed halide and monohalide perovskite devices. a J-V characteristics of the FAPbBr2I-based device after different pre-biasing voltages (Vb = 1, 1.5, 2V). Inset shows the behavior of EL amplitude after biasing under different Vb. b J-V characteristics of MAPbBr3 based device after pre- biasing at different voltages. Inset demonstrates corresponding EL peak intensity after pre-biasing at different voltages. (0≤x≤1) cation composition of perovskite. Fig. 1b demonstrates Vth dependence for different cation compositions FAxMA1-x (where x = 0, 0.2, 0.4, 0.6, 0.8, 1) after pre-biasings at Vb = 1.5V. Vth values obtained J-V characteristics demonstrate non-linear behavior upon cation FAxMA1-x (0≤x≤1) composition (Fig. 1b). Also we investigated influence of Vb value on Vth value, Fig. 2a shows J-V curves of FA-based LESC measured after pre-biasing at different Vb, the black line corresponds to the J-V characteristic after pre-biasing at 1V, the blue line to pre-biasing at 1.5V and the orange line after the pre-biasing at 2V. The increase of Vb leads to moving of the J-V characteristic "knee" to the low voltage region along with lowering of Vth value (Fig. 2a). Each Vb has its saturated value of Vth and than the longer time the device is subjected to the pre-biasing by Vb application results in more pronounced and shifted to the low voltage region "knee" of the J-V characteristic (Fig. S3) as well as attains saturated Vth value (Fig. S4). Moreover, increase of Vb also leads to higher amplitude of EL and as well as EL ignition occurs at the low voltage region, which reflects better charge injection conditions (inset in Fig. 3a). Devices with different cation compositions demonstrate different Vth values after several pre- biasing cycles at equal Vb. This behavior of Vth dependence and high total difference is quite surprising. We also recordered the optical power of devices in LED regime and external quantum efficiency, where the FA-based perovskite is much lower than MA-based (Fig. S5). While having similar characteristics to PV performance, LED emission line did not demonstrate any odd behavior, and all devices demonstrated an EL in the red region with dependence of peak position on FAxMA1-x (0≤x≤1) composition (Fig. S6). Anion segregation induced by high current injection produces domains enriched with iodine ions inside the perovskite layer, and these domains control the EL. It is important to note that, there is a high band mismatch between the perovskite layer and C60 (Fig 1.), but the device is still capable of demonstrating a relatively high LED efficiency [30] and low Vth. The difference in LED efficiency and in Vth values points to different conditions for charge injection into perovskite layer. MA-based perovskite offers better charge injection conditions than the FA-based. It turns out that the p-i-n structure in MA- and FA-based peroskites is different but in accordance with DFT calculations of defects in perovskite MAPbI3 [41], interstitial iodine (II) ion, and MA vacancy (VMA) produce an acceptor level near the valence band. At the same time, iodine vacancy (VI) and the interstitial MA molecule (MAI) produce a donor level near the conduction 6 Fig. 3. C-V characteristics of perovskite dual functional devices. a C-V characteristics before (black line) and after (red line) prebiasing the device. b Mott−Schottky curves of prebiased dual functional devices based on MAPbBr2I (green line), FAPbBr2I (red line), MAPbBr3 (blue line). Linear approximation gives Vbi of devices. band [41], and almost the same situation occurs with FAPbI3 [42], moreover, position of these type of defects does not change much in different perovskites [43, 44]. Therefore, the p-i-n structure in both cases should be similar, yet but experimental results shows the opposite. This contradiction points to additional effects influencing p-i-n development. Thus, we looked at properties of cation molecules MA and FA. Molecules of these cations have different and high dipole momentums of 2.29 and 0.21 Debye, respectively [38], and since MA molecules have a greater dipole momentum than FA molecules, MA-based device band bending will be stronger and their Vth lower than FA-based ones (Fig. 1b). To verify our hypothesis we conducted the same cycle of biasing on monohalide perovskite devices by investigating MAPbBr3-based devices in same device design (Fig. S7). Fig. 2b demonstrates obtained J-V, EL-V characteristics and EL spectra. J-V characteristics of MAPbBr3- based device, which do not change much event after biasing at high pre-biasing voltages (Vb = 2.5 V). The absence of the shift in J-Vs (Fig. 3b) and EL (inset in Fig 3b) after biasing at Vb = 1.5, 2.0, 2.5V corresponds to absence of MA+-ion movement under applied voltage, otherwise these characteristics would demonstrate considerable shift to lower voltages and increase of EL intensity in a sequence of measurements (inset in Fig. 2b). Although there is a small shift in J-V and EL-V characteristic (Fig. 2b and inset in Fig. 2b) but in can be attributed to movement of Br- ions and formation of weak p-i-n structure due to movement of Br-ions and it's vacancies. Such behavior of J-V characteristics, Vth value and electroluminescence in agreement with investigations of halide diffusion in MAPbI3, in which molecule of MA almost doesn't move, but atoms of I have considerable diffusion coefficient [33]. Meaning that, the ionic migration of only halide ion is not suffisient to effectively bend perovskite device band structure, which is clearly indicates by high Vth of MAPbBr3 based devices after pre-biasing the deice at high Vb. To more accurately determine the built-in voltage (Vbi) in our devices we conducted capacitance voltage characteristics of dual functional devices in LED regime. Fig.3a demonstrates the obtained C-V characteristics of dual functional devices based on MAPbBr2I (Fig. S8). The black curve corresponds to unbiased device and the red to the device biased at 1.5V for 1 minute. The presence of a "kink" on the red curve corresponds to charge accumulation in the device [45]. The Mott−Schottky analysis [46] of obtained C-V characteristics for devices based on MAPbBr2I, 7 Figure 4. Model. a Cation migration during the device pre-biasing (anion ions omitted for simplicity). The pre-biasings at different Vb lead to accumulation of cation ions at the perovskite\ETL layer. Higher Vb leads to higher cation density at the interface. As a result device will have different Vth values because different cation densities produce different device band bending. FAPbBr2I, and MAPbBr3 gave values of Vbi around 1.7, 2.6 and 2.9 V (Fig. 3b), which are in agreement with those obtained from LED J-V characteristics [45] (Figs. 2 and 3). According to latest investigations of halide diffusion in MAPbI3 the cation ion is almost intact as the iodine ion moves across the perovskite [33, 34]. Apparently, mixed halide perovskites demonstrate both cation and anion ions migration under applied external voltage. And pre-biasing plays important role in achievement of low Vth of the LED regime. During pre-biasing under applied voltage anion and cation ions move toward the perovskite/transport layer interfaces and form accumulation layers. The cation accumulation forming at the perovskite/C60 layer leads to band bending due to its non-zero dipole momentum. In this case, the cation accumulation layer works as a PEI layer [31,39]. A higher value of Vb results in a lower Vth (Fig 3a) due to the formation of a stronger accumulation layer (Fig. 4). Since MA+ and FA+ have different dipole momentums, this requires different Vb needed to bend band diagram and achieve low Vth (Fig. 2 and 4). Moreover, dipole momentum of MA molecule is comparable with dipole momentums of ETL widely used in organic light emitting diodes like BCP and TPBI [45]. Therefore, ionic migration of both cation and anion ions of perovskite along with high dipole momentum of organic cation make mixed halide perovskite more perspective than monohalide ones for development of LESC. Conclusions We have shown that the ionic migration in mixed halide perovskites demonstrates cation and anion movement, resulting in dipole layer formation near the perovskite/ETL interface, which lowers the threshold voltage for the LED operation mode. The presence of these accumulation layers leads to formation of more pronounced p-i-n structure inside the perovskite layer during device pre-biasing. Since molecules of perovskite organic cations MA and FA have non-zero dipole momentums of 2.29 and 0.21 Debye, respectively, band bending allows us to overcome high potential barrier between perovskite and transport layer in standard solar cell design based on PEDOT:PSS and C60 as transport layers. In case of MAPbBr2I perovskite, the Vth value can be as low as 1.7V, whereas the FAPbBr2I perovskite gives 2.6V, and MAPbBr3 perovskite barely reaches 2.9V. Thus, the device band diagram can be switched from solar cell to LED operation, and therefore creating the dual functional device mixed halide perovskites have more perspective than monohalide ones due to both cation and anion migration. 8 References [1] [2] [3] [4] [5] [6] Lu M., Zhang Y., Wang S., Guo J., Yu W. W., & Rogach A. L. Metal Halide Perovskite Light‐ Emitting Devices: Promising Technology for Next‐Generation Displays. Advanced Functional Materials, 2019, 1902008. Liu Z., Kruckemeier L., Krogmeier B., Klingebiel B., Marquez J. A., Levcenko S., Oz S., Mathur S., Rau U., Unold T., Kirchartz T. Open-circuit voltages exceeding 1.26 v in planar methylammonium lead iodide perovskite solar cells. ACS Energy Letters, 2018, 4, 110-117. Stranks S. D., Hoye R. L., Di D., Friend R. H., Deschler F. The physics of light emission in halide perovskite devices. Advanced Materials, 2018 1803336. Jung M., Ji S. G., Kim G., Seok S. I. Perovskite precursor solution chemistry: from fundamentals to photovoltaic applications. Chemical Society Reviews, 2019, 48, 2011-2038. Becker M. A., Vaxenburg R., Nedelcu G., Sercel P. C., Shabaev A., Mehl M. J., Michopoulos J. G. , Lambrakos S. G. , Bernstein N., Lyons J.L. , Stöferle T., Mahrt R.F. , Kovalenko M.V., Norris D.J. , Rainò G. and Efros A.L. Bright triplet excitons in caesium lead halide perovskites. Nature, 2018, 553, 189. Yang W. S., Park B. W., Jung E. H., Jeon N. J., Kim Y. C., Lee D. U., S. S. Shin, J. Seo, E. K. Kim, J. H. Noh and Seok S. I. Iodide management in formamidinium-lead-halide -- based perovskite layers for efficient solar cells. Science, 2017, 356, 1376-1379. [7] McMeekin D.P., Sadoughi G., Rehman W., Eperon G.E., Saliba M., Hörantner M.T., Haghighirad A., N. Saka1, L. Korte, B. Rech, M.B. Johnston, L.M. Herz, H.J. Snaith. A mixed- cation lead mixed-halide perovskite absorber for tandem solar cells. Science, 2016, 351, 151-155. [8] [9] Seok S.I., Grätzel M. and Park N., Methodologies toward Highly Efficient Perovskite Solar Cells. Small, 2018, 14, 1704177. Lin K., Xing J., Quan L.N., de Arquer F.P.G., Gong X., Lu J., Xie L., Zhao W., Zhang D., Yan C., Li W., Liu X., Lu Yan, Kirman J., Sargent E.H., Xiong Q. and Wei Z. Perovskite light-emitting diodes with external quantum efficiency exceeding 20 per cent. Nature, 2018, 562, 245. [10] Brennan M.C., Draguta S., Kamat P.V and Kuno M., Light-Induced Anion Phase Segregation in Mixed Halide Perovskites. ACS Energy Lett., 2018, 3, 204 -- 213. [11] Yuan Y. and Huang J. Ion Migration in Organometal Trihalide Perovskite and Its Impact on Photovoltaic Efficiency and Stability. Acc. Chem. Res., 2016, 49, 286 -- 293. [12] Hoke E. T., Slotcavage D. J., Dohner E. R.,. Bowring A. R, Karunadasa H. I. and McGehee M. D. Reversible photo-induced trap formation in mixed-halide hybrid perovskites for photovoltaics. Chem. Sci., 2015, 6, 613 -- 617. [13] Braly I. L., Stoddard R. J., Rajagopal A., Uhl A. R., Katahara J. K., Je, A. K. Y. and Hillhouse H. W. Current-induced phase segregation in mixed halide hybrid perovskites and its impact on two-terminal tandem solar cell design. ACS Energy Letters, 2017, 2, 1841-1847. [14] Slotcavage D. J., Karunadasa H. I. and McGehee M. D. Light-Induced Phase Segregation in Halide-Perovskite Absorbers. ACS Energy Lett., 2016 1, 1199 -- 1205. [15] Knight A. J., Wright A. D., Patel J. B., McMeekin D. P., Snaith H. J., Johnston M. B. and Herz 9 L. M. Electronic Traps and Phase Segregation in Lead Mixed-Halide Perovskite. ACS Energy Letters, 2018, 4, 75-84. [16] Aristidou N., Eames C., Sanchez-Molina I., Bu X., Kosco J., Islam M. S. and Haque S. A. Fast oxygen diffusion and iodide defects mediate oxygen-induced degradation of perovskite solar cells. Nature communications, 2017, 8, 15218. [17] Besleaga C., Abramiuc L. E., Stancu V., Tomulescu A.G., Sima M., Trinca L., Plugaru N., Pintilie L. Nemnes G.A., Iliescu M., Svavarsson H.G., Manolescu A., Pintilie I. Iodine migration and degradation of perovskite solar cells enhanced by metallic electrodes. The journal of physical chemistry letters, 2016, 7, 5168-5175 [18] Eames C., Frost J. M., Barnes P. R. F., O'Regan B. C., Walsh A. and Islam M. S. Ionic transport in hybrid lead iodide perovskite solar cells. Nat. Commun., 2015, 6, 2 -- 9. [19] Smith E. C., Ellis C. L., Javaid H., Renna L. A., Liu Y., Russell T. P., Bag M. and Venkataraman D. Interplay between Ion Transport, Applied Bias, and Degradation under Illumination in Hybrid Perovskite pin Devices. The Journal of Physical Chemistry C, 2018, 122, 13986-13994. [20] Habisreutinger S. N., Noel N. K. and Snaith H. J. Hysteresis Index: A Figure without Merit for Quantifying Hysteresis in Perovskite Solar Cells. ACS Energy Lett., 2018, 3, 2472 -- 2476. [21] Ferdani D., Pering S., Ghosh D., Kubiak P., Walker A., Lewis S. E., Johnson A.L., Bakedr P.J., Islam M.S. and Cameron P. J. Partial Cation Substitution Reduces Iodide Ion Transport in Lead Iodide Perovskite Solar Cells. Energy & Environmental Science, 2019, 12, 2264-2272. [22] Yoo J.J., Wieghold S., Sponseller M., Chua M., Bertram S. N., Hartono N.T.P., Tresback J.S., Hansen E.C., Correa-Baena J.-P., Bulović V., Buonassisi T., Shin S.S. and Bawendi M.G. An Interface Stabilized Perovskite Solar Cell with High Stabilized Efficiency and Low Voltage Loss. Energy & Environmental Science, 2019, 12, 2192-2199 [23] Han T. H., Tan S., Xue J., Meng L.,. Lee J. W, and Yang Y. Interface and Defect Engineering for Metal Halide Perovskite Optoelectronic Devices. Adv. Mater., 2019, 1803515, 1 -- 35. [24] Hu M., Bi C., Yuan Y., Bai Y., and Huang J. Stabilized wide bandgap MAPbBrxI3-x perovskite by enhanced grain size and improved crystallinity. Adv. Sci., 2015, 3, 6 -- 11. [25] Walsh A. and Stranks S. D. Taking Control of Ion Transport in Halide Perovskite Solar Cells. ACS Energy Lett., 2018, 3, 1983 -- 1990. [26] Deng Y., Xiao Z. and Huang J. Light-Induced Self-Poling Effect on Organometal Trihalide Perovskite Solar Cells for Increased Device Efficiency and Stability. Adv. Energy Mater., 2015, 5, 1500721. [27] Lee J.-W., Kim S.-G., Yang J.-M., Yang Y., and Park N.-G.. Verification and mitigation of ion migration in perovskite solar cells. APL Mater., 2019, 7, 4, 041111. [28] Puscher B. M. D., Ayguler M. Docampo F., P., and Costa R. D. Unveiling the Dynamic Processes in Hybrid Lead Bromide Perovskite Nanoparticle Thin Film Devices. Adv. Energy Mater., 2017, 7, 1 -- 10. [29] Zhang T., Cheung S. H., Meng X., Zhu L., Bai Y., Ho C. H. Y., Xiao S., Xue Q. So S.K. and Yang, S. Pinning down the anomalous light soaking effect toward high-performance and fast- response perovskite solar cells: the ion-migration-induced charge accumulation. The journal of physical chemistry letters, 2017 8, 5069-5076. 10 [30] Gets D., Saranin D., Ishteev A., Haroldson R., Danilovskiy E., Makarov S., and Zakhidov A. Light-emitting perovskite solar cell with segregation enhanced self doping. Applied Surface Science, 2019 476, 486-492. [31] Kim H. B., Yoon Y. J., Jeong J., Heo J., Jang H., Seo J. H., Walker B., Kim J. Y. Peroptronic devices: perovskite-based light-emitting solar cells. Energy & Environmental Science, 2017, 10, 1950-1957. [32] Gil-Escrig L., Longo G., Pertegas A., Roldan-Carmona C., Soriano A., Sessolo M. and Bolink H. J. Efficient photovoltaic and electroluminescent perovskite devices. Chemical Communications, 2015, 51, 569-571. [33] Senocrate A., Moudrakovski I., Acarturk T., Merkle R., Kim G. Y., Starke U., Grätzel M. and Maier J. Slow CH3NH3+ diffusion in CH3NH3PbI3 under light measured by solid-state NMR and tracer diffusion. The Journal of Physical Chemistry C, 2018, 122, 21803-21806. [34] Futscher M. H., Lee J. M., McGovern L., Muscarella L. A., Wang T., Haider M. I., Fakharuddin A., Schmidt-Mende L., Ehrler B. Quantification of ion migration in CH3NH3PbI3 perovskite solar cells by transient capacitance measurements. Materials Horizons, 2019, 6, 1497-1503. [35] Chiba T., Kumagai D., Udagawa K., Watanabe Y. and Kido J. Dual mode OPV-OLED device with photovoltaic and light-emitting functionalities. Scientific reports, 2018, 8, 11472. [36] Ayguler M. F., Puscher B. M., Tong Y., Bein T., Urban A. S., Costa R. D., and Docampo P. Light- emitting electrochemical cells based on inorganic metal halide perovskite nanocrystals. Journal of Physics D: Applied Physics, 2018, 51, 334001. [37] Liu Y., Hangoma P. M., Tamilavan V., Shin I., Hwang I. W., Jung Y. K., Lee B.R., Jeong J.H., Park S.H., Kim K. H. Dual-functional light-emitting perovskite solar cells enabled by soft- covered annealing process. Nano Energy, 2019, 61, 251-258. [38] Frost J. M., Butler K. T., Brivio F., Hendon C. H., van Schilfgaarde M. and Walsh A. Atomistic Origins of High-Performance in Hybrid Halide Perovskite Solar Cells. Nano Lett., 2014, 14, 2584 -- 2590. [39] Zhou Y., Fuentes-Hernandez C., Shim J., Meyer J., Giordano A. J., Li H., Winget P., Papadopoulos T., Cheun H., Kim J., Fenoll M., Dindar A., Haske W., Najafabadi E., Khan T.M., Sojoudi H., Barlow S., Graham S., Brédas J.-L., Marder S.R., Kahn A., Kippelen B. A universal method to produce low -- work function electrodes for organic electronics. Science, 2012, 336, 327-332. [40] Bernard G. M., Wasylishen R. E., Ratcliffe C. I., Terskikh V., Wu Q., Buriak J. M. and Hauger T. Methylammonium Cation Dynamics in Methylammonium Lead Halide Perovskites: A Solid-State NMR Perspective. The Journal of Physical Chemistry A, 2018, 122, 1560-1573. [41] Yin W.-J., Shi T., and Yan Y. Unusual defect physics in CH3NH3PbI3 perovskite solar cell absorber. Appl. Phys. Lett., 2014, 104, 063903. [42] Liu N. and Yam C. First-principles study of intrinsic defects in formamidinium lead triiodide perovskite solar cell absorbers. Phys. Chem. Chem. Phys., 2018, 20, 6800 -- 6804. [43] Kang J. and Wang L. High Defect Tolerance in Lead Halide Perovskite CsPbBr3. J. Phys. Chem. Lett., 2017, 8, 489 -- 493. [44] Jiang Q., Chen M., Li J., Wang M., Zeng X., Besara T., Lu J., Xin Y., Shan X., Pan B., Wang C., 11 Lin S., Siergrist T., Xiao Q., Yu Z. Electrochemical doping of halide perovskites with ion intercalation. ACS nano, 2017, 11, 1073-1079. [45] Noguchi Y., Miyazaki Y., Tanaka Y., Sato N., Nakayama Y., Schmidt T. D., Brutting W., and Ishii H. Charge accumulation at organic semiconductor interfaces due to a permanent dipole moment and its orientational order in bilayer devices. Journal of Applied Physics, 2012, 111, 114508. [46] Almora O., Aranda C., Mas-Marza E. and Garcia-Belmonte G. On Mott-Schottky analysis interpretation of capacitance measurements in organometal perovskite solar cells. Applied Physics Letters, 2016, 109, 173903. 12
1708.02928	1	1708	2017-08-09T17:54:13	Quantum efficiency modeling for a thick back-illuminated astronomical CCD	[ "physics.app-ph" ]	The quantum efficiency and reflectivity of thick, back-illuminated CCD's being fabricated at LBNL for astronomical applications are modeled and compared with experiment. The treatment differs from standard thin-film optics in that (a) absorption is permitted in any film, (b) the 200--500~$\mu$m thick silicon substrate is considered as a thin film in order to observe the fringing behavior at long wavelengths, and (c) by using approximate boundary conditions, absorption in the surface films is separated from absorption in the substrate. For the quantum efficiency measurements the CCD's are normally operated as CCD's, usually at $T = -140^\circ$C, and at higher temperatures as photodiodes. They are mounted on mechanical substrates. Reflectivity is measured on air-backed wafer samples at room temperature. The agreement between model expectation and quantum efficiency measurement is in general satisfactory.	physics.app-ph	physics	Journal of Applied Physics 122, 055301 (published online 2 August 2017) Quantum efficiency modeling for a thick back-illuminated astronomical CCD D. E. Groom,∗ S. Haque, S. E. Holland, and W. F. Kolbe Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA (Dated: August 10, 2017) The quantum efficiency and reflectivity of thick, back-illuminated CCD's being fabricated at LBNL for astronomical applications are modeled and compared with experiment. The treatment differs from standard thin-film optics in that (a) absorption is permitted in any film, (b) the 200–500 µm thick silicon substrate is considered as a thin film in order to observe the fringing behavior at long wavelengths, and (c) by using approximate boundary conditions, absorption in the surface films is separated from absorption in the substrate. For the quantum efficiency measurements the CCD's are normally operated as CCD's, usually at T = −140◦C, and at higher temperatures as photodiodes. They are mounted on mechanical substrates. Reflectivity is measured on air-backed wafer samples at room temperature. The agreement between model expectation and quantum efficiency measurement is in general satisfactory. 7 1 0 2 g u A 9 ] h p - p p a . s c i s y h p [ 1 v 8 2 9 2 0 . 8 0 7 1 : v i X r a ∗ [email protected] I. INTRODUCTION 2 Fully depleted thick back-illuminated p-channel charge-coupled devices (CCD's) developed at Lawrence Berkeley National Laboratory (LBNL) for astronomical applications have useful quantum efficiencies (QE's) extending into the near infrared (IR)[1, 2]. The response is limited to < ∼1100 nm by the indirect bandgap of silicon. The QE typically falls to about 50% at 1000 nm, depending on the temperature and thickness of the CCD. Thicknesses range from 200 to 500 µm. The temperature range of interest is −140◦ C to 20◦ C. If T > ∼ −100◦ C, the QE is measured by operating the device as a photodiode. A highly schematic cross section of a typical back-illuminated LBNL CCD is shown in Fig. 1. Modeling the response depends crucially on the complex refractive indices and thicknesses of the materials involved. Especially critical is the absorption coefficient α(λ, T ) of silicon as the indirect bandgap is approached. Most of the other indices also present special problems. While the formalism presented here is applicable to any CCD, we specialize to the LBNL case. In these CCD's, a thin film (10–25 nm) of in-situ doped polysilicon (ISDP) is grown on the rear surface to serve as an ohmic contact. Absorption in this layer limits the blue response, particularly below 450 nm. Over the ISDP is an antireflective coating optimized for maximum transmission at desired wavelengths, particularly in the near-infrared. This paper and the corresponding code grew out of work reported by Groom[3] in 1999. At that time the treatment of absorption in the ISDP and surface films was ad hoc at best. Absorption by an indium-tin oxide (ITO) film was neglected. ISDP absorption was poorly modeled, so that results below about 550 nm were uncertain. Although oblique incidence with either E and B parallel to the surface is treated, oblique incidence is of relatively little importance, since even for the extreme case of an f1 system the incident cos θ0 is ≥ 0.89, and the ray is "quickly straightened" by refraction into the high-index silicon (n = 3.7–4). The design constraints are daunting: As shown in Fig. 2, the absorption length in silicon ranges over four orders of magnitude in the useful wavelength region, from a few nm at the atmospheric cutoff near 320 nm to the thickness of the CCD, typically 250 µm. At the blue end of the spectrum it is close to the thickness of the (absorptive) ISDP, while at the near-IR end the CCD approaches transparency, with multiple reflections producing fringes. The analysis uses the standard transfer matrix formalism described in multiple sources, e.g. in Refs. 8–10. Some departures are made in addressing CCD-specific absorption issues: 3 AR coating ISDP rear window (n+ contact) Bias Voltage Back y x Photo- sensitive volume (200-300 μm) n− − (> 4 kΩ-cm) Buried p channel 50 nm SiO2 50 nm Si3N4 3-phase CCD structure polysilicon gate electrodes 300 nm thick 1.8-2 μm SiO2 15 μm pixel Epoxy Aluminum nitride or silicon Front FIG. 1. Optical structure of the LBNL CCD. In different versions the substrate resistivity ranges from 4 to 20 kΩ-cm, and some of the CCD's have a 10.5 µm pixel width. Not to scale. 5 10 4 10 ) m μ ( h t g n e l n o i t p r o s b A 3 10 2 10 10 1 0.10 –2 10 –3 10 200 300 400 500 Atmospheric cutoff T = 77 K = –196C T = 133 K = –140C T = 173 K = –110 C T = 300 K = 27 C 700 600 Wavelength (nm) 800 900 1000 1100 1200 Indirect bandgap @ 300 K 1.11 eV, λg 1116 nm FIG. 2. The absorption length ℓ in silicona. The 300◦ K solid curves are from the Handbook of Optical Constants of Solids[4] extended to 1200 nm[5] (black) and Green[6] (gray or green). The dashed curves are calculated from the phenomenological fits by Rajkanan et al.[7]. a The imaginary part of the index k, absorption coefficient α, and absorption length ℓ are related by 4πk/λ = α = 1/ℓ. 1. Many sources do not consider absorption-after all, one tries to make optical coatings out of transparent materials. But once it is introduced, there is a sign ambiguity in the definition of the complex index of refraction: nc = n + i k or nc = n− i k. The choice is arbitrary: The positive convention is used in Refs. 9–13, while the negative convention is used in Refs. 8, 14, 4 and 15. Having chosen the sign, one must take care that the rest of the formalism ensures that light is attenuated in absorptive layers. We adopt the negative sign convention. 2. The silicon substrate is also treated as a "thin film," although it is opaque over much of the optical range of interest. Fringes are commonly observed near the red end of a back- illuminated CCD's sensitivity, implying coherence over the thickness of the device and, remarkably, near-specular reflection from the front surface gate structure and mechanical substrate. For a 200 µm thick CCD the fringe spacing at λ = 1000 nm is only 0.7 nm, and in any case the fringes are "washed out" by a finite aperture. 3. It is useful to find absorption in the antireflective (AR) coating and, separately, in the ISDP. This is done by modifying boundary conditions, valid when the absorption length in silicon is small compared to the substrate thickness. We assume 100% internal quantum efficiency, i.e., every photon absorbed in the sensitive region produces a collected electron or hole. Model and measurement results are discussed for three antireflective coating designs: ITO/SiO2, ITO/ZrO2/SiO2, and TiO2/SiO2. II. MULTILAYER REFLECTED AND TRANSMITTED AMPLITUDES AND INTENSITIES In the simplest case, a plane wave in medium a with real index n0 (usually air or vacuum) is incident at an angle θ0 on a film with complex index n1 and thickness d1. It exits into medium b with index ns. With the definition γ = (n/c) cos θ (appropriate if E is parallel to the surface), the boundary conditions relate the fields at the two interfaces:   Ea Ba   =   cos δ1 i sin δ1 γ1 i γ1 sin δ1 cos δ1     Eb Bb   ≡ M1   Eb Bb   (1) The phase lag in one traversal, δ1, is (2πd1/λ)n1 cos θ1. The product n1 cos θ1 can be calculated from the complex version of Snell's law: n1 sin θ1 = n0 sin θ0 (2) The transport matrix M1 contains only variables pertaining to that layer; if the light is trans- mitted into another film a similar matrix is introduced. For N films,   Ea Ba   = M1M2 . . . MN   Eb Bb   ≡ M   Eb Bb   . 5 (3) The reflected fraction of the light r and the transmitted fraction t can be extracted from the boundary condition equations, r = γ0(m11 + γsm12) − (m21 + γsm22) γ0(m11 + γsm12) + (m21 + γsm22) t = 2γ0 γ0(m11 + γsm12) + (m21 + γsm22) , (4a) (4b) where the mij are the components of the product matrix M. Over most of the spectral region of interest, the silicon substrate is essentially opaque. More specifically, sin δ and cos δ both contain potentially large factors exp(2πdk/λ). While it is easy to block numerical overflows, we have found it convenient to factor out the divergent behavior: Mj = e−δIjMF j (5) Here δIj is the imaginary part of δI , and, because of our negative sign convention for the imaginary part of indices, it is always negative. Eqs. (4) become r = t = γ0(mF γ0(mF 11 + γsmF 11 + γsmF 21 + γsmF 21 + γsmF 22) 22) 12) − (mF 12) + (mF 2γ0 exp (P δIj) 12) + (mF (γ0mF 11 + γ0γsmF 21 + γsmF 22) . (6a) (6b) Since the exponential factors cancel in Eq. (6a), the reflected amplitude r is calculable for any amount of absorption, while the transmitted amplitude t is (essentially) zero for high absorption. If B is parallel to the surface, then γ = (n/c)/ cos θ. Since the phase lag δ is geometrical, it is not affected by polarization. However, in Eqs. (4b) and (6b), γ0 (= (n0/c) cos θ0) in the numerator is replaced by (n0/c)/ cos θs. The fractional reflected intensity R is r2. The fractional intensity of the light transmitted into the mechanical substrate is T = ℜ(ns cos θs) n0 cos θ0 t2 . (7) This fraction is not per se interesting, but if a fraction A of the light is absorbed in intermediate layers, then A = 1− R− T . This is the QE of the CCD plus Acoat, the fraction of the light absorbed in the surface films including the ISDP. 6 The desired result, the QE, is the absorbed fraction in the substrate alone. It is necessary to separate absorption in the substrate from absorption in the complete coating and to separate absorption in the IDSP from (possible) absorption in the AR layers. We can rewrite Eq. (3) as   Ea Ba   = MARMISDPMSi   Eb Bb   ≡ McoatMSi   Eb Bb   , (8) where MAR is the product of the transfer matrices for the AR coating films, MISDP is the transfer matrix for the ISDP coating, MSi is the matrix for the silicon substrate, and Mcoat = MARMISDP. One wishes to "get inside the device" and sample the fields just after the light exits the AR coating, or after exiting the AR + ISDP coatings. Unfortunately, light is reflected back into the AR coating at the substrate interface, and we have not found an algebraic solution for the fields after the AR layers or after the AR layers plus the ISDP coating. However, an alternative method is quite accurate over almost the entire spectral region. It depends on two features of the problem: 1. Since no light is transmitted over most of the wavelength region (λ < ∼ 900 nm, depending on the CCD thickness), we can replace the device with just the surface layers on a semi-infinite silicon substrate, and via Eqs. (4) or (6) find the fraction of the light transmitted (Tcoat) and reflected (Rcoat) by the ISDP + AR layers alone:   Ea Ba   = MARMISDP   Eb Bb   ≡ Mcoat   Eb Bb   (9) Then the fraction of light absorbed by the ISDP plus AR layers (Acoat) is 1 − Rcoat − Tcoat. This can be subtracted from the absorption in the complete device to find the QE: QE = A − Acoat (10) This is exact over most of the CCD, where the absorption length is small compared with the substrate thickness. 2. No serious error is introduced by lumping the Si and ISDP together as the semi-infinite silicon substrate, then calculating the transmission and reflectivity of the AR films alone. The index difference between Si and the ISDP is very small: The reflectivity is <0.04% for λ > ∼ 400 nm and < 0.01% for λ > ∼ 650 nm. One thus obtains AAR, the intensity fraction absorbed by the AR coating alone, essentially by the ITO. Then AISDP = Acoat − AAR. 7 ZrO2 (zirconia) TiO2 (rutile, ne) TiO2 (rutile, no) ITO 10 k ITO SiO2 MgF2 Si kSi 7 6 5 4 3 2 1 x e d n i f o s t r a p ) d e h s a d ( y r a n i g a m i d n a ) d i l o s ( l a e R k TiO2 (rutile, ne) 0 200 300 400 500 600 800 Wavelength (nm) 700 900 1000 1100 1200 FIG. 3. Real and imaginary parts of the refractive indices of silicon and a few antireflective (AR) coating candidates. These are for bulk samples, and may not be realizable in sputtered or vapor deposited films. The ITO data are from SOPRA/ITO2.NK[16] and are significantly different from those used in our earlier work[3]. III. REFRACTIVE INDICES The indices of silicon and some candidate AR films are shown in Fig. 3. Indices of the transparent films (TiO2, ZrO2, HfO2, fused SiO2, MgF2, etc.) are widely tabulated for bulk samples[16], but must be used with caution, especially for sputtered or vapor-deposited films. For our application HfO2 is inferior to ZrO2. SiO2 and MgF2 have very similar properties, but since the LBNL MicroSystems Lab has wide experience with SiO2 films, it is used in preference to MgF2. With the exception of SiO2, Si, ISDP, and all of the AR films used present special problems that are discussed here. More extensive studies of candidate materials have been published by Smith & Baumeister[17] and by Lesser[18], focussing on UV transparent oxides and fluorides. A. Index of silicon While the coatings and boundary conditions at the surfaces of the CCD affect transmission and reflection, it is absorption in the silicon substrate that results in charge collection. Understanding k(T, λ) is therefore of paramount importance in understanding its QE. This is particularly true in the near infrared, where the the absorption length ℓ (= 1/α = λ/4πk) rapidly approaches the thickness of the silicon substrate as the indirect bandgap is approached (at 1.12 eV, or 1100 nm, depending on temperature), reducing the QE to nearly zero. Many papers over the last 60 years have been devoted to the refractive index of silicon, partic- ularly the absorptive part, because of its great importance in solar cell design. A subset of these studies is particularly relevant to the optical and near-IR response of a CCD[6, 7, 19–28]. Two of these are relevant to our studies: 8 1. Green[6] (2008) published tables of optical parameters at 300◦ K, together with empirical power-law temperature coefficients for α, n, and k[29]: "The self-consistent tabulation was derived from Kramers-Kronig analysis of updated reflectance data deduced from the liter- ature." Our earlier work[3] used values of n from the Handbook of Optical Constants of Solids[4] extended to 1100 nm via tables presented by Janesick[5]. Green's values are quite close to these, but in our view supersede them. Temperature coefficients "calculated from cited and additional data sets" describe simple power laws as given in his Eqs. (9) and (10). Calculations using these coefficients indicate that the real part of the index varies only weakly with temperature. Changes are most evident near 380 nm. Since Green's n(300◦ K) and n(133◦ K) are nearly indistinguishable, we have adopted Green's n(300◦ K) for model calculations at all temperatures. Absorption calculated using his coefficients yields model QE's seriously at variance with our data. An example is shown in Fig. 11. 2. Rajkanan et al.[7] (1979) developed a physics-based model of the absorption that used ex- perimental data from MacFarlane et al.[19] and unpublished NASA sources to determine model parameters[30]. The best accuracy was obtained "with indirect band gaps at 1.1557 and 2.5 eV and a direct allowed gap at 3.2 eV" (390 nm). A fairly abrupt change in the absorption coefficient as the photon energy crosses this thresh- old (which increases somewhat with temperature) is evident in Fig. 2. Although their paper implies only 20% accuracy, we find remarkable agreement between the model predictions and our measured CCD QE at different temperatures and substrate thicknesses. At wavelengths below 390 nm, the Rajkanan et al. curves differ somewhat from measured values. This is due to the simple, smooth curve for k obtained from the Rajkanan et al. model in the direct bandgap region. It makes little practical difference, since (a) our QE measurements extend down only to 320 nm, and (b) absorptions lengths are so short (a few 10's of nm) that the ISDP layer already confuses the issue. 9 Satisfactory agreement between QE modeled using the Rajhanan et al.'s absorption coefficient[7] and data is obtained in all cases for the fiducial region in which the QE drops from 90% to 20% of its maximum, spanning a CCD thickness range from 200 µm to 500 µm and tempera- ture range from 20◦ C to −140◦ C. A surprising low-energy "skirt" at the higher temperatures extending the QE well above the indirect bandgap energy is discussed in connection with the TiO2/SiO2 coated CCD's. B. Index of in-situ doped polysilicon (ISDP) During fabrication, a fairly thick layer of phosphorus-doped polycrystalline silicon on the rear surface of the CCD acts as an active getter, maintaining the necessary very low leakage currents through high-temperature processing steps. After thinning and polishing a backside ohmic contact is formed by depositing a thin ISDP layer (10–25 nm). The process is described in more detail in Ref. 1. Also shown in that paper is a secondary ion mass spectroscopy depth profile of a nominal 20 nm thick ISDP layer. The P concentration varies by a factor of three until a depth of 20 nm is reached, then drops exponentially at about one decade/7 nm. It is difficult to convert this profile to a single number for model calculations. In most cases the model thickness needs to be increased by 5–10 nm from the nominal value to obtain agreement with measurements. The real part of its index is about the same as that of silicon, but it is considerably more absorptive than silicon in the blue[31]. Tables SIPOLYM .NK in the SOPRA database[16] provide n and k for M = 10–90. The peak values of both n and k decrease as M increases. Documentation of the SOPRA tables is not available, but it is likely that M is the fraction (in %) of amorphous silicon present. For λ > 450 nm n rises slightly with M , while k increases significantly. Holland, Wang, and Moses fabricated and measured the QE of a series of photodiodes with successively thicker ISDP coatings[32]. A sample of their data together with model fits using SIPOLY10.NK are shown in Fig. 4 for ISDP thicknesses of 10, 30, and 100 nm. In the 10 nm case the data and model are in essential agreement above 390 nm. The agreement is worse for the 30 and 100 nm cases, but the disagreement is in different directions. In all cases, and in our CCD QE measurements, the model QE falls below data for λ < ∼ 380 nm; our assumed index is simply too absorptive in this region. Since the QE is already falling rapidly here, the error is of little practical consequence. In the model QE calculations, ISDP absorption peaks at about 350 nm, and falls to insignificance in the red. Since there are yield concerns if the ISDP is too thin, most recent CCD's use 25 nm 10 0.8 0.7 0.6 0 I f o 0.5 1 − R (bare Si) n o i t c a r F 0.4 0.3 0.2 0.1 0.0 200 300 400 500 10 nm 30 nm 100 nm 700 600 Wavelength (nm) 800 900 1000 1100 1200 FIG. 4. Quantum efficiency of bare silicon with ISDP coatings 10, 30, and 100 nm thick. Measurements were made at "room temperature," and the model calculations, using indices from the SOPRA SIPOLY10.NK table, were at 300◦ K. coatings at the expense of significant QE loss below 500 nm. C. Index of indium-tin oxide (ITO) ITO films are (nearly) transparent to visible and near-IR light. In spite of ITO's widespread use, it is not an optically invariant material. Its optical properties depend upon the method of application, temperature, pressure, sputtering atmosphere and power, composition, and annealing. It varies from amorphous to crystalline, and often has a graded index[33–37]. ITO was first used in our application to augment rear surface conductivity and also, with careful thickness choice, to act as an AR coating or first layer of an AR coating[32]. It was reactively sputtered at room temperature in a low-pressure O2/Ar atmosphere from a target composed of 90% In2O3 and 10% SnO2 by weight. To optimize conduction in the ITO, the oxygen content of the film (less than saturated) was controlled using deposition parameters described in Ref. 32. Annealing in N2 for an hour at 200◦ C substantially improved the transmittance. Our original sources of information about the optical constants for indium-tin oxide were the papers by Woollam, McGahan, & Johs[33] and Gerfin & Gratzel[34]. Reference 34 gives tables of 6-parameter fits to a dispersion formula for the dielectric constant ǫ. Figure 5 shows the index, calculated as the real part of √ǫµ0/c, for three ITO films obtained from different sources. The Gerfin & Gratzel fits are based on data for 350 < λ < 690 nm. Data from a figure in Ref. 33 as smoothed by Gerfin & Gratzel's dispersion formula are shown by the dash-dotted red lines, the basis of our modeling calculations until recently. The curvature change and decrease of the 11 4 3 n 2 1 0 200 n Gerfin & Gratzel : fitting range 0.4 0.3 0.2 k 0.1 : Gerfin & Gratzel Woollam et al. SOPRA ITO2 A B C B C k 400 600 800 Wavelength (nm) 1000 0 1200 FIG. 5. Black curves show phenomenological fits to Gerfin & Gratzel's[34] spectroscopic ellipsometric measurements of ITO indices for samples A, B, and C between 1.8 and 3.5 eV (λλ = 689–354 nm). Dotted extrapolations were made using their dispersion formula. The dash-dotted red curves are functions of the same form drawn through the measurements by Woollam et al.[33]. Solid magenta curves are the ITO2 indices from the SOPRA database (250–850 nm), with dashed linear extrapolations to 200–1200 nm. index above 700 nm are expected from Drude-type absorption, indicating free carriers in the film. The solid magenta curves 250–850 nm are from ITO2.NK in the SOPRA database. The SOPRA indices, linearly extrapolated to λ = 1200 nm, have been used recently, and provide somewhat better agreement with the measured QE. Since the dispersion fits show increasing slope of k with λ and decreasing slope of n with λ, the SOPRA data and in particular our linear extrapolation may be unphysical. Attempts to fit a Gerfin & Gratzel-style dispersion function to the SOPRA data have failed to converge. Index uncertainties in the extrapolated region are relatively unimportant in our application. D. Index of zirconium oxide The refractive index of ZrO2 from three sources is shown in Fig. 6. The SOPRA data[16] show a curious inflection at about 600 nm, perhaps from combining indices from different sources. The LBNL MicroSystems Lab (MSL) sputtered film spectroscopic ellipsometer measurements were made at the LBNL Molecular Foundry. The cubic zircona sample measured by Wood & Nassau[38] contained 12-mole % yttria. Since the index of Y2O5 is considerably lower than that of ZrO2[39], one might expect the Wood & Nassau sample to have a smaller index than pure ZrO2, as the figure 12 2.6 2.5 x e d n i 2 O r Z 2.4 2.3 2.2 2.1 200 SOPRA MSL Wood82 400 600 800 Wavelength (nm) 1000 1200 FIG. 6. The upper (blue) curve is the ZrO2 index of refraction as given in the SOPRA database (ZRO2.NK)[16]. The index shown in red (MSL, LBNL MicroSystems Lab) is from spectroscopic ellip- someter measurements made at the LBNL Molecular Foundry. The lower (green) curve is from Wood & Nassau (1982)[38]; their sample contained 12-mol % yttria. Dashed curve segments indicate extrapolations. suggests. The MSL and Wood & Nassau curves have roughly the same smooth shape, without the unex- pected structure near 600 nm. We use the MSL index for our calculations. E. Index of titanium dioxide The index of the rutile form of TiO2 shown in Fig. 3 (highest curve) is from SOPRA TIO2.NK, and tables are available through Filmetrics[16]. It may have the highest index for any transparent AR film candidate. Its wavelength dependence is remarkably similar to that of silicon, making it the near-ideal material for a wideband antireflective coating. But there are problems: the SOPRA TIO2.NK index is evidently for the extraordinary ray in this birefringent material[4, 40]. Moreover, reactively sputtered TiO2 films at temperatures consistent with CCD fabrication are mostly the (also birefringent) anatase form: "Annealing of the films in air at 850 ◦C showed that anatase-rutile transformation strongly depends on the deposition temperature; the films deposited at temperature below 400◦C were converted to the anatase-rutile mixture films, and the films deposited at 400◦C to complete rutile films"[41]. The index of anatase is considerably below that of rutile, and tables are not readily available. The many papers on the subject do not present consistent results[42– 46]. For example, Dakka et al.[42] show different indices for "new target" (NT) and "used target" 13 Rutile ne (SOPRA) Rutile ne (DeVore, SOPRA) Rutile (Dakka) Rutile no (DeVore) UT (Dakka) 1.025 N U T (Dakka) T (Dakka) x e d n i 2 O T i 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 200 300 400 500 700 600 800 Wavelength (nm) 900 1000 1100 1200 FIG. 7. TiO2 refractive indices reported by Dakka et al.[42], DeVore[40], and SOPRA[16]. "NT" indicates "new target," "UT" indicates "used target." The DeVore 1951 result for ne is evidently used in SOPRA TIO2.NK for λ > 436 nm. The Dakka et al. UT result, extrapolated and scaled by 1.025 (red curve), gives the best fit to our QE data. The star indicates the measurement supplied by the manufacturer of the film, Hionix Inc. (UT) samples, and describe porosity and voids. The Dakka et al. results, together together with DeVore[40] and SOPRA[16] indices, are shown in Fig. 7. F. "Index" of the mechanical substrate When transmission becomes important in the near IR (> ∼ 900 nm, depending on the thickness of the silicon), the light exits into the gate structure and a mechanical substrate, which we have often modeled as exit into air: ns = 1.0. It is is considerably more complicated than this, as shown in Fig. 1. The light encounters 50 nm of SiO2, then 50 nm of Si3N4, polysilicon gates 300 nm thick (with 40% overlap), a thick layer of SiO2, and epoxy that binds the device to thick silicon or aluminum nitride. Remarkably, a single index ns seems to describe this region adequately. Modeled QE turns out to be very insensitive to its value: ns = 1.5 or even ns = 2.0 moves the model calculation in the steep IR falloff region only slightly to the left, where it agrees slightly better with the measurements. However, asymptotic 1−R and T increase significantly with increasing ns. Examples are shown below for both ns = 1.0 and ns = 1.5. 14 IV. MODEL CALCULATIONS COMPARED WITH QE AND REFLECTIVITY MEASUREMENTS Only the reflectivity R and the transmitivity T are directly modeled. As per the discussion of Sec. II, 1 − R − T can be approximately decomposed into the QE, the light fraction absorbed in the ISDP layer (AISDP), and the fraction absorbed in the antireflective coatings (AAR). If the AR coating is not significantly absorptive, the QE and 1−R curves should be nearly identical for a fairly wide region in the red and near IR, where there is no transmission and the ISDP is essentially transparent. If the AR coating does absorb significantly at all wavelengths, then there is a calculable gap between the QE and 1−R curves, as is evident in Figs. 9 and 10. The measured QE is subject to amplifier gain uncertainties and other problems at the few-percent level. The absolute measurement of R is used to normalize the QE measurements. The QE can be measured in either the normal CCD mode or by reading out the entire or a masked subsection of the CCD as a photodiode (PD mode). This provides some additional redundancy, and measurements can be made at higher temperatures than are possible in the CCD mode. An example is discussed in Sec. IV D. Our setup for measuring the QE[47] is fairly standard: light from a monochromator enters an integrating sphere and exits a large aperture. It arrives at the dewar containing the CCD after an 0.8 m drift space in a baffled box. Slit widths are varied with light intensity; the bandwidth can be as small as 10 nm. A room-temperature standard photodiode at the CCD's position and behind the same dewar window is used to calibrate a similar photodiode at a small port in the integrating sphere that is then used as the reference for the QE measurement. The reflectometer is described in Ref. 48. The intensity of a light beam from the monochromator is measured by a photodiode after several mirror reflections. One mirror is then moved so that a reflection from the (air-backed) CCD wafer sample is included in the optical path. The ratio yields the absolute reflectivity R. These measurements are at room temperature. The CCD's have from 4 to 16 readout amplifiers whose gain calibrations can be uncertain at the few percent level. In a broad red spectral band the QE should be nearly 1−R; this is used to normalize the QE measurements. The CCD's fabricated and studied so far use various combinations of ISDP, ITO, ZrO2, TiO2, and SiO2 films on the silicon substrate. While the index of SiO2 is very well known, the others all require special attention. In particular, the near IR response of the CCD depends crucially on the absorption coefficient of silicon, which is discussed in detail. 100 80 0 . 8 2 0.84 0.86 0.88 0.90 60 0.92 ) m n ( 2 O r Z 40 y + r +2 ir ISDP = 25 nm, ITO = 20 nm max(SiO2, ZrO2) = 0.938 at (80 ,44) 0 . 8 8 0.90 0 .92 0.92 0.90 0.88 0.80 0.82 0.84 0.86 20 0 40 15 0 0 . 9 0 . 8 8 6 0 . 8 0.84 60 80 100 120 140 160 SiO2 (nm) FIG. 8. Model QE contours as a function of ZrO2 and SiO2 thicknesses, where the QE is the average response of Hyper-Supreme-Cam g and i filters[49]. The maximum shown, at ISDP:ITO:ZrO2:SiO2 = 10:20:106:42 nm, is close to the values chosen for the DESI CCD's. The maximum of the model QE maximum using these values is 96.9% at 875 nm. A. Two-layer AR coating design methodology Recent designs use a minimal-thickness ITO coating over the ISDP, followed by high-index and low-index layers. Although this is technically a 3-layer AR coating, the thickness of the ITO is held constant in optimizing the other thicknesses for maximum response. In the absence of any clear criterion for AR coating optimization, the response was maximized for several linear combinations of broadband and narrowband filter responses. Figure 8 shows an example for an ITO (fixed 20 nm thick)/ZrO2/SiO2 coating, where the model QE is the average of HSC-g and HSC-i responses[49]. Response was calculated for a matrix of ZrO2 and SiO2 thicknesses to make the QE contour plot. Thickness tolerances can be estimated from the "flatness" of the peak. Although still somewhat subjective, the method has been moderately successful. Subsequent thickness tuning on the basis of experiment has also been useful. One problem, still not understood, is a discrepancy of 10–15% between the deposited SiO2 thickness and the modeled thickness-the CCD behaves as though excessive SiO2 has been deposited[50] . We correct for this empirically by depositing a thinner SiO2 layer. 16 1.0 0.9 0.8 0.7 0 I f o 0.6 n o i t c a r F 0.5 0.4 0.3 0.2 0.1 0.0 1–R QE T Absorbed in ITO + ISDP Absorbed in ISDP Absorbed in ITO 200 300 400 500 600 800 Wavelength (nm) 700 900 1000 1100 1200 FIG. 9. Model calculation for a 200 µm thick CCD at −140◦ C with 80 nm SiO2 and a 58 nm ITO AR coating. The ISDP layer is 25 nm thick. The differences between the QE and 1−R at ≈ 800 nm and between the asymptotic T and 1−R are due to absorption in the ITO. Boxcar averages over the fringe bands are indicated by the white curves. Here and in Fig. 10 the mechanical substrate index was taken as 1.0. The approximations made in calculating coating absorption break down when there is appreciable transmission, producing the spurious "foot" of the QE for λ > ∼ 1100 nm. 1.0 0.9 0.8 0.7 0 I f o 0.6 n o i t c a r F 0.5 0.4 0.3 0.2 0.1 Filmetrics 1–R 1–R QE ITO + ISDP T ISDP ITO 0.0 200 300 400 500 600 800 Wavelength (nm) 700 900 1000 1100 1200 FIG. 10. Same as Fig. 9 with fringe bands removed and QE measurements added. The blue circles on the descending part of the curve indicate the measured values corrected for the monochromator bandwidth. The dashed 1−R curve is from the online Filmetrics Reflectance Calculator[16]. B. ITO/SiO2 coating 17 In initial designs, indium-tin oxide was used both to ensure adequate rear-surface conductivity and to serve as the first layer of the AR coating. The model output is shown in Figs. 9, 10, and 13. The full calculation is shown in Fig. 9. In this and in most other examples, light is normally incident on the CCD. The decomposition of the absorption into QE, absorption in the ISDP+AR coating, ISDP, and AR is discussed in Sec. II. Above 900 nm the CCD becomes increasingly transparent, resulting in interference between the reflected and transmitted amplitudes in the silicon substrate. The responses change rapidly with increasing wavelength, and appear as bands in Fig. 9. White lines show box-car averaged T , 1−R, and QE, the intensity fraction absorbed in the sensitive region. The discrepancy between 1−R and the QE in the near-IR is primarily caused by absorption in the ITO. The approximations made in separating the QE and absorption in the ISDP and ITO break down as fringing becomes more pronounced, so the modeled QE does not quite go to zero at long wavelengths. Figure 10 shows the same curves, but with the fringing bands replaced by boxcar averages, experimental QE measurements (red triangles) added, and a dashed curve comparing the modeled 1−R with results from the Filmetrics Reflectivity Calculator[16] has been added. Blue circles on the descending part of the QE response illustrate a correction for the finite monochromator bandpass. For this model calculation ns, the effective index of the mechanical substrate, ns, was taken as 1.00. The value is relevant only in the transparency region, λ > ∼ 900 nm. CCD's with this coating have been used in the BOSS[51] camera, in DECam[52], in the red leg of the KECK Low Resolution Spectrometer[53], and in other applications. C. (ITO)/ZrO2/SiO2 coating A design has been developed for the Dark Energy Spectroscopic Instrument (DESI)[54] that uses ZrO2 rather than ITO for most of the high-index layer. There is a 20–25 nm ITO film between the ISDP and ZrO2. The ITO is included to avoid direct contact with a new material that might introduce reliability issues. (We have had substantial experience with ITO in direct contact with ISDP.) In any case, the ITO does not seriously compromise the QE in the DESI red and IR channels[54]. An example is shown in Fig. 11. Compared with the ITO/SiO2 response (Figs. 9 and 10), the blue QE is increased and flattened, and the QE is only slightly below the 1−R limit in a 18 1.0 0.9 0.8 0.7 0 I f o 0.6 n o i t c a r F 0.5 0.4 0.3 0.2 0.1 0.0 QE 1–R Green T Absorbed in ITO + ISDP (coat) Absorbed in ISDP Absorbed in ITO (AR) 200 300 400 500 600 800 Wavelength (nm) 700 900 1000 1100 1200 FIG. 11. The nominal ISDP:ITO:ZrO2:SiO2 thickness were 11:20:38:106 nm, while a better fit (shown) is for thicknesses 18:20:38:118 nm. The wafer was 250 µm thick, the temperature was −140◦ C (model and QE data) and the mechanical substrate index ns was 1.5. Reflectivity was measured at room temperature. Fringing bands are deleted from the figure. Curves show the model results; points show measurements. QE data for 12 channels are scaled for agreement near the model QE peak. The dashed curve is calculated using Green's temperature coefficients[6]. broad red region. For the model calculation shown here, ns, the effective index of the mechanical substrate, was taken as 1.50. A comparison with Fig. 9, where it was 1.00, shows that the asymptotic transparency and 1−R are both greater, although their difference (nearly the QE in this case) is almost unchanged. The near-IR QE using Green's k(λ,−140◦ C)[6] (Sec. III A) is also shown in Fig. 11. Green's agreement with Rajkanan et al. is better when the comparison is made at T = 300◦ C. D. TiO2/SiO2 coating Since the LBNL MicroSystems Lab did not have a sputtering target for TiO2 deposition, a TiO2/SiO2 coating was applied to an otherwise-complete CCD by the Hionix corporation[55]. They reported a TiO2 index of 2.44 at 633 nm. The SiO2 film was about 14% thicker than requested, resulting in a QE somewhat lower than expected. The response is shown in Fig. 12. Good fits to the measured QE were obtained using the Dakka "UT" index scaled by 1.025, in agreement with the Hionix measurement at 633 nm. This CCD 1.0 0.9 0.8 0.7 0 I f o 0.6 n o i t c a r F 0.5 0.4 0.3 0.2 0.1 0.0 Acoat AAR AISDP 200 300 400 500 19 Measured 1− R Modeled 1− R −140 C −60 C +25 C 600 Wavelength (nm) 700 800 900 1000 1100 1200 FIG. 12. Response of an ISDP/TiO2/SiO2 (32 nm/57 nm/132 nm) coated CCD at −140 C (CCD mode), −60 C (PD mode) and 25 C (PD mode). Curves are the model results; points are the measurements. Substrate index is 1.5 for model QE and 1.0 for model 1−R. A dotted vertical line at 1116 nm indicates the indirect bandgap at 300◦ K. At the higher temperatures, particularly at 25 C, the QE does not fall to zero as rapidly as the model predicts, evidently due to e-h production via a photon-two phonon processes[56]. has a better response than any of the others we have tested. Nonetheless, given the difficulty and likely unpredictability of TiO2 films and only marginal improvement from the (ITO)/ZrO2/SiO2 AR coating, there is little incentive to pursue this approach. A remarkable feature of the room-temperature measurements is the QE "skirt" extending well beyond 1116 nm, the wavelength corresponding to the indirect bandgap at 1.1108 eV at 300◦K (vertical dotted line in Fig. 12). In order to conserve both energy and momentum, absorption involving indirect transitions requires the absorption or emission of one or more phonons. The Rajkanan et al. model[7] takes into account the two lowest phonon excitations, 0.0183 eV and 0.0577 eV that produce contributions to the absorption coefficient with displacements of ±19 nm and ±61 nm. As a result, the model QE at 25◦ C shown in Fig. 12 is about 8% at 1100 and falls to zero by 1177 nm. However, the measured response is above 10% at 1120 nm and falls to zero only just below 1200 nm. This e-h production by low-energy photons is thought to be due to higher-temperature double-phonon processes producing offsets of ±99 nm and ±149 nm[19]. 20 1–R QE T 1.0 0.9 0.8 0.7 0 I f o 0.6 n o i t c a r F 0.5 0.4 0.3 0.2 0.1 0.0 990 992 994 996 1000 1002 1004 1006 1008 1010 998 Wavelength (nm) FIG. 13. Detail of the 990–1010 nm region of Fig. 9. The fringe spacing is δλ = λ2/2nd, where d is the silicon thickness and n is the real part of its index (at 1000 nm, n = 3.57 and δλ = 0.7 nm). The dashed central curves show the results of averaging the light's incidence angles over an f1.5 aperture. These are close to the boxcar averages, which appear to be "below center" because of the asymmetry of the Haidinger fringes. E. Fringing The sensitive region of a "traditional" thin back-illuminated CCD is an epitaxial layer ≈ 20 µm thick. At longer wavelengths light reaches the back surface, and the resulting multiple reflections produce interference "fringes." A white-light exposure with the KECK low-resolution spectrograph (LRIS)[57] is shown in Ref. 3. The fringes are somewhat irregular because of thickness variations in the epitaxial layer, but one can infer from the fringe average spacing δλ that the layer is 20– 22 µm thick. (For a thickness d at a wavelength where the real part of the silicon index is n, δλ = λ2/2nd.) In this case, the spacing is about 7 nm at λ = 1000 nm, providing at once a way to measure the epitaxial layer thickness[58] and a nuisance for observers. Broadband photometry using R and IR filters is plagued by swirled interference patterns whose amplitude is modulated by time-varying OH sky lines. Removal of these fringes is discussed by McLean[59] and others, and algorithms to treat the problem are part of observatory software. That the fringes exist means that the light remains coherent over at least 20 µm and that reflection from the mechanical substrate is remarkably specular. An expansion of our modeled 990–1100 nm region for a 200 µm thick CCD's is shown in Fig. 13. For normally incident light the peak-to-valley QE variation is nearly 0.2 of the incident intensity, or 21 0.34 of the average QE. However, the fringes are ten times closer than those in a 20 µm thick CCD, or about 0.7 nm at 1000 nm. Only the highest-resolution spectrometers could resolve these fringes. In particular, they are not resolved in our QE measurements because of the much-wider bandpass of the monochromator. Even if they are not observed, the agreement of the boxcar averaged QE with measured QE argues for coherence and fairly specular reflection. An actual instrument has a finite aperture, so that light arrives at different angles, each with a fringe spacing characteristic of the slant depth. The fringing pattern tends to average out. The central dashed lines in Fig. 13 were calculated for an f1.5 aperture. The remaining oscillations are the result of a beat pattern characteristic of this wavelength range; in the model they disappeared and reappeared as the aperture was changed. Results of the calculation are nearly independent of polarization. Using a DECam CCD, Stevenson et al.[60] searched for water in the atmosphere of an exoplanet with the LDSS-3 spectrometer (focal ratio f2.5) at the Magellan telescope at Las Campanas Obser- vatory. Fringes were not observed. A more detailed study of one of the images by A. Seifahrt[61] set limits on the fringe intensity of 0.25% (peak-to-valley). DECcam Y-band images have shown fringing swirls at the 0.4% peak-to-valley level[62]. The pattern and spacing is probably inconsistent with interference in the CCD itself, given its thickness and expected thickness variations. It is possible that the interference originates in the epoxy layer binding the CCD to the mechanical substrate, which was aluminum nitride for the DECam CCD's. In any case, the steep QE falloff in the near IR limits any possible fringing to the 900–1050 nm region. The fringing amplitude, especially in the cases of the DESI (ITO)/ZrO2/SiO2 CCD(Fig. 11) and the TiO2/SiO2 CCD (Fig. 12), is minimized by the very low reflectivity in this region. For normally incident light the maximum peak-to-valley QE variation is 0.10 for the DESI CCD and 0.08 for the TiO2/SiO2 CCD. V. PROSPECTS FOR QE IMPROVEMENT It was mentioned above that the model thicknesses had to be "tuned" somewhat to obtain agreement with the measurements. To some extent this was because of variation in the actual film thicknesses, or, more likely, that the thin film indices were different from those measured in bulk samples. (In the case of TiO2, there was even ambiguity about the crystal form.) Thus by perturbing the actual film thicknesses the QE can be improved. Experimental AR coatings are being made to explore this. It is likely that response of the (ITO)/ZrO2/SiO2 (DESI) CCD can be 22 enhanced in the 450–600 nm region from about 84% to about 90% by increasing the design ZrO2 thickness by 7–10 nm-a spectral region just below the DESI red channel, but the increased blue response might be useful for other applications. Our blue response is limited by absorption in the ISDP rear electrode. An alternate approach is delta doping, depositing approximately a monolayer of dopant atoms on the rear surface via molecular beam epitaxy (MBE). This technique has been developed at the Jet Propulsion Lab (Caltech) for a variety of spacecraft and sounding rocket missions[63]. The object is to obtain reflection-limited ultraviolet quantum efficiency down to about 100 nm. Antimony layers about 5 nm thick were applied in this way to 2k×4k and 1k×1k LBNL CCD's similar to those described here[64, 65]. QE measurements with and without a (nomi- nally) Si3N4/SiO2 AR coating are shown in Ref. 64. The results are consistent with ours except that their QE remains high down to about 400 nm. There has been no further work on the delta-doping approach. VI. CONCLUSIONS We have extended standard thin-film optics methods to model the quantum efficiency (QE) of LBNL thick, back-illuminated CCD's, and compared the results with experimental measurements of QE and reflectivity for three antireflective coating examples. The calculations included (a) considering the thick substrate as one of the films and (b) separating absorption in an ISDP coating serving as the rear contact and in the antireflective coating from absorption in the silicon substrate (the QE) by modifying boundary conditions. Problems encountered with the indices involved are discussed. While agreement with experimental QE and reflectivity measurements are regarded as adequate, it is limited by uncertainties in the ISDP index, the ITO and TiO2 indices, and film deposition thickness variations. Fringing is neither expected nor observed. VII. ACKNOWLEDGEMENTS The QE "machine" was built by Jens Stecker and the reflectometer by Maximilian Fabricius, both in collaboration with and under the supervision of Armin Karchar. With the exception of TiO2, AR coatings were applied in the LBNL MicroSystems Lab. We are also grateful for Chris Bebek's encouragement and guidance. . This work was supported by the U.S. Department of Energy under Contract No. DE-AC02- 05CH11231. 23 [1] S. Holland, D. Groom, N. Palaio, R. J. Stover, and M. Wei, in IEEE Trans. Electron Devices, Vol. 50 (2003) pp. 225–238. [2] S. E. Holland, C. J. Bebek, K. S. Dawson, J. H. Emes, M. H. Fabricius, J. A. Fairfield, D. E. Groom, A. Karcher, W. F. Kolbe, N. P. Palaio, N. A. Roe, and G. Wang, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Proc. SPIE, Vol. 6276 (2006) p. 62760B. [3] D. E. Groom, S. E. Holland, M. E. Levi, N. P. Palaio, S. Perlmutter, R. J. Stover, and M. Wei, in Sensors, Cameras, and Systems for Scientific/Industrial Applications, Proc. SPIE, Vol. 3649, edited by M. M. Blouke and G. M. Williams (1999) pp. 80–90. [4] D. F. Edwards, Handbook of Optical Constants of Solids, edited by E. D. Palik (Academic Press, 1985) pp. 547–569. [5] J. R. Janesick, IS&T/SPIE Symposium on Electronic Imaging Science and Technology , 233 (no data source or year given). [6] M. A. Green, Solar Energy Materials and Solar Cells 92, 1305 (2008). [7] K. Rajkanan, R. Singh, and J. Shewchun, Solid-State Electronics 22, 793 (1979). [8] F. L. Pedrotti and L. S. Pedrotti, Introduction to Optics (Prentice Hall, New Jersey, 1993). [9] E. Hecht, Optics, 3rd ed. (Adddison-Wesley, 1998). [10] H. A. Macleod, Thin-Film Optical Filters, 3rd ed. (Institute of Physics, Bristol, 2001). [11] F. A. Jenkins and H. E. White, Fundamentals of Optics (2nd edition) (McGraw Hill, New York, 1950). [12] W. K. Panofsky and M. Phillips, Classical Electricity and Magnetism (Addison-Wesley, Redding, MA, 1955). [13] J. C. Slater and N. H. Frank, Electromagnetism (McGraw-Hill, New York, 1947). [14] M. Born and E. Wolf, Principles of Optics, 7th ed. (Pergamon, Cambridge, 1999). [15] A. Sommerfeld, Optics (Academic Press, New York, 1964). [16] Most useful indices can be downloaded from filmetrics.com. Many are from the SOPRA Material Database, given without citation. In some cases polynomial or dispersion fits can be found. [17] D. Smith and P. Baumeister, Applied Optics 18, 111 (1979). [18] M. Lesser, Optical Engineering 26, 911 (1987). [19] G. G. Macfarlane, T. P. McLean, J. E. Quarrington, and V. Roberts, Phys. Rev. 111, 1245 (1958). [20] W. C. Dash and R. Newman, Phys. Rev. 99, 1151 (1955). [21] H. R. Philipp and E. A. Taft, Phys. Rev. 120, 37 (1960). [22] H. A. Weakliem and D. Redfield, J. App. Phys. 50, 1491 (1979). [23] G. E. Jellison, Jr. and F. A. Modine, J. App. Phys. 53, 3745 (1982). 24 [24] G. E. Jellison, Jr. and F. A. Modine, Appl. Phys. Lett. 41, 180 (1982). [25] G. E. Jellison, Jr. and H. H. Burke, J. App. Phys. 60, 841 (1986). [26] G. E. Jellison, Optical Materials 1, 41 (1992). [27] K. Bucher, J. Bruns, and H. G. Wagemann, J. App. Phys. 75, 1127 (1994). [28] M. A. Green and M. Keevers, Progress in Photovoltaics 3, 189 (1995). [29] To maintain the required significance at longer wavelengths it is necessary to recalculate k from Green's absorption coefficients. Tables are most easily copied from the web version: sciencedirect.com/science/article/pii/S0927024808002158#. [30] A nearly-identical expression for α is given by Bucher et al.[27] without reference to Rajkanan et al.'s earlier paper. With two exceptions, the parameters are the same or nearly the same. However, the direct allowed bandgap contribution is given as A(¯hω − Egd)3/2/¯hω, rather than Rajkanan et al.'s A(¯hω − Egd)1/2. Evidently the exponent should be 1/2 for a direct bandgap transition[56]. Although both forms give similar results for photon energies below Egd, we choose Rajkanan et al.'s form. [31] G. Lubberts, B. C. Burkey, F. Moser, and E. A. Trabka, J. Appl. Phys. 52, 6870 (1981). [32] S. E. Holland, N. W. Wang, and W. W. Moses, IEEE Trans. Nucl. Sci. 44, 443 (1997). [33] J. A. Woollam, W. A. McGahan, and B. Johs, Thin Solid Films 241, 44 (1994). [34] T. Gerfin and M. Gratzel, J. Appl. Phys. 79, 1722 (1996). [35] S. B. Lee, J. C. Pincenti, A. Cocco, and D. L. Naylor, J. Vac. Sci. Tech. A 11, 2742 (1993). [36] H. E. Rhaleb et al., Applied Surface Science 201, 138 (2002). [37] C. Guill´en and J. Herrero, J. App. Phys. 101, 073514 (2007). [38] D. L. Wood and K. Nassau, Appl. Optics 21, 2978 (1982). [39] Y. Nigara, Jpn. J. Appl. Phys. 7, 404 (1968). [40] J. R. DeVore, J. Opt. Soc. Am. 41, 416 (1951). [41] D. Wicaksana, A. Kobayashi, and A. Kinbara, J. Vac. Sci. Technol. A 10, 1479 (1992). [42] A. Dakka, J. Lafait, M. Abd-Lefdil, and C. Sella, M.J.Condensed matter 2, 153 (1999). [43] H. Tang, H. Berger, P. Schmid, and F. Levy, Solid State Communications 92, 267 (1994). [44] P. G. R. Dannenberg, Thin Solid Films 360, 122 (2000). [45] G. E. Jellison et al., J. Appl. Phys. 93, 9537 (2003). [46] L. Miao, P. Jin, K. Kaneko, A. Terai, N. Nabatova-Gabain, and S. Tanemura, App. Surface Sci. 212-213, 255 (2003). [47] D. E. Groom, C. J. Bebek, M. Fabricius, A. Karcher, W. F. Kolbe, N. A. Roe, and J. Steckert, in Sensors, Cameras, and Systems for Scientific/Industrial Applications VII , Proc. SPIE, Vol. 6068, edited by M. M. Blouke (2006) pp. 133–143. [48] M. H. Fabricius, C. J. Bebek, D. E. Groom, A. Karcher, and N. A. Roe, in Sensors, Cameras, and Systems for Scientific/Industrial Applications VII , Proc. SPIE, Vol. 6068, edited by M. M. Blouke (2006) pp. 144–154. [49] S. Kawanomoto et al., in prep. 25 [50] Blacksberg et al. report measured film indices somewhat different than were expected[64]. For SiO2 they find a film index about 2% higher than for bulk fused quartz, in the right direction but not enough to explain our apparent excess. [51] S. A. Smee, J. E. Gunn, A. Uomoto, N. Roe, D. Schlegel, et al., Astron. J. 146, 32 (2013). [52] B. Flaugher, H. T. Diehl, K. Honscheid, et al., Astron. J. 150, 150 (2015), arXiv:1504.02900. [53] C. Rockosi, R. Stover, et al., in Ground-based and Airborne Instrumentation for Astronomy III , SPIE, Vol. 7735 (2010) p. 77350R. [54] C. Bebek, J. Emes, D. Groom, S. Haque, S. Holland, A. Karcher, W. Kolbe, J. Lee, N. Palaio, and G. Wang, J. Instr. 10, C05026 (2015). [55] Hionix, Inc., hionix.com. [56] J. I. Pankove, Optical processes in semiconductors (Dover Publications, New York, 1971) pp. 36–42. [57] J. B. Oke et al., P. A. S. P. 107, 375 (1995). [58] J. R. Janesick, Scientific charge-coupled devices, Bellingham, WA: SPIE Optical Engineering Press, 2001, xvi, 906 p. SPIE Press monograph, PM 83 (2001). [59] I. S. McLean, ed., Electronic imaging in astronomy. Detectors and instrumentation (UK Wiley, Chich- ester, 1997). [60] K. B. Stevenson, J. L. Bean, A. Seifahrt, et al., Astrophys. J. 817, 141 (2016), arXiv:1511.08226. [61] Andreas Seifahrt (Univ. of Chicago), personal communication (2016). [62] P. Martini (Ohio State Univ.), personal communication (2012). [63] S. Nikzad, A. D. Jewell, M. E. Hoenk, T. Jones, J. Hennessy, T. Goodsall, A. Carver, C. Shapiro, S. R. Cheng, E. Hamden, G. Kyne, D. C. Martin, D. Schiminovich, P. Scowen, K. France, S. McCandliss, and R. E. Lupu, ArXiv e-prints (2016), arXiv:1612.04734 [astro-ph.IM]. [64] J. Blacksberg, S. Nikzad, M. E. Hoenk, S. E. Holland, and W. F. Kolbe, IEEE Transactions on Electron Devices 55, 3402 (2008). [65] B. C. Jacquot, S. P. Monacos, T. J. Jones, J. Blacksberg, M. E. Hoenk, and S. Nikzad, in High Energy, Optical, and Infrared Detectors for Astronomy IV , Proc. SPIE, Vol. 7742 (2010) p. 77420I.
1710.04857	1	1710	2017-10-13T09:48:25	Power Synthesis of Maximally-Sparse Linear Arrays Radiating Shaped Patterns through a Compressive-Sensing Driven Strategy	[ "physics.app-ph", "cs.IT", "cs.IT" ]	We present an innovative approach to the synthesis of linear arrays having the least possible number of elements while radiating shaped beams lying in completely arbitrary power masks. The approach, based on theory and procedures lend from Compressive Sensing, has two innovative key features. First, it exploits at best the multiplicity of equivalent field solutions corresponding to the many different power patterns lying in the given mask. Second, it a-priori optimizes those parameters that affect the performance of Compressive Sensing. The overall problem is formulated as two convex programming routines plus one local optimization, with the inherent advantages in terms of computational time and solutions optimality. An extensive numerical comparison against state-of-the-art procedures proves the effectiveness of the approach.	physics.app-ph	physics	Power Synthesis of Maximally-Sparse Linear Arrays Radiating Shaped Patterns through a Compressive-Sensing Driven Strategy Andrea Francesco Morabito, Antonia Rita Laganà, Gino Sorbello, and Tommaso Isernia1  Abstract-We present an innovative approach to the synthesis of linear arrays having the least possible number of elements while radiating shaped beams lying in completely arbitrary power masks. The approach, based on theory and procedures lend from Compressive Sensing, has two innovative key features. First, it exploits at best the multiplicity of equivalent field solutions corresponding to the many different power patterns lying in the given mask. Second, it a-priori optimizes those parameters that affect the performance of Compressive Sensing. The overall problem is formulated as two convex programming routines plus one local optimization, with the inherent advantages in terms of computational time and solutions' optimality. An extensive numerical comparison against state-of-the-art procedures proves the effectiveness of the approach. Index Terms-Compressive Sensing, Maximally Sparse Arrays, Shaped Beams, Power Synthesis. I. INTRODUCTION The synthesis of non-uniformly spaced arrays [1] is of interest in a large number of applications. While the first design techniques date back to the Sixties (see for instance [1],[2]), the problem has been the subject of a renewed interest in the last years [3]-[20]. In particular, attention has been devoted to the synthesis of 'maximally-sparse' arrays [6]-[15], i.e., arrays fulfilling assigned mission requirements by exploiting the minimum number of active elements. Many different solution approaches have been proposed to deal with such problem, recently including global optimization [15], the Forward Backward Matrix Pencil Method (FBMPM) [11]- [13], and the Linear Programming (LP) procedure proposed in [14]. Finally, an important body of work is given by a number of techniques based on the Compressive Sensing (CS) theory [6]-[10]. On the other side, a brief look to the characteristics of the different approaches shows that there is indeed room for improvement, as detailed below. First, it is interesting to note that the MPM and FBMPM, although being very effective, have some theoretical flaw (as the authors themselves state at the end the introduction of [12]). In fact, in order to find the locations of the array elements, the MPM requires that the poles of the equation identified by organizing the desired pattern data as an Hankel matrix and performing its singular value decomposition lie on the unit circle in the complex plane [11]. However, this property usually does not hold true in the case of shaped beams, this being the main reason of replacing the MPM with the FBMPM in the synthesis of such kind of fields [12]. On the other side, as stated by authors, the additional relationships enforced by FBMPM, although being very useful, represent a necessary but not a sufficient condition in order to guarantee the unitary amplitude of the mentioned poles [12], so that it is still possible to further improve the achieved results. Second, both the FBMPM and the existing CS-based approaches aim to fit at best both the amplitude and phase distributions of a reference far-field pattern. Such a circumstance completely ignores the fact in array antennas synthesis one is generally interested in power distributions, than amplitude and phase far-field distributions. rather Obviously, a priori establishing given phase distributions implies a reduction of the degrees of freedom available to the designer, thus reducing the number of possible solutions [21]. The approach in [14] apparently overcomes such a limitation, as a formulation is given in terms of a 'mask' for the pattern. Unfortunately, this technique is based on the assumption that the radiated field is a real function, while it is known that in case of shaped beams the optimal far-field pattern, i.e., the one with the least dimension for given performances, is in general a complex function [21]. Hence (see Sections II and IV), a number of possibly convenient solutions are hidden to the designer also in such a case. In the following, we propose, discuss and test a new approach to the synthesis of shaped patterns by means of (maximally) sparse linear arrays. The approach is able to take into account the multiplicity of possible fields able to fulfill given upper and lower bounds on the power pattern. Although developed for the case of linear arrays, the approach can be extended to the case of patterns requiring a circular symmetry, and it can be used in the case of two-dimensional factorable patterns, which are of actual interest in several applications. The paper is organized as follows. In Section II we recall basic results of CS and discuss its strengths and weaknesses in relationship with synthesis problems. Moreover, we also briefly recall useful results already available in the synthesis of shaped beams, with particular emphasis on the case of linear arrays. The proposed approach is presented and assessed in Sections III and IV, respectively, the latter providing successful numerical comparisons with all methods and results in [7],[10],[12],[14]. Conclusions follow. II. A COUPLE OF BASIC RESULTS In order to better clarify the usefulness of CS in synthesis problems, as well as the meaning and the interest of the approach we are proposing, in this Section we briefly review a couple of general theoretical results respectively related to the CS theory and to the synthesis of shaped beams. II.1 Interest and weaknesses of Compressive Sensing in synthesis problems The well-known circumstance that the aperture source and its far-field distribution are essentially related by a Fourier Transform constitutes a formidable basis in order to apply to array synthesis all theories and procedures arising from [22] and related (see for instance [23],[24]). Roughly speaking, the measurements one would use in the recovery of a sparse signal are substituted by the far-field specifications, while the array layout and excitation represent the sparse signal to be 'recovered' through CS [6]-[10]. In applying CS to such a problem, the common approach is that of looking for a fitting of both the amplitude and phase distributions of a reference far-field pattern, and the problem's actual unknowns are both the array elements excitations and locations. In pursuing such a goal, inspired from [22]-[24], the crucial antenna-design operation is usually that of minimizing the (weighted) ℓ1 norm of excitations [6]-[10], as defined on a very dense auxiliary grid. One may argue that both terms 'recovery' and 'Compressive Sensing' are questionable, as one cannot be sure that a sparse array able to realize a given pattern certainly exists, so that one could talk in terms of 'sparsity enhancing' rather than of 'Compressive Sensing'. On the other side, if a sparse array able to fulfill the assigned technical requirements exists then the CS theory provides the rigorous conditions and the actual algorithms allowing to unveil it, so that the adopted terminology (and framework) makes indeed sense. The CS-based procedures in [6]-[10] show excellent capabilities for the synthesis of sparse arrays. On the other hand, they also have a number of drawbacks, and it is indeed the CS theory which allows understanding weaknesses and developing a number of improvements. First, while the CS theory provides theoretical conditions and practical procedures for a faithful recovery in case of linear problems, it must be stressed that a power-pattern synthesis problem, even when looking for a generic (non- sparse) source distribution, is indeed a non-linear problem. In fact, as already stated, one is generally interested in power to than amplitude and phase far-field distributions, rather distributions, a-priori establishing given phase and distributions, as it is done in the present CS-based procedures, implies a reduction of the degrees of freedom available to the designer, thus hiding possible solutions [21]. As a consequence, one should either develop a CS theory for non-linear problems, or devise possible ways take advantage from the fact that many different complex patterns could equally fulfill the given power pattern specifications (which is our choice, see Section III). A second weakness of present procedures can be understood from a number of results including Theorem 1.3 in [22] and/or condition (6) in [23]. In these references, it is shown that the recovery of a signal having an original dimension equal to N and a 'sparse' or 'compressed' size equal to S results successful with overwhelming probability only if the number of independent measurement (say M) exceeds a threshold which grows with both S and logN [22]- [24]. Now, in the CS application to array synthesis, M, N, and S respectively map to the cardinalities of far-field specifications, initial array layout (including elements with zero excitation), and final sparse array layout (counting only active elements) [6]-[10]. As a consequence, since any field radiated by a finite-dimensional source has a finite number of degrees of freedom (and hence a limited value of M) [25], a very dense original array layout (and hence a very large value of N) may imply a violation of the conditions under which CS guarantees an effective recovery, i.e., an effective array synthesis. As a consequence, one should be careful about the choice of N, i.e., of the initial grid wherein CS is eventually applied, while still allowing in the overall procedure to locate the active antennas anywhere. II.2 A theoretical result in the synthesis of linear sources A number of procedures for the synthesis of shaped beams through standard 'non-sparse' arrays (such as the Woodward Lawson technique [26] and Gradient Search and Conjugate Matching) are reviewed in [20]. Another relevant and widely diffused approach to synthesis (which includes conjugate matching) is the so-called Alternating Projection technique [27]. On the other side, all these synthesis methods do not take into account that the synthesis of sources able to realize power patterns lying within a given mask is indeed a non-linear and non-convex problem. As a consequence, the final result of all the above synthesis methods depends indeed on the initial assumptions or on the starting point of the adopted algorithm. In fact, an exceptionally large number of procedures based on global optimization procedures has been devised (see [28],[29] and references therein). In case of equispaced linear arrays a completely different approach, which overcomes the non-convexity of the problem as well as the computational burden (and weaknesses [30]) associated to global optimization, is possible [21]. The approach can be naturally extended to the case of continuous sources by considering the limiting case of vanishing small spacings (and adding constraints on the invisible part of the spectrum) [31]. Such an alternative approach is based on the circumstance that the square-amplitude array factor of a linear equispaced array of Q antennas can be expressed as uP )(  Q  1  Qq  qeD 1 jqu (1) with u=βdsinα and β=2π/λ, λ, α, and d being the wavelength, the angle between the boresight and observation directions, and the spacing between adjacent elements, respectively. Since P must be a real and non-negative function, the conditions DD *   q q q  ,2,1 ..., Q  1 where * means complex conjugation, and Q  1  Qq  qeD 1 jqu  0 (2) (3) must hold true. Hence, one can identify a power pattern fulfilling given shape constraints by solving the following optimization problem: jqu  uUB )( jqu  uLB )(  ,2,1 ..., Q  1 Q 1   eD q Qq 1  Q 1   eD q Qq 1  qD *  q  Q 1              D q eD q Qq 1  jqu  0 (4.a) (4.b) (4.c) (4.d) in Then, wherein the arbitrary (real and non-negative) functions UB and LB respectively denote the upper and lower bounds exploited to shape the field amplitude according to the particular application at hand. By taking into account the band-limitedness of P, expressions (4) can be substituted with a sufficiently fine discretization [21],[25] so that it can be seen as a system of ordinary linear inequalities in the Dq coefficients. Notably, such a problem can be easily solved by using any LP tool. factored (1) can be representation two polynomials being the complex conjugate each of the other, which can be interpreted as the array factor and its complex conjugate. As a consequence, one is able to solve the shaped beam synthesis problem through solution of (4) and suitable factorization techniques [21]. Notably, more than a single solution can be found, as the factorization of the reference power pattern P in complex conjugate factors is generally non- unique, so that the same power pattern corresponds to a (generally large) number of different far-field distributions. All of the solution can be easily generated, at least in principle, by means of the so called 'zero-flipping' operation [21]. With respect to the specific subject of this paper, the above result has a two-fold usefulness. First, it offers a natural way field to specifications, thus allowing to exploit the standard (linear) power-pattern specifications translate into CS framework. Second, the consideration of all possible field determinations allows to keep under control (and take advantage from) the multiplicity of solutions available for each fixed power pattern. III. THE PROPOSED APPROACH TO SYNTHESIS the The aim of the procedure is to perform the power synthesis of a shaped beam lying in a fixed and arbitrary mask through an array antenna composed by the minimum number of elements. By taking into account limitations discussed in Section II.1, and taking advantage from the result recalled in Section II.2, the power synthesis problem at hand has been conveniently three-steps procedure. formulated as following In the first step, determine some convenient nominal power pattern to be pursued, and find all the different field patterns corresponding to it. In the second step, by taking advantage from the field patterns determined in step one, use CS procedures in order to synthesize a sparse array having the minimum possible number of elements. The above two steps allow taking relevant advantage from the fact that a single power pattern corresponds to a large number of different patterns, but they still do not exploit all the available degrees of freedom. In fact, there may be many different power patterns lying in the same mask. The third step is hence devoted to perform a refinement of the solution by taking advantage from this circumstance. In particular, modifications of the initial power pattern are pursued in such a way to realize a further sparsification of the array layout. Obviously, the original mask is still enforced on the on the final power pattern. Details on the three steps are respectively given in the three following subsections. III.1. Step 1: Exploiting the multiplicity of solutions for a given power pattern Let us suppose a solution to (4) has been found. Then, by introducing the auxiliary variable z=eju, and exploiting the Fundamental Theorem of Algebra, representation (1) can be factored in terms of its zeroes. Their proper clustering allows the factorization of the reference power pattern P in complex conjugate factors. Notably, one will have many different clustering possibilities corresponding to number of different far-field distributions. To express this multiplicity of solutions, let us consider the matrix A  1 a a 2  Wa             (5) where W is the overall number of different array factors corresponding to P and the vector aw=[aw,1,aw,2,…,aw,Q] denotes the excitation vector corresponding to the w-th array factor for w=1,2,…,W. As discussed in [21], the number of equivalent solutions is W=2R'/2, R' being the number of zeroes of P which do not lie on the unit circle. As W grows very rapidly with R', it makes sense to discuss the choice of the parameters Q and d, which determine the total number of roots of P, i.e., R=2Q-2, and the reference-array size, i.e., T=(Q-1)d. Due to the bandlimitedness of the fields radiated by non- superdirective sources [25],[31], the actual feasibility or unfeasibility of a power pattern mask is essentially determined by the antenna's electrical size T/λ. In fact, for a fixed T, increasing Q means increasing R, thus apparently furnishing a larger number of degrees of freedom. On the other side, increasing Q also means to reduce the array elements spacing. The reduction of d, in turn, forces the user to add constraints in the invisible part of the spectrum in order to avoid superdirective sources [21],[31]. As a consequence, the additional roots which are apparently available must be used to bind the antenna's reactive energy (and are indeed expected to lie on the unit circle). Therefore, provided d≤λ/2, different values of Q and d corresponding to the same value of T are expected to lead to very similar radiation performances in the visible part of this circumstance in a large set of synthesis problems, solving each of them by choosing d=0.5λ, d=0.4λ, and d=0.3λ. At the end of the overall proposed approach, they all led to identical results in terms of both radiation performance and sparse array's final number of elements. the spectrum. We have verified Let us discuss now the choice, amongst all the equivalent solutions (5), of the field to be pursued through CS. A first possibility could be that of applying a CS-based procedure to all the different fields arising from the factorization step, and then to observe the resulting number of active array elements. A similar strategy has been adopted in [12] by Liu and co- workers which, starting from a power solution in [21], have applied the FBMPM to all the equivalent field solutions. A second possibility could be that of a-priori selecting (by exploiting some convenient ad-hoc criterion) one amongst the different field distributions fulfilling the initial constraints. This strategy, which is the one exploited by our approach, may allow avoiding a heavy computational burden in those test cases having a very large number of equivalent solutions. Following the CS-theory [22]-[24], an intuitive choice is to consider the excitations set corresponding to the array layout with the maximum amount of sparsity. Therefore, amongst all the available far-field distributions, a convenient choice is to select Q f r      ea qr , jq d  sin q  1 where fr(u)fr*(u)=P(u), and w  a a r 1 w 1  ,...,2,1 W (6) (7) as reference field for step 2 of our procedure. In fact, according to the CS-theory concerning relaxation of ℓ0-norm optimization problems, fr represents, by construction, the far-field distribution that more easily lends itself to a 'sparsification' process (see [22]-[24] for further details). As discussed with the help of some examples in the Section IV, application of (7) works very well even in those cases where d is chosen in the order of λ/2, which can be attributed again to the finite number of degrees of freedom of finite-dimensional sources [25]. III.2. Step 2: CS-driven engine to recover a preliminary array layout Once the 'optimal' reference far-field distribution has been identified, we apply a CS-inspired procedure to synthesize it. We denote with fb the optimized far field and with the vectors x and b the locations and excitations of the N-elements linear array radiating it, respectively. such as greedy approaches Many different procedures are commonly used in sparse recovery, iterative thresholding [33], Basis Pursuit [34] or Least Absolute Shrinkage and Selection Operator (LASSO) [35]. Taking into account the theoretical results presented in [22], we adopt an approach based on a ℓ1 relaxation of the original problem, which is similar in spirit to the LASSO approach. In particular, we formulate the problem as follows: [32], min b b 1 subject to         2    f b  g   1   f b    f r r f        2 2   2 wherein f b N      n  1 xj  n sin eb n (8.a) (8.b) (8.c) (9) Notably, the problem is reduced in such a way to a Convex Programming (CP) optimization. In such a formulation:  the cost function (8.a) is based on CS theory and has the same aim of the functional minimized in [7]-[10], i.e., to induce a minimization of the arrays' active elements number;  constraint (8.b) is devoted to control the sidelobes amplitude and enforces an upper bound g(α) on the synthesized power pattern in the angular region denoted with τ1. Note g can be an arbitrary (real and positive) function so that field 'notches' can be eventually realized;  constraint (8.c) ensures that, in the angular region denoted by τ2, the optimized array factor fits the reference one with a precision ε. Notably, given its mathematical formulation, provided an intersection between constraints (8.b) and (8.c) does exist, problem (8) provides always a unique solution provably representing the global optimum. Moreover, provided conditions for the validity of CS sparse recovery hold true, such a solution also will be a maximally sparse solution to the synthesis problem [6]-[10]. As a final comment on this step, let us discuss the choice of N and x in (9). Once the array aperture size has been fixed on the basis of the particular radiation and geometrical specifications, following the theory in [6]-[10], we define x as a vector uniformly spanning it. Note that, in order to optimize the CS performance, we use values of N considerably smaller than those adopted in previous procedures. Such a choice (see also Section II.1) will boost potentialities of CS from both point of views of speed of convergence and optimality of final results [22]-[24]. On the other side, possible accuracy losses in identifying the array locations, deriving from choosing in such a step an elements spacing larger than in other approaches, are then recovered in the third step of the design procedure. III.3. Step 3: Further sparsification and solution refinement The last part of the synthesis is devoted to a further 'sparsification' and optimization of the array layout. In the first part of it, which is similar in spirit to the so- called iterative thresholding approaches [33], we further reduce the number of array elements by: 1) discarding the antennas having an excitation amplitude lower than a threshold υ; 2) recursively substituting each couple of remaining elements whose distance is lower than a threshold σ with a single element placed in the middle point between them. In the following, the elements number and the locations set coming out from these operations are denoted with S and y, respectively, while a sketch illustrating such modifications is given in Fig. 1. Fig. 1. Third step of the design procedure: sketch concerning the sparsification of the array layout coming out from step 3. Then, the solution is further improved by determining slight shifts on array locations (and the optimal excitations associated to them) so as to improve the overall radiation performances without increasing the number of elements. In particular, we look for vectors c and Δy such that the final far field distribution S f y 1      s  1  j  y s y  s  sin  ec s (10) is given by the solution the following optimization problem: min c y ,  f y 1    f r r f        2 2   2 subject to 2    f y 1  g    1  (11.a) (11.b) In fact, provided that the minimum value of the functional in (11.a) does not result larger than ε, solving problem (11) guarantees a radiation performance equivalent to the one coming out from problem (8) while saving N-S array elements. Unfortunately, due to the way the unknown Δy appears in (10), the optimization problem (11) is a non-convex one. On the other side, one is just looking for 'small' shifts, so that (10) can be reasonably approximated by S f y 2      s  1 ec s yj  s sin  1(  j   y s sin )  (12) As a consequence, the last step can be formulated as in (11) by using fy2 instead of fy1 and adding the convex constraint    ys s (13) being η a user-defined constant such that 0<η<<1. Notably, even if the optimization problem, unlike (4) and (8), is still not of the CP class, it is nearly so by virtue of (13), so that occurrence of sub-optimal solutions can be avoided while local optimization procedures can still be used. Finally, the adopted formulation allows the addition of the convex constraint       y S S s y 1 S  1 S  1 S    2,1   ,..., y  y  (14) as a powerful way to avoid a too small spacing between adjacent elements (possibly leading to undesired mutual- coupling effects). It is worth noting that this last step allows counteracting possible drawbacks deriving from a value of N smaller than in other CS-based approaches, and compensating for possible mismatches deriving from the 'discarding' and 'unification' of elements described above. Moreover, it pursues a field fitting just in the main-beam zone, while upper bounds are used for the sidelobes regions. As already stated, the possibility of modifying the power pattern with respect to the nominal one fixed in step 1 allows recovering a significant number of degrees of freedom with respect to more usual approaches. IV. NUMERICAL ASSESSMENT In order to assess the actual capabilities of the proposed approach, we have compared its outcomes with those of the state-of-the-art techniques available for the synthesis of shaped beams through 'maximally-sparse' arrays. In order to provide a comparison as complete as possible, we discuss in the following six test cases. In particular, the first two examples respectively concern a comparison with two techniques which tackle the problem at hand without exploiting the CS theory, namely the LP procedure presented in [14] and the FBMPM introduced in [12]. Then, in the third, fourth, and fifth numerical examples two state-of-the-art techniques exploiting the CS theory, namely the Bayesian Compressive Sensing (BCS)-based procedure described in [7] and CS-based weighted-norm approach proposed in [10], are considered. Finally, the sixth test case concerns a simple extension of the approach to the synthesis of a planar array. By the sake of clarity, we resume in Table 1 the main parameters used for the different linear arrays. Test case W (Number of equivalent solutions) ε (Fitting precision inside 2 region) Cfr. vs. [14] (flat-top) Cfr. vs. [12] (cosecant) Cfr. vs. [7] (flat-top) Cfr. vs. [7] (cosecant) Cfr. vs. [10] (cosecant) 512 2x10-2 32 17x10-2 16 5x10-3 32 32 17x10-2 22x10-2 (Fitting region) 2 L/λ (Size) ψ/λ (Minimum elements spacing) S (Final elements number) Elements number saving vs. best case available in literature sinα≤0.37 -0.05≤sinα≤0.85 sinα≤0.31 0≤sinα≤0.83 -0.18≤sinα≤0.59 11.68 0.40 13 58% 7.28 0.48 11 15% 6.80 0.46 10 9% 6.93 0.51 12 8% 7.53 0.55 12 8% Tab. I. Simulation parameters adopted for the test cases concerning one-dimensional arrays. Notably, by virtue of the extremely low computational burden of the overall design algorithm, each example required less than 15 seconds to be generated by a calculator having an Intel Core i7-3537U 2.50GHz CPU and a 10 GB RAM. Finally, it is worth noting that, although the proposed procedure can manage any element factor in the synthesis, all the following power-pattern figures just refer to the square- amplitude distribution of the array factors. approach, we identified a 13-elements sparse array able to radiate the power-pattern distribution depicted in red color in Fig. 2. As it can be seen, the synthesized solution results equivalent, in terms of radiation performance, to the reference one. Therefore, the proposed procedure has allowed, with respect to the best solution shown in [14], a 58% elements saving without experiencing any power-pattern worsening. The synthesized sparse array's excitation amplitudes and phases are respectively depicted in figures 3 and 4. Fig. 2. Reference (blue curve, from [14]) and synthesized (red curve) power pattern in the test case of Susection IV.1. Exploited upper-bound constraint (green curve) is also depicted. Simulation parameters: d/λ=0.5; Q=29; N=201; ν=0.05; σ/=0.1; η/=0.025. Fig. 3. Excitation amplitudes of the maximally-sparse array radiating the red pattern of Fig. 2. IV.1 Comparison with [14] (flat-top beam) We adopted as reference the power pattern depicted in blue color in Fig. 2, which has been synthesized in [14] by means of a 31-elements sparse array. This flat-top field guarantees a ripple equal to 0.4455 dB for -20°≤α≤20°, and fulfills an upper-bound constraint (depicted in green color in Fig. 2) of - 30 dB for -90°≤α≤-25° and 25°≤α≤90°. By factorizing this power pattern, we have been able to determine 512 equivalent field solutions. Then, we have selected amongst them the one corresponding the minimum ℓ1 norm and, by applying to the latter the proposed the excitation distribution having to IV.2 Comparison with [12] (cosecant beam) We used as reference square-amplitude array factor the power pattern depicted in blue color in Fig. 5. In [12], the FBMPM allowed generating it thorugh a sparse array composed by 13 elements located over an aperture of 7.5λ. By exploiting the proposed approach, we have been able to generate the power pattern depicted in red color in Fig. 9 by exploiting a sparse array composed by 11 elements located over an aperture of 7.4λ. Therefore, without experiencing any loss in terms of radiation performance, and exploiting a practically equivalent aperture size, we reduced of the 15.4% the elements number achieved in [12]. The amplitude and phase distributions of the synthesized array excitations are shown in figures 6 and 7, respectively. Fig. 4. Excitation phases of the maximally-sparse array radiating the red pattern of Fig. 2. Fig. 7. Excitation phases of the maximally-sparse array radiating the red pattern of Fig. 5. IV.3 Comparison with [7] (flat-top beam) In the third test case, we used as reference power pattern the flat-top beam depicted in blue color in Fig. 8, which has a maximum ripple of ±0.58 dB, a peak sidelobe level of -35.6 dB with respect to the maximum power pattern value for α≥27.5°. In [7], BCS has been effectively exploited to generate it through a sparse array composed by 11 elements located over an aperture of 7λ. Also, this power pattern has been generated in [12] by exploiting the FBMPM and using 10 radiating elements. Fig. 5. Adopted upper bound (green curve) and reference (blue curve, from [12]) and synthesized (red curve) power patterns for the second test case. Simulation parameters: d/λ=0.5; Q=16; N=321; ν=0.005; σ/=0.025; η/=0.05. Fig. 8. Third test case: adopted upper bound (green curve); reference (blue curve, from [7]) and synthesized (red curve) power patterns. Simulation parameters: d/λ=0.5; Q=15; N=38; ν=0.04; σ/=0.19; η/=0.09. As first design operation, we have generated this power pattern without changing any parameter with respect to [7] (see Table 1). Then, we have factorized the resulting polynomial and applied our CS-based routine to all the equivalent far-field distributions corresponding to it, with the aim of assessing the effectiveness of the adoption of (7). In particular, in each instance we have analyzed the ℓ1 norm of the excitation set of the reference array factor and counted the active elements of the sparse array coming out from the CS routine. The outcomes of such experiments have been Fig. 6. Excitation amplitudes of the maximally-sparse array radiating the red pattern of Fig. 5. summarized in Fig. 9. As it can be seen, the excitation set having the minimum ℓ1 norm is also the one that leads to the minimum number of active elements in the array designed through CS. This circumstance (which has been also verified in the previous and following test cases) confirms the arguments in Section III.1 about the capability of identifying in a simple fashion the far-field distribution which more easily lends itself to a CS-based 'sparsification' process. Notably, in this test case the results achieved by means of the proposed strategy are equivalent to those one reported in [12], while they turn out being improved with respect to [7] in all terms of number of elements, dimensions and performances. Finally, we highlight the important circumstance that the execution of the third step of the procedure allowed us to reduce N of about the 50% with respect to the case wherein just the first two steps of the approach are exploited. Fig. 9. Concerning the impact on the CS performance of the ℓ1 norm of the excitations of the different solutions coming out from the statement of the reference power pattern. Fig. 11. Excitation phases corresponding to the red pattern of Fig. 8. IV.4 Comparison with [7] (cosecant beam) We adopted as reference power pattern the square-cosecant field shown in blue color in Fig. 12. In [7], by adopting BCS this pattern has been synthesized through a sparse array composed by 13 elements located over an aperture of 7.5λ. By exploiting the proposed approach, we have been able to reduce again the array's number of elements and size, respectively, without any radiation-performance loss. Fig. 10. Excitation amplitudes corresponding to the red pattern of Fig. 8. By applying the post-processing procedure to the array layout having the minimum number of elements at the end of the CS routine, we have achieved a radiation performance which favorably compares to [7] through a sparse array composed by 10 elements located over an aperture of 6.8λ with a minimum spacing of 0.46λ. The achieved array locations and excitations are depicted in Fig. 10 (amplitude distribution) and Fig. 11 (phase distribution), respectively, while Fig. 8 shows the exploited upper-bound function and a superposition of the reference and synthesized square-amplitude array factors. As it can be seen, the achieved power pattern perfectly fits the reference one. Fig. 12. Synthesis of a maximally-sparse array radiating a square-cosecant pattern with a specific sidelobes decay. Reference (blue curve, from [7]) and synthesized (red curve) square-amplitude array factors. Adopted upper bound (green curve) also depicted. Simulation parameters: d/λ=0.5; Q=16; N=161; ν=0.03; σ/=0.05; η/=0.01. This circumstance is shown in Fig. 12, wherein a comparison between reference and synthesized patterns is reported. The designed array is composed by 12 elements located over an aperture of 6.9λ with a minimum spacing of 0.51λ. The corresponding excitation amplitudes and phases are depicted in figures 13 and 14, respectively. exploited to bind the sidelobes as desired, which also points out the flexibility of the overall synthesis procedure. Fig. 13. Excitation amplitudes of the maximally-sparse array radiating the red pattern of Fig. 12. Fig. 15. Synthesis of a maximally-sparse array radiating a square-cosecant pattern with a non-uniform, piecewise-constant sidelobe level behavior. Adopted upper bound (green curve); reference (blue curve, from [10]) and synthesized factors. Simulation parameters: d/λ=0.5; Q=16; N=161; ν=0.005; σ/=0.05; η/=0.08. square-amplitude array (red curve) Fig. 14. Excitation phases of the maximally-sparse array radiating the red pattern of Fig. 12. Fig. 16. Array excitation amplitudes corresponding to red pattern of Fig. 15. IV.5 Comparison with [10] (cosecant beam) In the fifth test case, we aimed at synthesizing the square- cosecant power pattern depicted in blue color in Fig. 15. In [10], CS has been exploited to generate it by means of a sparse array composed by 13 elements located over an aperture of 8λ. Notably, by means of the presented approach, we have achieved an equivalent radiation performance while reducing both the elements number and the aperture size. In particular, the synthesized sparse array is composed by 12 elements located over an aperture of 7.5λ with a minimum spacing of 0.56λ. The achieved antenna layout and excitations are depicted in Fig. 16 (amplitude distribution) and Fig. 17 (phase distribution), respectively. A superposition of the reference and synthesized patterns is shown in Fig. 15, wherein it can be noticed that all the design goals and constraints have been fulfilled. The same figure shows the upper-bound function Fig. 17. Array excitation phases corresponding to the red pattern of Fig. 15. IV.6 Synthesis of a factorable beam through a planar array As a final test case, we show the outcomes of a numerical experiment concerning a simple way to extend the proposed procedure to the synthesis of planar sparse arrays with factorable patterns. Coming to details, we aimed at synthesizing a beam having a flat-top behavior along the azimuth angle and a square- cosecant behavior along the elevation angle, which is of interest in radar applications as well as for radio-base stations. In so doing, we took as a reference the pattern generated from the factorization of the fields depicted in blue color in figures 8 and 12, respectively. Notably, exploitation of the results in [7] would require 13x11=143 elements to generate the desired radiation pattern. On the other side, a straightforward application of our results in Subsections IV.3 and IV.4 already reduces to 12x10=120 the number of required elements. This circumstance is coherent with the fact that, as long as a pattern is factorable, the elements number reduction in the planar array is roughly doubled with respect to the one experienced in the two underlying one-dimensional arrays. Fig. 18. Array layout achieved by factorizing the one-dimensional solutions shown in Subsections IV.3 and IV.4 and discarding those elements having a normalized excitation amplitude lower than 0.04: elements' location before (blue circles) and after (red dots) running the final 'refinement' algorithm. Fig. 19. Normalized excitation amplitudes associated to the refined layout shown in Fig. 18. Fig. 20. Power pattern of the planar array having the refined layout of Fig. 18 and the excitations whose amplitude distribution is shown in Fig. 19. (a) (b) Fig. 21. Main cuts along ordinate [subplot (a)] and abscissa [subplot (b)] of the power pattern shown in Fig. 20: reference (blue color) and synthesized (red color) solutions. Interestingly, a simple application of the third step of the proposed procedure, i.e., of the techniques presented in Subsection III.3, allows a further considerable reduction of elements. In fact, by simply 'eliminating' those elements having an excitation amplitude lower than υ=0.04 we achieved the array layout depicted in Fig. 18 (blue circles), which is composed by 94 elements. Then, by adapting to the two- dimensional layouts case the approach in (10),(11), in such a way to jointly refine both the x and y array locations, we finally achieved the layout depicted in Fig. 18 (red dots), and the corresponding optimal excitations depicted in Fig. 19 (just amplitude is shown). The achieved power pattern is shown in figures 20 (in the two-dimensional spectral plane) and 21 (along its two main cuts), wherein θ and ϕ respectively represent the elevation and azimuth angles. As it can be seen, a very good agreement is achieved between the reference field and the main cuts of the synthesized pattern. Notably, the overall approach allows to save roughly the 58% of the elements with respect to a fully populated array with a uniform /2 spacing, and roughly the 34% of elements (tolerating a slight worsening of the very low sidelobe levels) with respect to a simple factorization of the solutions in [7]. IV. CONCLUSIONS We have introduced and tested a new approach, inspired from CS, for the synthesis of shaped beams by means of linear arrays having the minimum possible number of elements. The presented technique takes maximum advantage from the multiplicity of different far field and source distributions (having different amounts of sparsity) corresponding to a fixed power pattern, and also from the circumstance that different power patterns may equally fulfill the initial constraints. These features, as well as the use of a reduced cardinality of the tentative array, has allowed us to improve the amount of sparsity of the synthesized array in a number of benchmark problems present in the state-of-the-art literature. In particular, different methods based on LP [14], FBMPM [12], CS [10] and BCS [7] have been considered for comparison. In all cases, the proposed approach always performed equally or better than the considered techniques. The procedure can be easily extended to other cases of interest in several applications, including the design of arrays with factorable patterns (see Section IV.6) as well as one- dimensional simply reconfigurable arrays (by exploiting the ideas in [36]). REFERENCES [1] M. I. Skolnik, "Chapter 6. Nonuniform Arrays," in Antenna Theory, Part I, R. E. Collin and F. Zucker (eds), New York, McGraw-Hill, 1969. [2] R. E. Willey, "Space tapering of linear and planar arrays," IRE Trans. on Antennas and Propagation, vol. 10, n. 4, pp. 369-377, 1962. [3] T. A. Milligan, "Space-tapered circular (ring) array," IEEE Antennas and Propagation Magazine, vol. 46, n. 3, pp. 70-73, 2004. taper [4] O. M. Bucci, M. D'Urso, T. Isernia, P. Angeletti, and G. Toso, "Deterministic synthesis of uniform amplitude sparse arrays via new techniques," IEEE Transactions on Antennas and density Propagation, vol. 58, n. 6, pp. 1949-1958, 2010. [5] O. M. Bucci, T. Isernia, and A. F. Morabito, "An effective deterministic procedure for the synthesis of shaped beams by means of uniform- amplitude linear sparse arrays," IEEE Transactions on Antennas and Propagation, vol. 61, n. 1, pp. 169-175, 2013 [6] G. Oliveri and A. Massa, "Bayesian compressive sampling for pattern synthesis with maximally sparse non-uniform linear arrays," IEEE Transactions on Antennas and Propagation, vol. 59, n. 2, pp. 467-481, 2011. [7] G. Oliveri, M. Carlin, and A. Massa, "Complex-weight sparse linear IEEE array Transactions on Antennas and Propagation, vol. 60, n. 5, pp. 2309- 2326, 2012. synthesis by Bayesian Compressive Sampling," [8] G. Prisco and M. D'Urso, "Maximally sparse arrays via sequential convex optimizations," IEEE Antennas and Wireless Propagation Letters, vol. 11, pp. 192–195, 2012. [9] W. Zhang, L. Li, and F. Li, "Reducing the number of elements in linear and planar antenna arrays with sparseness constrained optimization," IEEE Transactions on Antennas and Propagation, vol. 59, n. 8, pp. 3106–3111, 2011. [10] B. Fuchs, "Synthesis of sparse arrays with focused or shaped beam pattern via sequential convex optimizations," IEEE Transactions on Antennas and Propagation, vol. 60, n. 7, pp. 3499-3503, 2012. [11] Y. Liu, Z. Nie, and Q. H. Liu, "Reducing the number of elements in a linear antenna array by the matrix pencil method, IEEE Transactions on Antennas and Propagation, vol. 56, n. 9, pp. 2955-2962, 2008. [12] Y. Liu, Q. H. Liu, and Z. Nie, "Reducing the number of elements in the synthesis of shaped-beam patterns by the forward-backward matrix pencil method," IEEE Transactions on Antennas and Propagation, vol. 58, n. 2, pp. 604-608, 2010. [13] Y. Liu, Q. H. Liu, and Z. Nie, "Reducing the number of elements in multiple-pattern linear arrays by the extended matrix pencil methods," IEEE Trans. on Antennas Propag., vol. 62, n. 2, pp. 652-660, 2014. [14] S. E. Nai, W. Ser, Z. L. Yu, and H. Chen, "Beampattern synthesis for linear and planar arrays with antenna selection by convex optimization," IEEE Transactions on Antennas and Propagation, vol. 58, n. 12, pp. 3923-3930, 2010. [15] A. Trucco and V. Murino, "Stochastic optimization of linear sparse arrays," IEEE Journal of Oceanic Engineering, vol. 24, n. 3, pp. 291- 299, 1999. [16] D. Caratelli and M. C. Viganò, "A novel deterministic synthesis for constrained sparse array design problem," IEEE technique Transactions on Antennas and Propagation, vol. 59, n. 11, pp. 4085- 4093, 2011. [17] D. Caratelli and M. C. Viganò, "Analytical synthesis technique for linear uniform-amplitude sparse arrays," Radio Science, vol. 46, n. 4, 2011. signal [18] A. K. Bhattacharyya, "Projection matrix method for shaped beam synthesis in phased arrays and reflectors," IEEE Transactions on Antennas and Propagation, vol. 55 n. 3, pp. 675-683, 2007. [19] A. F. Morabito, T. Isernia, and L. Di Donato, "Optimal synthesis of phase-only reconfigurable linear sparse arrays having uniform-amplitude excitations," Progress In Electromagnetics Research, vol. 124, pp. 405- 423, 2012. [20] A.K. Bhattacharyya, Phased Array Antennas: Floquet Analysis, Synthesis, BFNs, and Active Array Systems, Chapter 11, "Shaped-Beam Array Design: Optimization Algorithms," pp. 357-374, Wiley, 2006. [21] T. Isernia, O. M. Bucci, and N. Fiorentino, "Shaped beam antenna synthesis problems: feasibility criteria and new strategies," Journal of Electromagnetic Waves and Applications, vol. 12, pp. 103–137, 1998. [22] E. J. Candès, J. K. Romberg, and T. Tao, "Robust uncertainty principles: exact frequency information," IEEE Transactions on Information Theory, vol. 52, n. 2, pp. 489-509, 2006. reconstruction from highly incomplete [23] E. J. Candès and M. B. Wakin, "An Introduction To Compressive Sampling", IEEE Signal Processing Magazine, vol. 25, pp. 21–30, March 2008. [24] E. J. Candès, M. Wakin, and S. Boyd, "Enhancing sparsity by L1 Journal of Fourier Analysis and reweighted minimization," Applications, vol. 14, n. 5, pp. 877–905, 2008. [25] O. M. Bucci, G. Franceschetti, "On the degrees of freedom of scattered fields," IEEE Transactions on Antennas and Propagation, vol. AP-37, pp. 918-926, 1989. [26] P. M. Woodward and J. D. Lawson, "The theoretical precision with which an arbitrary radiation-pattern may be obtained from a source of finite size," Journal of the Institution of Electrical Engineers - Part III: Radio and Communication Engineering, vol. 95, n. 37, pp. 363-370, 1948. [27] O. M. Bucci, G. D'Elia, G. Mazzarella, and G. Panariello, "Antenna pattern synthesis: a new general approach," Proceedings of the IEEE, vol. 82, n. 3, pp. 358-371, 1994. [28] L. Manica, N. Anselmi, P. Rocca, and A. Massa, "Robust mask- constrained linear array synthesis through an interval-based particle SWARM optimization," IET Microw. Antennas Propag., vol. 7, n. 12, pp. 976-984, 2013. [29] A. Pirhadi, M. H. Rahmani, and Alireza Mallahzadeh, "Shaped beam array synthesis using particle swarm optimisation method with mutual coupling compensation and wideband feeding network," IET Microw. Antennas Propag., vol. 8, n. 8, pp. 549-555, 2014. [30] G. W. Wasilkowski, "On average complexity of global optimization problems," Mathematical Programming, vol. 57, n. 1-3, pp. 313-324, 1992. [31] O. M. Bucci, T. Isernia, A. F. Morabito, "Optimal synthesis of circularly symmetric shaped beams", IEEE Transactions on Antennas and Propagation, vol. 62, n. 4, pp. 1954-1964, 2014. [32] S. G. Mallat and Z. Zhang, "Matching pursuits with time-frequency dictionaries," IEEE Transactions on Signal Processing, vol. 41, n. 12, pp. 3397-3415, 1993. [33] T. Blumensath and M. E. Davies. "Iterative thresholding for sparse approximations," Journal of Fourier Analysis and Applications, vol. 14, n. 5, 2008. [34] S. Chen, D. Donoho D., and D. Saunders D., "Atomic decomposition by basis pursuit," SIAM Journal on Scientific Computing, vol. 20, n. 1, pp. 33-61, 1999. [35] R. Tibshirani, "Regression shrinkage and selection via the lasso," Journal of the Royal Statistical Society: Series B, vol 58, no.1, pp.267- 288, 1996. [36] A. F. Morabito, A. Massa, P. Rocca, and T. Isernia, "An effective approach to the synthesis of phase-only reconfigurable linear arrays," IEEE Transactions on Antennas and Propagation, vol. 60, n. 8, pp. 3622-3631, 2012.
1709.04062	1	1709	2017-09-12T21:24:51	Atomically-thin Femtojoule Filamentary Memristor	[ "physics.app-ph", "cond-mat.mes-hall" ]	The morphology and dimension of the conductive filament formed in a memristive device are strongly influenced by the thickness of its switching medium layer. Aggressive scaling of this active layer thickness is critical towards reducing the operating current, voltage and energy consumption in filamentary type memristors. Previously, the thickness of this filament layer has been limited to above a few nanometers due to processing constraints, making it challenging to further suppress the on-state current and the switching voltage. Here, we study the formation of conductive filaments in a material medium with sub-nanometer thickness, formed through the oxidation of atomically-thin two-dimensional boron nitride. The resulting memristive device exhibits sub-nanometer filamentary switching with sub-pA operation current and femtojoule per bit energy consumption. Furthermore, by confining the filament to the atomic scale, we observe current switching characteristics that are distinct from that in thicker medium due to the profoundly different atomic kinetics. The filament morphology in such an aggressively scaled memristive device is also theoretically explored. These ultra-low energy devices are promising for realizing femtojoule and sub-femtojoule electronic computation, which can be attractive for applications in a wide range of electronics systems that desire ultra-low power operation.	physics.app-ph	physics	Atomically-thin Femtojoule Filamentary Memristor Huan Zhao1, Zhipeng Dong2, He Tian1, Don DiMarzio3, Myung-Geun Han4, Lihua Zhang5, Xiaodong Yan1, Fanxin Liu6, Lang Shen1, Shu-jen Han7, Steve Cronin1, Wei Wu1, Jesse Tice3, Jing Guo2,, Han Wang1, 1Ming Hsieh Department of Electrical Engineering, University of Southern California, Los Angeles, CA 90089, USA 2Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32611, USA. 3NG Next, Northrop Grumman Aerospace Systems, 1 Space Park, Redondo Beach, CA 90278, USA. 4Condensed Matter Physics and Materials Sciences Department, Brookhaven National Laboratory, Upton, NY 11973, USA 5Center for Functional Nanomaterials, Brookhaven National Laboratory, Upton, NY 11973, USA 6Center for Optics & Optoelectronics Research (COOR), Collaborative Innovation Center for Information Technology in Biological and Medical Physics, and College of Science, Zhejiang University of Technology, Hangzhou 310023, PR China. 7IBM T. J. Watson Research Center, Yorktown Heights, NY 10598, USA *To whom correspondence should be addressed: [email protected] (H.W.), [email protected] (J.G.) The morphology and dimension of the conductive filament formed in a memristive device are strongly influenced by the thickness of its switching medium layer. Aggressive scaling of this active layer thickness is critical towards reducing the operating current, voltage and energy consumption in filamentary type memristors. Previously, the thickness of this filament layer has been limited to above a few nanometers due to processing constraints, making it challenging to further suppress the on-state current and the switching voltage. Here, we study the formation of conductive filaments in a material medium with sub-nanometer thickness, formed through the oxidation of atomically-thin two-dimensional boron nitride. The resulting memristive 1 device exhibits sub-nanometer filamentary switching with sub-pA operation current and femtojoule per bit energy consumption. Furthermore, by confining the filament to the atomic scale, we observe current switching characteristics that are distinct from that in thicker medium due to the profoundly different atomic kinetics. The filament morphology in such an aggressively scaled memristive device is also theoretically explored. These ultra-low energy devices are promising for realizing femtojoule and sub-femtojoule electronic computation, which can be attractive for applications in a wide range of electronics systems that desire ultra- low power operation. Resistive switching devices such as oxide based memristors[1-7] and conductive bridge random access memories (CBRAM)[8-10] utilize filamentary switching to realize low power memory and computation operations. In these devices, the formation and destruction of the conductive filament enables the reversible switching of the device conductivity,[11-13] resulting in compact two-terminal devices well suited for miniaturization and integration.[14-15] These devices are promising candidates for applications in low-current, low power memory and neuromorphic electronics.[16-23] However, the operating current in resistive switching devices are constrained by the scaling of the medium layer thickness that dictates the dimensions and morphology of the resulting filament.[2, 24-27] The thickness of the oxide based medium layer in existing low current memristive devices, typically deposited by atomic layer deposition method (ALD), is limited to above 3 nm[28-30] and the devices typically operate at the nano-ampere current level with the lowest record being 100 pA[31]. New materials and device designs are needed to form 2 ultrathin medium layers that is critical for reducing the filament length to the atomic scale, and for minimizing the energy required to form and rupture them. The recent emergence of two-dimensional materials has offered a new pathway for achieving a high quality atomically-thin medium layer to realize ultra-low power filamentary switching. In this work, we demonstrate the formation of a memristive switching medium layer with sub- nanometer thickness created through the oxidation of the atomically-thin two-dimensional (2D) hexagonal boron nitride. The reduction of the medium layer thickness allows the demonstration of sub-nanometer filamentary switching with sub-pA operation current and femtojoule per bit energy consumption. Moreover, fundamentally new atomic kinetics can arise when the medium layer thickness reduces to the atomic scale. The unique switching characteristics in this ultra- thin medium filamentary switching device is explained through Kinetic Monte Carlo simulations of the filament morphology and the distinct electrostatics in the atomically-thin medium. To create the atomically-thin switching medium, few layer hexagonal boron nitride (h-BN) flakes were first exfoliated onto the Si/SiO2 substrate with 285 nm SiO2 via a standard mechanical exfoliation method. The layer thicknesses were initially estimated by optical contrast and further confirmed by Atomic Force Microscopy (AFM). We conducted Raman spectroscopy on all the few-layer h-BN samples. A characteristic Raman peak near 1367 cm-1 corresponding to the E2g phonon vibration mode is observed in all the samples (Figure 1a). The few-layer samples were subsequently oxidized via an ultra-low power oxygen plasma treatment, which can oxidize the h-BN flakes up to a few nanometers in depth without damaging the flake 3 morphology (see Methods). AFM topographic images were obtained to verify that the flakes are atomically smooth after oxygen plasma treatment and have approximately the same thickness as the sample before the treatment. Raman measurements showed that the h-BN Raman peak completely vanished after the oxygen plasma treatment, which suggests the full amorphization of h-BN. Atomic resolution scanning transmission electron microscopy (STEM) reveals that pristine h-BN thin films have crystallized "honeycomb" lattice structure and hexagonal electron diffraction patterns, while oxidized h-BN flakes are amorphous (Figure 1b). Furthermore, electron energy loss spectroscopy (EELS) and elemental quantification identified that the oxidized few-layer hBN sample contains boron, nitrogen, and oxygen, among which the oxygen element dominates (Figure 1c). Therefore we use BNOx to denote the amorphized boron oxide material in the following text. Each BNOx flake was deterministically transferred from the substrate onto a few-layer graphene flake which serves as a thin, smooth and conformal bottom electrode (BE) of the memristor structure. 40 nm silver was subsequently deposited as the top electrode (TE), capped with another 40 nm gold to facilitate electrical probing (Figure 1a). Multiple devices were fabricated with BNOx layer thickness ranging from 0.9 nm to 2.3 nm, corresponding to bilayer to 5-layer h-BN before oxidation. The left panel of Figure 1d shows the AFM image of a 0.9 nm thick BNOx flake stacked on a 5-layer graphene BE and the corresponding AFM height profile. The right panel is a cross-section STEM image of the memristive device fabricated with this BNOx sample. We can clearly see a ~0.9 nm amorphous BNOx layer sandwiched between the graphene bottom electrode and the Ag top electrode. 4 Electrical measurements were conducted on devices with 0.9 nm, 1.3 nm, 1.8 nm, and 2.3 nm BNOx medium layer thickness under the current compliances of 5 pA, 90 pA, 500 pA, and 2 nA, respectively. The current compliance we applied for each device is the lowest current that can robustly form and rupture the filament while avoiding filament overgrowth. All the devices exhibit typical bipolar resistive switch behaviors in their I-V characteristics (Figure 2a). By applying a positive/negative voltage from BE to TE, a conductive filament of Ag atoms can be formed/ruptured, resulting in the switching to the low/high resistance state of the device. Once the positive voltage applied is larger than a critical voltage, i.e. the SET voltage, the Ag cations electrically connect the BE and TE, resulting in an abrupt jump of the current and therefore turning on the device. The migration of the Ag cations occurs at a relatively lower bias and dominates the conductance between the TE and BE. After a device is set, it will remain at the low resistance state (LRS) until a negative voltage is applied to reset the device to a high resistance state (HRS). The on/off ratio of our devices ranges from 100 to 1000, and this asymmetrical on-state current, which had enlarged the memory window and reduced the current level for rupturing the filament, is a signature of conductive bridge memristors.[8, 31] All these devices can reliably operate for at least hundreds of set/reset cycles in the ambient environment. The electrical characteristics of these memristive devices with atomically-thin switching layer demonstrated two distinctive features compared to the conventional thicker memristors. In previously reported memristive devices, the reset process is believed to form a depletion gap in the filament that electrically disconnects the TE and BE, while the set process re-connects the gap and therefore turns the device on. Several experiments have measured the filament depletion gap length to be ~6 nm[32-34], while a lower limit of 2-3 nm gap size was predicted via 5 theoretical analysis[35]. In conventional memristive devices, the filament, which is longer than the gap distance, is generated through a forming step requiring a significantly larger voltage than that used for setting the device. In contrast, the thickness of all our switching layers are smaller than the size of conventional filament gaps, and therefore no forming step is needed and the set voltage scales proportionally with the BNOx switching layer thickness. Another interesting feature of the BNOx devices is the strong dependence of its switching characteristics on the thickness of the BNOx layer. Devices with a thinner BNOx layer can be reliably operated under lower current and voltage in both the "set" and "reset" states. In particular, the set voltage increases linearly with the thickness of the BNOx layer (Figure 2b). Furthermore, when the voltage increases at the positive side, the current for the 0.9 and 1.3 nm samples show no notable increase until the abrupt jump occurs, while the current for the 1.8 nm and 2.3 nm samples increases gradually as the voltage increases until the voltage reaches the threshold to set the device, which is similar to conventional memristive devices. To understand the unique thickness dependent properties of the switching characteristics, we carried out kinetic Monte Carlo (KMC) simulations. The KMC simulations describes cationic generation, hopping and reduction behaviors at a microscopic level, which provides qualitative understanding of the electrical characteristics in these aggressively scaled memristive devices. Figure 3a shows the filament morphology for the different switching medium thicknesses, tox, in which the blue dots denote the energy minimum sites in the BNOx layer; the yellow dots denote the active electrode atoms or cations and the gray dots denote the inert electrode atoms. It shows that an extremely thin filament down to a single atomistic chain can be formed between 6 the electrodes for device with tOX ≈ 0.9 nm oxidized from bilayer hBN. In contrast, in the case of tOX ≈ 4.5 nm, because the filament grows in both the lateral and vertical directions before a percolative filament path can be formed, the filament expands much wider than a single atomistic chain in its lateral dimensions. Hence, the percolative filament path formed through a thicker oxide in the SET process would typically have a wider diameter. In addition, an enhanced positive feedback process facilitates the thin filament formation in the device for the case of tOX ≈ 0.9 nm. For a resistive switching device with BNOx thickness tOX and an applied voltage of V0, the electric field increases from 𝑉0 tOX to 𝑉0 tOX−𝑎0 if a cation (yellow dot) reaches the bottom electrode through hopping as shown in Figure 3b. Here 𝑎0 ≈ 0.45 nm is the grid spacing, which is treated as an approximate value for layer spacing in layered materials. For tOX ≈ 0.9 nm and V0 630 mV, the vertical electric field approximately increases from ~0.7 V/nm to ~1.4 V/nm after the first cation hops into the switching medium, which results in an exponential increase in the local hopping rate and hence facilitates the subsequent vertical filament formation. In comparison, for tOX ≈ 2.3 nm and V0 1.6V, the vertical electric field approximately increases from ~0.7 V/nm to ~0.86 V/nm after a cation hops into the switching medium during the initial filament formation stage. The change is much less significant compared to the vertical electrical field enhancement in the 0.9 nm switching layer case. The enhancement of the vertical growth rate, which has an exponential dependence on the vertical electric field, thereby, is much smaller. The larger increase in the vertical electric field in devices with ultrathin BNOx layer leads to more pronounced positive feedback process of vertical filament growth, which enhances the formation of a narrow atomistic filament chain. Furthermore, through KMC simulation, we found that the intrinsic set 7 time reduces as the BNOx layer thickness decreases (see Figure S1 in the supplementary information) since it is easier to form a percolative path in a thinner BNOX layer. In addition, the number of hopping cations required to electrically connect the BE and TE becomes considerably smaller when the switching medium thickness scales down since not only the filament becomes shorter, but the lateral dimensions of the filament also becomes smaller (Figure 3a). Hence, the atomically-thin memristive device is expected to have a faster intrinsic operation speed than the memristors with thicker switching medium. Figure 4 shows the characteristics of the thinnest memristive device we have demonstrated with 0.9 nm BNOx layer. Figure 4a shows the set and reset I-V characteristics of the device with 0.9 pA, 5 pA, and 9 pA current compliances. The high resistance state (HRS) current level fluctuates around 10-100 fA with no obvious gradual increase until the abrupt rise in the current occurs. The 10 -100 fA current level is dominated by the equipment noise. The device can be written and erased at sub-pA current level with a on/off current ratio ~10. The energy consumption for the programming (SET) and erasing (RESET) operations are estimated to be less than 10 fJ and 1 fJ, respectively, when we applied 10 ms voltage pulses to execute these operations. The device is expected to operate well with voltage pulses much shorter than 10 ms, which is the speed limit of our existing high sensitivity current measurement system. The set voltages under different current compliance are reliably near 0.6-0.7 V, which is consistent with the aforementioned Ag cation migration mechanism. Figure 4b shows the data retention of the memory device with a 0.9 nm BNOx switching layer, measured at 85°C. The device resistance at the low resistance state (LRS) was measured after the device was switched on under 5 pA 8 current compliance, and HRS resistance was measured after the device was switched off. The data was read at 0.18 V without applying current compliance, with a total measurement time of 4 hours. The device shows excellent data retention, suggesting that the Ag filament is robust even for the operation at low power and elevated temperature. Figure 4c shows HRS and LRS current levels obtained during 100 continuous set-reset DC cycles, with a reading voltage of 0.15 V and current compliance of 5 pA. The set voltages statistics for the 100 cycles exhibit a normal distribution centered at 0.63 V and a standard deviation of 0.088 V (Figure 4d), indicating that the filament formation process is highly reproducible and consistent, even operated at ultra-low power. The device has a record low operation current that is about two orders of magnitude lower than previously reported[8]. The device is attractive for application as an ultralow power memristive computational device, for example as a binary synapse for neuromorphic computing. In summary, the resistive switching of conductive filaments in a sub-nanometer thickness material medium is studied for the first time. The atomically-thin switching medium layer is formed through a unique process involving the oxidation of two-dimensional hexagonal boron nitride. It is observed that the confinement of the filament to atomic-scale thickness results in distinct atomic kinetics for the filament formation. The resulting memristor device can operate at sub-picoampere on-state current, less than 1 V operating voltage, and energy consumption per bit of less than 10 fJ for SET and less than 1 fJ for RESET. The device brings us one step closer towards realizing memristive electronic computation at the energy level of sub- 9 femtojoule per bit. It is promising for applications in many emerging applications that desire ultra-low power operation, such as binary synapse in neuromorphic computation systems. Experimental Section Few-layer hBN Flake Preparation and Oxidation: Few-layer h-BN flakes were exfoliated onto the Si/SiO2 substrate with 285 nm SiO2 layer. After THE exfoliation and layer number identification, samples with bilayer and trilayer hBN were treated with low power remote oxygen plasma (Evactron® Plasma Decontaminator™ ) to fully oxidize the sample. An RF power of 12 W with 200 mTorr O2 pressure and 20-30 seconds treatment time was applied. The samples were kept at least ~10 cm away from the plasma source to avoid damaging the surface morphology. Thicker h-BN flakes (>3 layers) were oxidized by Reactive Ion Etching (Oxford Instruments, Plasmalab 80 RIE system) under 200 mTorr O2 pressure, 10-15 W RF power, and 15-30 s treatment time. The oxidized sample was then deterministically transferred onto a multilayer graphene flake, which serves as the inert bottom electrode of the device. AFM and Raman spectroscopy: The AFM image was captured using a Bruker Dimension Icon System under an automatic peak force tapping mode. The Raman spectroscopy was measured with a Renishaw InVia spectrometer with a 532 nm incident laser. 50X objective and 100 uW laser power were used to avoid damaging the samples. STEM imaging and EELS measurements: Cross-sectional STEM samples were prepared using an FEI Helios660 dual beam FIB/SEM system. In situ FIB lift-out method using FEI Helios 660 FIB/SEM was used to prepare the cross-sectional STEM samples with a protective C layer. Low energy (2 keV) Ga+ ion beam was used for final milling. A Hitachi HD2700 and JEOL 10 ARM 200CF with an aberration corrector were used for high-angle annular dark-field (HAADF) scanning transmission electron microscopy (STEM) images were obtained with collection angles ranging from 68 to 280 mrad. EELS data were obtained with the JEOL ARM 200CF equipped with a cold field-emission gun. The energy dispersion and energy resolution were 0.25 eV/channel and better than 1 eV, respectively. Atomic ratios derived from the EELS data was performed by the quantification routine provided in Gatan Digital MicrographTM. Device Fabrication and Electrical Characterization: The device was patterned by a standard EBL technique and the metal electrodes were deposited by electron beam metal evaporation. We used 40 nm Ag as the top electrode, with another 40 nm Au on top. Ag was deposited onto the BNOx layer at a slow rate (~ 0.5 A/s) to minimize metal ion penetration. All the devices were measured by a Keysight B1500A parameter analyzer using a lakeshore probe station at ambient environment. Kinetic Monte Carlo Simulations: In two-dimensional (2D) KMC simulations, the hopping rate between the adjacent sites is described by a well-known thermodynamic relation, (1) where 𝐸𝑎 is the barrier height for a hopping mechanism, 𝑎0 is the grid spacing, kT is the thermal energy, 𝜀 is the electric field, 𝛾0 = 1012𝑠−1 is the hoping rate constant in the order of 𝑘𝑇/ℎ where ℎ is the Planck constant. 𝛽 is a unitless parameter that defines the efficiency at which the electric field assists in lowering the hopping/oxidation barrier height. A rectangular grid is used in the simulation by assuming the hopping sites are abundant for simplicity. The oxidation process at the top active electrode (AE) is described by the same thermodynamic relation in Eq. 1, in which 𝐸𝑎 is the oxidation barrier height and the electric field 𝜀 is taken 11 00ahEaexpkT at the AE surface. The filament formation is assumed to be limited by the oxidation and hopping processes, rather than reduction at the bottom inactive electrode. The potential distribution is calculated from a resistive network,[36] in which the resistance value between two metallic adjacent sites is low, and it is high otherwise. The concise model above has made a simplified assumption on the spatial distribution of hopping sites and has uncertainty of the hopping/ionization barrier height and rate constant parameters, but it is sufficient to study qualitative scaling trends as a function of the oxide thickness, which is insensitive to the exact values of the parameters. Acknowledgement This work was supported by Army Research Office (Grant no. W911NF-16-1-0435), Air Force Office of Scientific Research FATE MURI program (Grant no. FA9550-15-1-0514) and National Science Foundation (Grant no. CCF-1618038 and CCF-1618762). The work at Brookhaven National Laboratory was supported by the US DOE Basic Energy Sciences, Materials Sciences and Engineering Division, as well as the Center for Functional Nanomaterials which is U.S. DOE Office of Science Facilities, operated at Brookhaven National Laboratory under Contract No. DE-SC0012704. This research was supported in part by the Department of Energy (DOE) Award No. DE-FG02-07ER46376 (L.S). References [1] H. Akinaga, H. Shima, Proc. IEEE 2010, 98, 2237. [2] H.-S. P. Wong, H.-Y. Lee, S. Yu, Y.-S. Chen, Y. Wu, P.-S. Chen, B. Lee, F. T. Chen, M.-J. 12 Tsai, Proc. IEEE 2012, 100, 1951. [3] F. Pan, S. Gao, C. Chen, C. Song, F. Zeng, Mater. Sci. Eng. R 2014, 83, 1. [4] D. B. Strukov, G. S. Snider, D. R. Stewart, R. S. Williams, Nature 2008, 453, 80. [5] C.-L. Tsai, F. Xiong, E. Pop, M. Shim, Acs Nano 2013, 7, 5360. [6] T. Lee, Y. Chen, MRS Bull. 2012, 37, 144. [7] J. J. Yang, M. D. Pickett, X. Li, A. A. OhlbergDouglas, D. R. Stewart, R. S. Williams, Nat. Nanotechnol. 2008, 3, 429. [8] M. Kund, G. Beitel, C. U. Pinnow, T. Rohr, J. Schumann, R. Symanczyk, K. Ufert, G. Muller, IEDM Tech. Dig. 2005, 754. [9] R. Waser, M. Aono, Nat. Mater. 2007, 6, 833. [10] R. Waser, R. Dittmann, G. Staikov, K. Szot, Adv. Mater. 2009, 21, 2632. [11] L. Vandelli, A. Padovani, L. Larcher, G. Broglia, G. Ori, M. Montorsi, G. Bersuker, P. Pavan, IEEE Int. Electron Devices Meet. 2011, 421. [12] U. Celano, L. Goux, A. Belmonte, K. Opsomer, A. Franquet, A. Schulze, C. Detavernier, O. Richard, H. Bender, M. Jurczak, Nano Lett. 2014, 14, 2401. [13] H. Sun, Q. Liu, C. Li, S. Long, H. Lv, C. Bi, Z. Huo, L. Li, M. Liu, Adv. Funct. Mater. 2014, 24, 5679. [14] Y. Huang, R. Huang, Y. Pan, L. Zhang, Y. Cai, G. Yang, Y. Wang, IEEE Trans. Electron Devices 2012, 59, 2277. [15] S. Tanachutiwat, M. Liu, W. Wang, IEEE Trans. Very Large Scale Integr. VLSI Syst. 2011, 19, 2023. [16] S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder, W. Lu, Nano Lett. 2010, 10, 1297. 13 [17] E. Covi, S. Brivio, A. Serb, T. Prodromakis, M. Fanciulli, S. Spiga, Front. Neurosci. 2016, 10, 482. [18] G. Indiveri, B. Linares-Barranco, R. Legenstein, G. Deligeorgis, T. Prodromakis, Nanotechnology 2013, 24, 384010. [19] K.-H. Kim, S. Gaba, D. Wheeler, J. M. Cruz-Albrecht, T. Hussain, N. Srinivasa, W. Lu, Nano Lett. 2011, 12, 389. [20] M. Prezioso, F. Merrikh-Bayat, B. Hoskins, G. Adam, K. K. Likharev, D. B. Strukov, Nature 2015, 521, 61. [21] Z. Wang, S. Joshi, S. E. Savel'ev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan, Z. Li, Nat. Mater. 2017, 16, 101. [22] J. Woo, K. Moon, J. Song, S. Lee, M. Kwak, J. Park, H. Hwang, IEEE Electron Device Lett. 2016, 37, 994. [23] J A. M. Shen, C.-L. Chen, K. Kim, B. Cho, A. Tudor, Y. Chen, ACS Nano 2013, 7, 6117. [24] U. Russo, D. Ielmini, C. Cagli, A. L. Lacaita, IEEE Trans. Electron Devices 2009, 56, 186. [25] D.-H. Kwon, K. M. Kim, J. H. Jang, J. M. Jeon, M. H. Lee, G. H. Kim, X.-S. Li, G.-S. Park, B. Lee, S. Han, Nat. Nanotechnol. 2010, 5, 148. [26] J. J. Yang, F. Miao, M. D. Pickett, D. A. Ohlberg, D. R. Stewart, C. N. Lau, R. S. Williams, Nanotechnology 2009, 20, 215201. [27] T. Diokh, E. Le-Roux, S. Jeannot, M. Gros-Jean, P. Candelier, J. Nodin, V. Jousseaume, L. Perniola, H. Grampeix, T. Cabout, IEEE Int. Reliab. Phys. Symp. Proc. 2013, 3. [28] Y.-S. Chen, H.-Y. Lee, P.-S. Chen, T.-Y. Wu, C.-C. Wang, P.-J. Tzeng, F. Chen, M.-J. Tsai, C. Lien, IEEE Electron Device Lett. 2010, 31, 1473. 14 [29] Y. Chen, H. Lee, P. Chen, C. Tsai, P. Gu, T. Wu, K. Tsai, S. Sheu, W. Lin, C. Lin, IEEE Int. Electron Devices Meet. 2011, 717. [30] H. Lee, P. Chen, T. Wu, Y. Chen, C. Wang, P. Tzeng, C. Lin, F. Chen, C. Lien, M.-J. Tsai, IEEE Int. Electron Devices Meet. 2008, 297. [31] S. Gaba, F. Cai, J. Zhou, W. D. Lu, IEEE Electron Device Lett. 2014, 35, 1239. [32] Z Q. Liu, J. Sun, H. Lv, S. Long, K. Yin, N. Wan, Y. Li, L. Sun, M. Liu, Adv. Mater. 2012, 24, 1844. [33] Y. Yang, P. Gao, S. Gaba, T. Chang, X. Pan, W. Lu, Nat. Commun. 2012, 3, 732. [34] Y. Yang, P. Gao, L. Li, X. Pan, S. Tappertzhofen, S. Choi, R. Waser, I. Valov, W. D. Lu, Nat. Commun. 2014, 5, 4232. [35] S. Larentis, F. Nardi, S. Balatti, D. C. Gilmer, D. Ielmini, IEEE Trans. Electron Devices 2012, 59, 2468. [36] S. Pengxiao, L. Su, L. Ling, L. Ming, J. Semicond. 2014, 35, 104007. 15 Figure 1: Device structure and BNOx characterization. (a) Upper panel: The schematic of an atomically-thin memristive device. A 0.9 nm thick BNOx is sandwiched between the multi- layer graphene bottom electrode and the Ag top electrode. Bottom panel: Raman spectra of a bi-layer h-BN flake, before (red) and after (blue) oxygen plasma treatment. (b) High-resolution STEM images show the crystalline lattice of h-BN before the oxygen plasma treatment (left panel) and the sample becomes amorphous after the treatment (right panel). The scale bar is 2 nm. The insets of left panel is the electron diffraction pattern of the crystalline h-BN, where crystal planes and inter-planar spacing are indicated by Miller indices. (c) Electron energy loss spectroscopy (EELS) spectrum of a BNOx flake. The element atom ratio confirmed the abundance of oxygen in the sample after the oxygen plasma treatment. (d) The lower left panel is the AFM image of a BNOx stacked on a 5-layer graphene, of which the height profile is 16 shown at the upper left panel. The right panel is the cross-section STEM image of a device made from the sample in the left panel. The layered structure of 5-layer graphene and the amorphous morphology of BNOx can be clearly observed. Figure 2: Thickness dependent ultra-low power filamentary switching. (a) The set-reset I- V characteristics of devices with 0.9 nm, 1.3 nm, 1.8 nm, and 2.3 nm BNOx layer thicknesses measured under current compliances of 5 pA, 90 pA, 500 pA, and 2 nA, respectively. (b) The linear dependence of the device set voltage on the BNOx layer thickness. For each device, several (>5) switching cycles were measured to obtain a statistical distribution, indicated by the error bar. 17 Figure 3: Monte Carlo Simulation of the Filament Morphology. (a) KMC simulations of the filament formation process in mediums with 4.5 nm, 2.25 nm, and 0.9 nm oxide sites. The filament in the 4.5 nm medium is wider because it grows both vertically and laterally, while the simulation on 0.9 nm switching medium shows the possibility of forming a single atomic chain filament. The applied bias for the 4.5 nm and 0.9 nm structure is 3.15 V and 0.63 V, respectively. (b) Electrical potential profile during the filament growth in the 0.9 nm switching medium, as shown in the left panel. The applied voltage is 0.63V. The right panel demonstrates the local electrical field enhancement induced by a hoping of Ag cation, which is a situation depicted in the left panel. 18 Figure 4: Characteristics of a device with 0.9 nm BNOx. (a) Set-reset I-V characteristics of a device with 0.9 nm BNOx switching layer subject to different current compliances of 0.9 pA, 5 pA, and 9 pA. (b) Data retention measured at 85 °C. Both HRS and LRS resistances were measured with a 0.18 V read voltage. (c) The current read at 0.15 V over 100 continuous switching cycles. (d) The statistical distribution of the set voltage for 100 switching cycles and the Guassian fit curve. The set voltages exhibit a normal distribution centered at 0.63 V, with a standard deviation of 0.088 V. 19
1810.01056	1	1810	2018-10-02T04:02:15	CMOS-Integrated Diamond Nitrogen-Vacancy Quantum Sensor	[ "physics.app-ph", "quant-ph" ]	The nitrogen vacancy (NV) center in diamond has emerged as a leading solid-state quantum sensor for applications including magnetometry, electrometry, thermometry, and chemical sensing. However, an outstanding challenge for practical applications is that existing NV-based sensing techniques require bulky and discrete instruments for spin control and detection. Here, we address this challenge by integrating NV based quantum sensing with complementary metal-oxide-semiconductor (CMOS) technology. Through tailored CMOS-integrated microwave generation and photodetection, this work dramatically reduces the instrumentation footprint for quantum magnetometry and thermometry. This hybrid diamond-CMOS integration enables an ultra-compact and scalable platform for quantum sensing and quantum information processing.	physics.app-ph	physics	CMOS-Integrated Diamond Nitrogen-Vacancy Quantum Sensor Donggyu Kim,1, 2, ∗ Mohamed I. Ibrahim,3, 4, ∗ Christopher Foy,1, 4, ∗ Matthew E. Trusheim,1, 4 Ruonan Han,3, 4, † and Dirk R. Englund1, 3, 4, ‡ 1Research Laboratory of Electronics, Massachusetts Institute of Technology, 50 Vassar St, Cambridge, MA, 02139, USA 2Department of Mechanical Engineering, Massachusetts Institute of Technology, 77 Mass Ave, Cambridge, MA, 02139, USA 3Microsystems Technology Laboratories, Massachusetts Institute of Technology, 60 Vassar Street, Cambridge, MA, 02139, USA 4Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, 50 Vassar St, Cambridge, MA, 02139, USA The nitrogen vacancy (NV) center in diamond has emerged as a leading solid-state quantum sensor for applications including magnetometry, electrometry, thermometry, and chemical sensing. However, an outstanding challenge for practical applications is that existing NV-based sensing tech- niques require bulky and discrete instruments for spin control and detection. Here, we address this challenge by integrating NV based quantum sensing with complementary metal-oxide-semiconductor (CMOS) technology. Through tailored CMOS-integrated microwave generation and photodetection, this work dramatically reduces the instrumentation footprint for quantum magnetometry and ther- mometry. This hybrid diamond-CMOS integration enables an ultra-compact and scalable platform for quantum sensing and quantum information processing. In particular, Quantum metrology based on solid-state spins has shown outstanding sensing capabilities for various en- vironmental physical quantities. the NV center in diamond has emerged as a leading room-temperature quantum sensor for temperature1 -- 4, strain5 -- 7, electric fields8 -- 10, and magnetic fields11 -- 17 espe- cially to determine atomic species18 -- 25. The advances of NV-based quantum metrology are based on its long spin coherence time26 and its efficient optical interface for spin polarization and readout. Furthermore, picotesla mag- netic field sensitivity at DC under ambient conditions has been achieved27 by interrogating NV center ensembles. This hybrid NV-CMOS platform is a highly advanced, scalable and compact platform for quantum sensing and will serve as a foundation for a new class of quantum systems. Conventional approaches for NV magnetometry based on optically detected magnetic resonance (ODMR)28 in- volve discrete off-the-shelf instruments that limit practi- cal applications and scalability. NV ODMR requires (i) a microwave signal generator, amplifier, and delivery in- terface for NV spin manipulation, (ii) an optical filter to reject the pump laser, (iii) a photodetector for NV spin- dependent fluorescence measurement, and (iv) a pump laser. The use in conventional quantum sensing exper- iments of bulky instruments makes NV magnetometry hard outside of the lab or within mobile devices. Here, we realize a custom CMOS architecture that inte- grates requirements (i-iii) directly with a diamond sensor. This architecture29 stacks the microwave inductor, filter, and photodiode into a 200 µm × 200 µm footprint. We use this hybrid diamond-CMOS platform to demonstrate ambient quantum magnetometry and thermometry. ∗ These authors contributed equally † [email protected] ‡ [email protected] CHIP-SCALE QUANTUM SENSING Figure 1a illustrates the device for on-chip ODMR. A diamond slab is irradiated and annealed to produce NV centers at a density of ∼ 0.01 ppm. A 45◦ cut in the cor- ner of the diamond directs the off-chip green pump beam along the length of the diamond slab. This side-excitation reduces pump laser background into the photodetector lo- cated below the diamond. An on-chip microwave genera- tor and inductor drives the NV electron spin transitions. NV magnetometry detects external magnetic fields through the Zeeman shift induced on the NV's spin ground state sublevels28, shown in Fig. 2a. Specifically, an external magnetic field B induces an energy shift γeBz on the NV ground state spin triplet (ms = 0,±1(cid:105)), where Bz is the magnetic field component along the NV sym- metry axis. The spin transition frequencies, ν±, between sublevels 0(cid:105) and ±1(cid:105), are given by ν± = (Dgs − βT ∆T ) ± γeBz, (1) where Dgs = 2.87 GHz is the room-temperature natural ground-state splitting between sublevels 0(cid:105) and ±1(cid:105), γe is the electronic gyromagnetic ratio (28 GHz/T), βT = 74 kHz/K30, and ∆T is the temperature shift from room temperature. Measuring ν± gives Bz and ∆T in their difference and sum, respectively. In addition, measuring Bz for at least three of the four possible NV orientations in diamond (Inset in Fig. 1a) quantifies all components of B for vector magnetometry31 -- 34. The NV ground state transitions ν± are measured by ODMR under green laser excitation, as illustrated in Fig. 2a. The spin magnetic sublevel 0(cid:105) has a bright cy- cling transition, where it emits red fluorescence. In con- trast, the ±1(cid:105) can undergo an intersystem crossing into a metastable, dark spin-singlet state, from where it decays back into the 0(cid:105) sublevel. This has two consequences: op- tical spin polarization into sublevel 0(cid:105) and lower average fluorescence of the ±1(cid:105) spin populations. The microwave field moves spin population between 0(cid:105) and ±1(cid:105). Sweep- ing the applied microwave frequency leads to the ODMR 2 FIG. 1. CMOS-integrated quantum sensing architecture. a, A green pump laser excites an NV ensemble in the diamond slab. Microwave fields generated on-chip manipulate NV electron spins through an on-chip inductor, leading to ODMR. A metal/dielectric grating absorbs the green pump beam and transmits the NV spin-dependent fluorescence to the on-chip photodiode. Inset: NV atomic structure. Top-view micrograph of the fabricated CMOS chip without (b) and with (c) the diamond slab. Scale bar is 200 µm. spectra in Fig. 2b, from which ν± are determined. ON-CHIP MICROWAVE GENERATION AND DELIVERY In our chip-scale NV magnetometer, the ground-state spin transitions are driven by the on-chip generated mi- FIG. 2. NV energy level diagram and ODMR spectra. a, NV energy level diagram: green optical pump (green arrow) excites NV electrons from 3A2 to 3E. NV centers then emit red fluorescence by radiative decay (red arrow). Intersystem crossing rate (black dashed arrow) depends on NV spin states, resulting in spin-dependent fluorescence. b, Top: For B = 0, the ODMR spectrum shows one fluorescence dip for ν± = Dgs. Bottom: For an external magnetic field Bext, this resonance splits into two Zeeman-shifted spin transitions (black curve), whose difference (mean) gives Bz (∆T ). The gray curves show ODMR for the other three possible NV orientations. crowave fields. Figure 3a shows the circuitry for on-chip microwave generation and delivery. This circuitry is com- posed of a frequency synthesis loop, a current driver, and a resonant loop inductor. The frequency synthesis loop generates the microwave sweep signal from 2.6 GHz to 3.1 GHz required for the ODMR experiment. The main com- ponent of this loop is the on-chip CMOS ring voltage con- trolled oscillator (VCO). The ring VCO has a wide tuning range and avoids inductors35, which minimizes the cross- talk between the oscillator and the microwave inductor which drives the NV ensemble. The loop is closed with off-chip components to enhance the stability and decrease the phase noise of the signal. The microwave fields are delivered to the NV ensem- ble through the loop inductor (Fig. 3a) implemented on the top most metal layer (Metal 9). To efficiently deliver microwave field, the loop inductor and a pair of shunt capacitors form a resonating load for the current driver. The load resonates near Dgs. This current driver is fed by the output of the ring VCO. Therefore, the current flowing in the inductor is at the microwave frequency. To improve the performance of this inductor for advanced NV sensing protocols we wish to increase the emitted MW field amplitude36. The amplitude is enhanced by a factor Q of the driver DC bias current (I0(t) ∼ 5 mA), where Q (∼ 15) is the quality factor of the inductor. In addition, we use a three-turn loop to multiply the mi- crowave field strength. Overall, we have 25× enhanced microwave field strength compared to a non-resonant sin- gle turn loop (as plotted in Fig. 3b). These protocols also require highly uniform microwave fields over the excita- tion volume. To achieve this, three capacitive parasitic loops are inserted29. We tailor the radius of the these loops, so that their opposite induced field homogenize the overall generated field. Another degree of freedom is the capacitive gaps in the parasitic loops. This controls the amount of current flowing in these loops. Therefore, we 3 FIG. 3. On-chip CMOS microwave generation and inductor characteristics. a, Schematic of microwave generation circuitry. b, High-frequency electromagnetic fields simulations (HFFS) for on-chip inductor: Magnetic field amplitude is plotted as a function of distance from the inductor center (dashed line in a). The resonant multi-turn loop inductor (blue) produces 25× higher amplitude compared to the nonresonant single turn inductor (red) at the same DC current. The insertion of the parasitic capacitive loops yields a microwave uniformity to 95% over 50 µm. optimized these two parameters (i.e. the parasitic loop ra- dius and the capacitive gap) for the three parasitic loops to achieve > 95% uniformity. ON-CHIP SPIN READOUT The NV spin transitions are detected using an on- chip photodetector. A CMOS-compatible periodic metal- dielectric structure (Fig. 4a) in the Metal 8 interconnect layer filters green pump light. Specifically, incident light couples to the surfacepplasmon polariton (SPP) at the metal-dielectric interface, where green light rapidly de- cays due to frequency-dependent Ohmic loss37,38. The inset in Fig. 4a plots the intensity map for the green (λ = 532 nm, Top) and red light (λ = 700 nm, Bottom), showing the ∼ 95 % and ∼ 5 % absorption for green and red light through the structure, respectively. The photodiode consists of a P+/N-well/P-sub junc- tions (Inset in Fig. 4b), which is preferable for long wave- length detection39. Since we place the photodiode with its conductive layers below the inductor (Fig. 1a), large eddy currents near 2.87 GHz can be induced. This reduces the quality factor of the inductor, resulting in microwave am- plitude reduction. We reduce this eddy current by half, by dividing the photodiode area into four subareas as shown in Fig. 4b (See Methods for detailed analysis). Further- more, the anode/cathode connectors are arranged in a similar way to patterned ground shielding used in CMOS inductors40. This arrangement avoids any closed loops, which helps to cut the eddy current that may flow in the metallic connections. The photodiode has a measured re- sponsivity of 0.23 A/W at the wavelength of 532 nm and a noise-equivalent-power of ∼ 4.9 ± 2.6 nW/ Hz at 1.5 kHz. √ FIG. 4. Optical detection of NV spin-dependent flu- orescence. a, CMOS-compatible optical pump beam filter: the periodic metal-dielectric grating absorbs the green laser. Inset plots finite-difference-time-domain (FDTD) calculation of the optical intensity map inside the structure for green (top) and red light (bottom). Incident light polarization is perpen- dicular to the grating line. b, Photodiode geometry: photodi- ode area is divided into four subareas, separated by a shallow trench isolation, to reduce eddy current (yellow loops) losses. Inset: cross-section along dashed line. EXPERIMENTAL RESULTS We detect NV-ODMR with a lock-in technique. The green laser beam continuously excites the NV ensem- ble, and the frequency-modulated (FM) microwave fields (fm = 1.5 kHz and modulation depth of 6 MHz) drive the NV electron spin transition. The spin-dependent flu- orescence produces photo-current within the on-chip pho- todiode (Fig. 4b). Then, we read out the modulated 4 FIG. 5. On-chip detection of ODMR and NV-based quantum magnetometry. a, Frequency-modulated (FM) lock-in signal of NV spin-dependent fluorescence at zero external magnetic field (in addition to B ∼ 100 µT of the earth magnetic field). b, FM lock-in signal with a permanent magnet (B = 6.27mT): Bz is the magnetic field along the NV-axis with the spin transition at ν±. The linewidth of the ODMR is 7 MHz. Slopes dV/df at ν− = 2.8303 GHz and ν+ = 2.9330 GHz are 42.969 nV/MHz and 42.450 nV/MHz, respectively. c, On-chip magnetometry (Blue) and temperature effect (Red) separation: lock-in signals at both ν± are observed while switching the polarity of external electromagnet with a period of 26 min. Inset √ plots the histogram of measured magnetic field Bz with the standard deviation of 6.3 µT. This uncertainty corresponds to a Hz. Measurement is conducted with a time magnetic field sensitivity of 32.1 µT/ constant of 1 second. d, Noise spectral density monitored at ν−. Hz. The temperature sensitivity is 10.8 K/ √ photo-current at fm within an bandwidth of 0.078 Hz (a time constant of 1 second, considering the 24 dB/oct roll- off) with a lock-in amplifier (SR865A, Stanford Research System). and (cid:32) (cid:12)(cid:12)(cid:12)(cid:12)ν+ ∆V dV /df (cid:33) (cid:12)(cid:12)(cid:12)(cid:12)ν− − ∆V dV /df ∆Bz = 1 2γe . (3) Figure 5a shows the lock-in signal for the ODMR ex- periment under zero external magnetic field applied. This spectrum corresponds to the derivative of the ODMR spectrum shown in Fig. 2b. Next, we align a perma- nent magnet (6.27 mT) to split the spin transitions of the four NV orientations. Figure 5b shows the ODMR spectrum, which exhibits the expected eight spin transi- tions (Fig. 2b). In particular, we note the spin transitions at ν− = 2.8303 GHz and ν+ = 2.9330 GHz of the NV en- semble. Monitoring the lock-in signal V at ν− and ν+ allows in- dependent measurements of magnetic field and temper- ature, as described above. Specifically, the sum of the lock-in signal change ∆V at ν± is proportional to ∆T , while the difference provides ∆Bz: (cid:32) ∆T = 1 2βT (cid:12)(cid:12)(cid:12)(cid:12)ν+ ∆V dV /df (cid:12)(cid:12)(cid:12)(cid:12)ν− + ∆V dV /df (cid:33) (2) Figure 5c plots the detected ∆Bz induced by an electro- magnet (blue) and measured ∆T (red). The magnetic field sensitivity is given by the following relation: S = σBz√ ENBW . (4) Here, σBz is the noise in ∆Bz measurement, and ENBW is the equivalent noise bandwidth of the lock-in detec- tor. In our measurement, ENBW = 5/(64τ ) with a time constant τ of 1 second, accounting for the 24 dB/oct of the lock-in amplifier roll-off. By measuring σBz of 6.3 µT √ (Inset in Fig. 5c), we determine our DC magnetic field sensitivity of 32.1 µT/ Hz with a lock-in frequency of 1.5 kHz. Figure 5d plots the magnetic noise density measured at ν− (no temperature compensation) with the noise floor of ∼ 27 µT/ Hz. √ DISCUSSION The magnetic field sensitivity of our magnetometer is mainly limited by the green laser. Improving the on- chip optical filter performance by additional 60 dB (in- cluding nanophotonic structures41 in diamond or using multiple metal layers in CMOS) can improve the sen- sitivity by three orders of magnitude. In addition, re- cycling the green optical pump beam with a diamond wave-guide geometry27, and implementing dynamical de- coupling techniques28,36 would also lead to orders of mag- nitude sensitivity improvements. In conclusion, we demonstrate chip-scale quantum magnetometry by integrating diamonds with CMOS tech- nology. Throughout the CMOS multi-layers, essential components to detect NV ODMR - a microwave gener- ator, an inductor, an optical pump beam filter, and a photodetector - are fabricated. NV spin ensembles in- tegrated on the CMOS chip measure external magnetic fields with the sensitivity of 32.1 µT/ Hz. This com- pact spin-CMOS platform can be extended toward on- chip sensing of other quantities such as electric fields. We emphasize that the CMOS circuit in this work provides direct physical interactions with the NV quantum states beyond electronic I/O signaling42. √ In addition to chip-scale quantum sensing capability, our CMOS-based spin control and readout scheme will uniquely provide a scalable solution for implementing spin quantum-bit controls. This is in particular essential to de- velop a large-scale quantum information processor42 -- 45, which enables quantum enhanced sensing36,46,47 and quantum information processing48 -- 50. METHODS Diamond and CMOS Preparation To avoid the direct injection of the green laser pump beam on the CMOS, we cut the CVD-grown diamond sin- gle crystal (Element 6) as shown in Fig. 1a, which enables the optical pumping in parallel with the CMOS-diamond interface with total internal reflection. This diamond is irradiated by the electron beam with a dosage of 1018 at 1 MeV. Then, the diamond is annealed for 2 hours at 850◦ Celsius. Our CMOS chip is fabricated with 65 nm LP TSMC technology. To reduce the background red fluo- rescence from the CMOS passivation layer, we etched the layer with plasma reactive-ion. Measurement Setup A linearly polarized DPSS green laser beam (500 mW, λ = 532 nm, Verdi G2, Coherent) is delivered to the di- amond through a telescope of f1 = 35 mm and f2 = 150 mm. The beam diameter incident on the diamond is ∼ 500 micron. A half-wave plate rotates the polar- ization of the laser beam to maximize the laser absorp- 5 tion through the periodic metal/dielectric structure in the Metal 8 layer. A permanent magnetic was used in Fig. 5b to split the NV orientations. The square-wave mag- netic field applied in Fig. 5c was generated by an electro- magnet. Alternating electrical current was used to avoid magnetization. Eddy Current Analysis For a square photodiode with a side length of L, the eddy current power Peddy is quadratically proportional to the change of the magnetic flux dφ(t)/dt: (cid:1)2 ∝ L4(cid:0) dB dt L (cid:18) dB (cid:19)2 ∝ L3 Peddy ∝ (dφ(t)/dt)2 R . dt Here, t is the time; R is the resistance; and B is the mag- netic field generated by the loop inductor in Metal 9. By dividing the photodiode active area into N by N subar- eas, the eddy current is reduced by N 2 × (L/N )3/L3 = 1/N . ACKNOWLEDGEMENTS This research is supported in part by the Army Re- search Office Multidisciplinary University Research Ini- tiative (ARO MURI) biological transduction program. D.K. acknowledges financial support from the Kwanjeong Educational Foundation. M.I. acknowledges support from Singaporean-MIT Research Alliance (SMART), and MIT Center of Integrated Circuits and Systems. C.F. ac- knowledges support from Master Dynamic Limited. M.T. acknowledges support by an appointment to the Intelli- gence Community Postdoctoral Research Fellowship Pro- gram at MIT, administered by Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and the Office of the Director of National Intelligence. AUTHOR CONTRIBUTIONS D.E. and R.H. initially conceived the diamond-CMOS integration. M.I. conceived the idea of stacking the mi- crowave inductor, plasmonic filter, and photodiode in a 3D architecture. M.I., C.F., and D.K. contributed to chip specifications, design and the experiment. M.I. constructed the CMOS chip prototype. D.K. performed FDTD simulation for the optical filter design and the di- amond transfer on the CMOS chip. C.F. prepared the control software for the experiment. C.F. and D.K. con- structed the optical setup and etched CMOS passivation layers. All authors contributed to the discussion of the experimental results and writing the manuscript. COMPETING INTERESTS MATERIALS AND CORRESPONDENCE The chip-scale spin control and detection scheme in this work was filed as a United States Provisional Patent Application (62/623151). The correspondence should be addressed to Ruo- nan Han ([email protected]) or Dirk R. Englund (en- [email protected]) 6 [1] Georg Kucsko, PC Maurer, Norman Ying Yao, Michael Kubo, HJ Noh, PK Lo, Hongkun Park, and Mikhail D Lukin, "Nanometre-scale thermometry in a living cell," Nature 500, 54 (2013). [2] Philipp Neumann, Ingmar Jakobi, Florian Dolde, Chris- tian Burk, Rolf Reuter, Gerald Waldherr, Jan Honert, Thomas Wolf, Andreas Brunner, Jeong Hyun Shim, et al., "High-precision nanoscale temperature sensing using sin- gle defects in diamond," Nano letters 13, 2738 -- 2742 (2013). [3] Taras Plakhotnik, Marcus W Doherty, Jared H Cole, Robert Chapman, and Neil B Manson, "All-optical ther- mometry and thermal properties of the optically detected spin resonances of the nv -- center in nanodiamond," Nano letters 14, 4989 -- 4996 (2014). [4] Abdelghani Laraoui, Halley Aycock-Rizzo, Yang Gao, Xi Lu, Elisa Riedo, and Carlos A Meriles, "Imaging ther- mal conductivity with nanoscale resolution using a scan- ning spin probe," Nature communications 6, 8954 (2015). [5] Preeti Ovartchaiyapong, Kenneth W Lee, Bryan A My- ers, and Ania C Bleszynski Jayich, "Dynamic strain- mediated coupling of a single diamond spin to a mechan- ical resonator," Nature communications 5, 4429 (2014). [6] J Teissier, A Barfuss, P Appel, E Neu, and P Maletinsky, "Strain coupling of a nitrogen-vacancy center spin to a diamond mechanical oscillator," Physical review letters 113, 020503 (2014). [7] Matthew E Trusheim and Dirk Englund, "Wide-field strain imaging with preferentially aligned nitrogen- vacancy centers in polycrystalline diamond," New Jour- nal of Physics 18, 123023 (2016). [8] Florian Dolde, Helmut Fedder, Marcus W Doherty, Tobias Nobauer, Florian Rempp, Gopalakrishnan Bal- asubramanian, Thomas Wolf, Friedemann Reinhard, Lloyd CL Hollenberg, Fedor Jelezko, et al., "Electric-field sensing using single diamond spins," Nature Physics 7, 459 (2011). [9] Edward H Chen, Hannah A Clevenson, Kerry A John- son, Linh M Pham, Dirk R Englund, Philip R Hemmer, and Danielle A Braje, "High-sensitivity spin-based elec- trometry with an ensemble of nitrogen-vacancy centers in diamond," Physical Review A 95, 053417 (2017). [10] D. A. Broadway, N. Dontschuk, A. Tsai, S. E. Lillie, C. T. K. Lew, J. C. McCallum, B. C. Johnson, M. W. Do- herty, A. Stacey, L. C. L. Hollenberg, and J. P. Tetienne, "Spatial mapping of band bending in semiconductor de- vices using in situ quantum sensors," Nature Electronics 1, 502 -- 507 (2018). [11] JR Maze, PL Stanwix, JS Hodges, S Hong, JM Tay- lor, P Cappellaro, L Jiang, MV Gurudev Dutt, E Togan, AS Zibrov, et al., "Nanoscale magnetic sensing with an individual electronic spin in diamond," Nature 455, 644 (2008). [12] Gopalakrishnan Balasubramanian, IY Chan, Roman Kolesov, Mohannad Al-Hmoud, Julia Tisler, Chang Shin, Changdong Kim, Aleksander Wojcik, Philip R Hemmer, Anke Krueger, et al., "Nanoscale imaging magnetometry with diamond spins under ambient conditions," Nature 455, 648 (2008). [13] MS Grinolds, M Warner, Kristiaan De Greve, Yuliya Dovzhenko, L Thiel, Ronald Lee Walsworth, S Hong, P Maletinsky, and Amir Yacoby, "Subnanometre reso- lution in three-dimensional magnetic resonance imaging of individual dark spins," Nature nanotechnology 9, 279 (2014). [14] Kasper Jensen, Nathan Leefer, Andrey Jarmola, Yan- nick Dumeige, Victor M Acosta, Pauli Kehayias, Brian Patton, and Dmitry Budker, "Cavity-enhanced room- temperature magnetometry using absorption by nitrogen- vacancy centers in diamond," Physical review letters 112, 160802 (2014). [15] Thomas Wolf, Philipp Neumann, Kazuo Nakamura, Hi- toshi Sumiya, Takeshi Ohshima, Junichi Isoya, and Jorg Wrachtrup, "Subpicotesla diamond magnetometry," Physical Review X 5, 041001 (2015). [16] David R Glenn, Kyungheon Lee, Hongkun Park, Ralph Weissleder, Amir Yacoby, Mikhail D Lukin, Hakho Lee, Ronald L Walsworth, and Colin B Connolly, "Single-cell magnetic imaging using a quantum diamond microscope," Nature methods 12, 736 (2015). [17] Jens M Boss, KS Cujia, Jonathan Zopes, and Christian L Degen, "Quantum sensing with arbitrary frequency reso- lution," Science 356, 837 -- 840 (2017). [18] HJ Mamin, M Kim, MH Sherwood, CT Rettner, K Ohno, DD Awschalom, and D Rugar, "Nanoscale nuclear mag- netic resonance with a nitrogen-vacancy spin sensor," Sci- ence 339, 557 -- 560 (2013). [19] Tobias Staudacher, Fazhan Shi, S Pezzagna, Jan Meijer, Jiangfeng Du, Carlos A Meriles, Friedemann Reinhard, and Joerg Wrachtrup, "Nuclear magnetic resonance spec- troscopy on a (5-nanometer) 3 sample volume," Science 339, 561 -- 563 (2013). [20] T Haberle, D Schmid-Lorch, F Reinhard, and J Wrachtrup, "Nanoscale nuclear magnetic imaging with chemical contrast," Nature nanotechnology 10, 125 (2015). [21] D Rugar, HJ Mamin, MH Sherwood, M Kim, CT Ret- tner, K Ohno, and DD Awschalom, "Proton magnetic resonance imaging using a nitrogen -- vacancy spin sensor," Nature nanotechnology 10, 120 (2015). [22] Nabeel Aslam, Matthias Pfender, Philipp Neumann, Rolf Reuter, Andrea Zappe, Felipe F´avaro de Oliveira, An- drej Denisenko, Hitoshi Sumiya, Shinobu Onoda, Junichi Isoya, et al., "Nanoscale nuclear magnetic resonance with chemical resolution," Science 357, 67 -- 71 (2017). [23] Igor Lovchinsky, AO Sushkov, E Urbach, NP de Leon, Soonwon Choi, Kristiaan De Greve, R Evans, R Gertner, E Bersin, C Muller, et al., "Nuclear magnetic resonance detection and spectroscopy of single proteins using quan- tum logic," Science 351, 836 -- 841 (2016). [24] Igor Lovchinsky, JD Sanchez-Yamagishi, EK Urbach, S Choi, S Fang, TI Andersen, K Watanabe, T Taniguchi, A Bylinskii, E Kaxiras, et al., "Magnetic resonance spec- troscopy of an atomically thin material using a single-spin qubit," Science , eaal2538 (2017). [25] David R Glenn, Dominik B Bucher, Junghyun Lee, Mikhail D Lukin, Hongkun Park, and Ronald L Walsworth, "High-resolution magnetic resonance spec- troscopy using a solid-state spin sensor," Nature 555, 351 (2018). [26] Gopalakrishnan Balasubramanian, Philipp Neumann, Daniel Twitchen, Matthew Markham, Roman Kolesov, Norikazu Mizuochi, Junichi Isoya, Jocelyn Achard, Jo- hannes Beck, Julia Tissler, et al., "Ultralong spin coher- ence time in isotopically engineered diamond," Nature materials 8, 383 (2009). [27] Hannah Clevenson, Matthew E Trusheim, Carson Teale, Tim Schroder, Danielle Braje, and Dirk En- glund, "Broadband magnetometry and temperature sens- ing with a light-trapping diamond waveguide," Nature Physics 11, 393 (2015). [28] JM Taylor, P Cappellaro, L Childress, L Jiang, D Budker, PR Hemmer, A Yacoby, R Walsworth, and MD Lukin, "High-sensitivity diamond magnetometer with nanoscale resolution," Nature Physics 4, 810 (2008). [29] Mohamed I. Ibrahim, Christopher Foy, Donggyu Kim, Dirk R. Englund, and Ruonan Han, "Room-temperature quantum sensing in cmos: On-chip detection of electronic spin states in diamond color centers for magnetometry," IEEE VLSI Circuits Symposium (2018). [30] VM Acosta, E Bauch, MP Ledbetter, A Waxman, L-S Bouchard, and D Budker, "Temperature dependence of the nitrogen-vacancy magnetic resonance in diamond," Physical review letters 104, 070801 (2010). GD [31] BJ Maertz, AP Wijnheijmer, Fuchs, ME Nowakowski, and DD Awschalom, "Vector magnetic field microscopy using nitrogen vacancy centers in diamond," Applied Physics Letters 96, 092504 (2010). [32] Pengfei Wang, Zhenheng Yuan, Pu Huang, Xing Rong, Mengqi Wang, Xiangkun Xu, Changkui Duan, Chenyong Ju, Fazhan Shi, and Jiangfeng Du, "High-resolution vec- tor microwave magnetometry based on solid-state spins in diamond," Nature communications 6, 6631 (2015). [33] Hannah Clevenson, Linh M Pham, Carson Teale, Kerry Johnson, Dirk Englund, and Danielle Braje, "Robust high-dynamic-range vector magnetometry with nitrogen- vacancy centers in diamond," Applied Physics Letters 112, 252406 (2018). [34] Jennifer M Schloss, John F Barry, Matthew J Turner, and Ronald L Walsworth, "Simultaneous broadband vec- tor magnetometry using solid-state spins," arXiv preprint arXiv:1803.03718 (2018). [35] Behzad Razavi, RF microelectronics, Vol. 2 (Prentice Hall New Jersey, 1998). [36] Christian L Degen, F Reinhard, and P Cappellaro, "Quantum sensing," Reviews of modern physics 89, 035002 (2017). [37] Anatoly V Zayats, Igor I Smolyaninov, and Alexei A Maradudin, "Nano-optics of surface plasmon polaritons," Physics reports 408, 131 -- 314 (2005). [38] Lingyu Hong, Hao Li, Haw Yang, and Kaushik Sengupta, "Fully integrated fluorescence biosensors on-chip employ- ing multi-functional nanoplasmonic optical structures in 7 cmos," IEEE Journal of Solid-State Circuits 52, 2388 -- 2406 (2017). [39] Kartikeya Murari, Ralph Etienne-Cummings, Nitish Thakor, and Gert Cauwenberghs, "Which photodiode to use: a comparison of cmos-compatible structures," IEEE sensors journal 9, 752 -- 760 (2009). [40] C Patrick Yue and S Simon Wong, "On-chip spiral in- ductors with patterned ground shields for si-based rf ics," IEEE Journal of solid-state circuits 33, 743 -- 752 (1998). [41] Song Peng and G Michael Morris, "Resonant scattering from two-dimensional gratings," JOSA A 13, 993 -- 1005 (1996). [42] Edoardo Charbon, Fabio Sebastiano, Masoud Babaie, Andrei Vladimirescu, Mina Shahmoham- madi, Robert Bogdan Staszewski, Harald AR Homulle, Bishnu Patra, Jeroen PG Van Dijk, Rosario M Incandela, et al., "15.5 cryo-cmos circuits and systems for scalable quantum computing," in Solid-State Circuits Conference (ISSCC), 2017 IEEE International (Ieee, 2017) pp. 264 -- 265. [43] Norman Y Yao, Liang Jiang, Alexey V Gorshkov, Peter C Maurer, Geza Giedke, J Ignacio Cirac, and Mikhail D Lukin, "Scalable architecture for a room temperature solid-state quantum information processor," Nature com- munications 3, 800 (2012). [44] M Veldhorst, HGJ Eenink, CH Yang, and AS Dzurak, "Silicon cmos architecture for a spin-based quantum com- puter," Nature communications 8, 1766 (2017). [45] Bishnu Patra, Rosario M Incandela, Jeroen PG Van Dijk, Harald AR Homulle, Lin Song, Mina Shahmohammadi, Robert Bogdan Staszewski, Andrei Vladimirescu, Ma- soud Babaie, Fabio Sebastiano, et al., "Cryo-cmos circuits and systems for quantum computing applications," IEEE Journal of Solid-State Circuits (2018). [46] Vittorio Giovannetti, Seth Lloyd, and Lorenzo Maccone, "Advances in quantum metrology," Nature photonics 5, 222 (2011). [47] Thomas Unden, Priya Balasubramanian, Daniel Louzon, Yuval Vinkler, Martin B Plenio, Matthew Markham, Daniel Twitchen, Alastair Stacey, Igor Lovchinsky, Alexander O Sushkov, et al., "Quantum metrology en- hanced by repetitive quantum error correction," Physical review letters 116, 230502 (2016). [48] Hannes Bernien, Bas Hensen, Wolfgang Pfaff, Ger- win Koolstra, MS Blok, Lucio Robledo, TH Taminiau, Matthew Markham, DJ Twitchen, Lilian Childress, et al., "Heralded entanglement between solid-state qubits sepa- rated by three metres," Nature 497, 86 (2013). [49] Wolfgang Pfaff, BJ Hensen, Hannes Bernien, Suzanne B van Dam, Machiel S Blok, Tim H Taminiau, Marijn J Tiggelman, Raymond N Schouten, Matthew Markham, Daniel J Twitchen, et al., "Unconditional quantum tele- portation between distant solid-state quantum bits," Sci- ence 345, 532 -- 535 (2014). [50] Peter C Humphreys, Norbert Kalb, Jaco PJ Morits, Ray- mond N Schouten, Raymond FL Vermeulen, Daniel J Twitchen, Matthew Markham, and Ronald Hanson, "De- terministic delivery of remote entanglement on a quantum network," Nature 558, 268 (2018).
1910.10326	2	1910	2019-10-25T02:47:22	Electrical manipulation of spin pumping signal through nonlocal thermal magnon transport	[ "physics.app-ph" ]	We study the magnon transport in the nonlocal configuration composed of two Pt strips on top of yttrium iron garnet, with and without the presence of RF microwave generated by an on-chip antenna. We find that the spin-Hall induced thermal magnon heating/cooling, the Oersted field as well as the Joule heating generated by the a.c. current in the Pt injector can significantly influence the spin-pumping signal measured by the Pt detector in the presence of RF microwave, forcing the spin-pumping voltage to show up in the first and second harmonic signals in the nonlocal magnon transport measurement. These results indicate that nonlocal magnon transport configuration can serve as a structure to electrically detect and manipulate the spin-pumping signal. Furthermore, certain caution is needed when studying the interplay between incoherent magnon and coherent magnon spin transport in the nonlocal transport configuration, since the change in microwave-induced spin-pumping voltage can overwhelm the incoherent magnon transport signals.	physics.app-ph	physics	Electrical manipulation of spin pumping signal through nonlocal thermal magnon transport Yabin Fan1, Justin T. Hou1, Joseph Finley1, Se Kwon Kim2, Yaroslav Tserkovnyak3 and Luqiao Liu1 1Microsystems Technology Laboratories, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA 2Department of Physics and Astronomy, University of Missouri, Columbia, Missouri 65211, USA 3Department of Physics and Astronomy, University of California, Los Angeles, California 90095, USA Abstract: We study the magnon transport in the nonlocal configuration composed of two Pt strips on top of yttrium iron garnet, with and without the presence of RF microwave generated by an on-chip antenna. We find that the spin-Hall induced thermal magnon heating/cooling, the Oersted field as well as the Joule heating generated by the a.c. current in the Pt injector can significantly influence the spin-pumping signal measured by the Pt detector in the presence of RF microwave, forcing the spin-pumping voltage to show up in the first and second harmonic signals in the nonlocal magnon transport measurement. These results indicate that nonlocal magnon transport configuration can serve as a structure to electrically detect and manipulate the spin-pumping signal. Furthermore, certain caution is needed when studying the interplay between incoherent magnon and coherent magnon spin transport in the nonlocal transport configuration, since the change in microwave-induced spin-pumping voltage can overwhelm the incoherent magnon transport signals. Electronic mail: [email protected] 1 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.Recently, the incoherent magnon spin current generated electrically or thermally in magnetic insulators 1-3,4,5, compensated ferrimagnets6,7 and antiferromagnetic insulators 8 have drawn great attention, due to their potential applications in miniaturized magnonic devices. The incoherent magnons can be generated through spin-Hall effect (SHE) in heavy metals 1,2,5, e.g., in the Pt/yttrium iron garnet (YIG) structure by passing a charge current in the Pt layer, which creates a gradient in the magnon chemical potential in YIG 9. They can also be generated through the process called spin-Seebeck effect (SSE) in the magnetic insulator 3,7 when a temperature gradient is present. These incoherent magnons have short wavelength at room temperature 10 since they have energy up to 𝑘(cid:2886)𝑇/ℎ ~ 6THz, where the exchange energy dominates. These short- wavelength, high-energy incoherent magnons could offer the avenue to miniaturize magnonic devices, as they have a diffusion length up to 10µm 1 and they are electrically controllable 11,12. Besides incoherent magnons, the coherent magnons have been well-studied in the past. The coherent magnons can be generated by RF microwave at the ferromagnetic resonance (FMR) state 13,14,15, and they have well-defined frequency and long wavelength, as they can propagate on the centimeter scale. Therefore, they have been anticipated to have rich applications in magnonic logic devices 14,15,16. The microwave generated coherent magnons usually fall into the GHz frequency region, where the magnetic dipole interaction dominates. Moreover, spin-orbit torque can also be employed to generate coherent magnons 17. Naturally, it becomes important to investigate the interplay between coherent and incoherent magnons for their concurrent applications in magnonic devices, since they have different properties and they are generated and controlled by different means. However, despite large efforts, their interplay has still been under debate 15,18,19. One natural way to investigate the interplay between them would be to study the electrically/thermally induced incoherent magnon transport in the nonlocal transport 2 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.configuration in the presence of RF microwave 20. We will show in the following that in the presence of RF microwave, the nonlocal magnon transport configuration composed of two Pt strips on top of magnetic insulator can pick up the spin-pumping signals caused by the microwave, which could overwhelm the incoherent magnon transport signals in this configuration. Moreover, our work also indicates that the spin-pumping signal can be manipulated electrically by nonlocal thermal magnon transport, offering a remote control of coherent magnetic dynamics. To begin with, we have grown 60nm YIG films on top of the gadolinium gallium garnet (GGG) (110) substrate by magnetron sputtering 21, followed by rapid thermal annealing in oxygen atmosphere. To characterize the film quality, we have performed FMR measurement by placing the sample on top of a coplanar waveguide (CPW) to measure the RF microwave absorption while sweeping the in-plane magnetic field, as schematically shown in Fig. 1(a) inset. A typical FMR data at 6GHz microwave is plotted in Fig. 1(a). By fitting the FMR absorption spectrum with Lorentzian formula 22, we can obtain the resonance field 𝐻(cid:2928)(cid:2915)(cid:2929) and linewidth ∆𝐻 of the YIG film at the specific frequency. In Fig. 1(b), we show the obtained 𝐻(cid:2928)(cid:2915)(cid:2929) and ∆𝐻 under different frequencies. By fitting ∆𝐻 versus frequency with a linear curve 23, we get the damping factor 𝛼(cid:3404)1.8(cid:3400)10(cid:2879)(cid:2872). To further test the YIG film quality for incoherent magnon transport, we have fabricated Pt patterns on top by photolithography, as shown in Fig. 1(c) inset, to probe the nonlocal magnon transport. The pattern consists of two Pt bars (width:6µm, length:100µm, thickness:10nm) with square contact pads for wire bonding. The two Pt bars have a separation of 12µm (center to center). The idea is to use one of the two Pt bars as an injector and the other as a detector, to generate incoherent magnons by the SHE and the Joule heating in the Pt injector, and to measure 3 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.the incoherent magnon signal by 1st and 2nd harmonic signals in the Pt detector, as has been reported previously 1,2. We carried out the nonlocal magnon transport experiment by sending an a.c. current 𝐼(cid:2911)(cid:2913)(cid:3404)16mA (r.m.s.) at 15.5Hz in the injector, and meanwhile measuring the 1st and 2nd harmonic voltages in the Pt detector bar when rotating the external field 𝐻(cid:3404)300Oe in-plane. As shown in Fig. 1(c), while the 2nd harmonic voltage shows cos(cid:4666)𝜃(cid:4667) behavior, resulted from the inverse spin-Hall effect (ISHE) converting the thermally generated magnons to voltage signal, the 1st harmonic signal does not show explicit angle dependence, indicating that the SHE- induced magnons arriving at the detector is too weak to be detected, presumably due to the relatively high damping factor and large bar-to-bar separation on our film compared with previous reports 1,2. To probe the RF microwave influence on the non-local measurement, we fabricated an on-chip antenna beside the two Pt bars in order to apply RF microwave, as schematically shown in Fig. 2(a). In this setup, a constant 4GHz 25dBm microwave generated by a microwave synthesizer is sent into the antenna, and the nonlocal magnon transport measurement is simultaneously carried out using a lock-in amplifier. As shown in Fig. 2(b), when we apply a 5mA (r.m.s.) 15.5Hz a.c. current through the Pt injector, we detected some dips in the 1st harmonic nonlocal voltage V1w on the Pt detector at the YIG resonance field positions during sweeping magnetic field along x- direction. These dips scale linearly with the a.c. current applied through the injector in the low current regime (Iac≤5mA). To understand the mechanism of these dips, we carried out another experiment with a different measurement configuration. In this second experiment, we apply amplitude modulated (50% modulation) microwave onto the antenna, and at the same time measure the voltage at the Pt detector bar with a lock-in amplifier at the same modulation frequency 24 when sweeping the magnetic field. This is in nature a spin 4 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.pumping experiment configuration 25, where the RF spin pumping is induced by the on-chip microwave antenna. In the meanwhile, we can still apply d.c. current through the injector Pt bar of Fig. 2(a), which can provide additional influence onto the detected signal. As the current in the injector branch is at a different frequency from the detected signal (d.c. vs amplitude- modulation frequency), it does not directly contribute to the detected spin pumping voltage. As shown in Fig. 2(c), we can indeed measure the spin-pumping voltages Vsp at the resonance field when the d.c. current is 0mA and ±5mA in the injector bar. Compared with the case Idc=0mA, the spin-pumping voltage measured when Idc=±5mA shows a smaller value and the resonance field moves towards a higher field, presumably due to the Joule heating effect from the injector which alters the material or interface property at the Pt detector area. In order to compare the 1st harmonic nonlocal signal measured in Fig. 2(b) with these data, we did subtraction between the Vsp(Idc=+5mA) and the Vsp(Idc=-5mA) data and present the result in Fig. 2(d). Since the 1st harmonic nonlocal voltage V1w in Fig. 2(b) is proportional to the current (rather than current's absolute value), it should correlate with the voltage difference between Vsp(Idc=+5mA) and Vsp(Idc=-5mA). As shown in Fig. 2(d), the differential ΔVsp indeed shows some dip features around the resonance field, and in order to understand these features, we take a closer look at the spin-pumping signals in the following. In Fig. 3(a), we plot the zoom-in image of the spin-pumping voltage for Idc=±5mA of Fig. 2(c) in the positive resonance field regime. We find for both Idc=+5mA and Idc=-5mA, the spin-pumping peak splits into several peaks and the linewidth is greatly broadened, which could be attributed to the Joule heating effect that can reduce the YIG magnetization and introduce temperature inhomogeneity in the Pt detector area. More importantly, we notice that the Idc=-5mA spin- pumping curve shows a larger linewidth broadening in the resonance high-field edge than the 5 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.Idc=+5mA curve. This difference is most possibly due to the SHE in the Pt injector: when Idc=- 5mA, the SHE in the Pt injector can result in spin accumulation pointing along x-direction at the Pt/YIG interface, as shown in Fig. 3(b) lower panel, and these spins can generate incoherent magnons in YIG by interfacial spin scattering 1, which can be regarded as a magnon "heating" process in addition to the Joule heating effect 11,19; on the other hand, when Idc=+5mA, the SHE in the Pt injector can result in spin accumulation pointing along -x -direction at the Pt/YIG interface, as shown in Fig. 3(b) upper panel, and these spins will absorb incoherent magnons in YIG by interfacial spin scattering 1, which can be regarded as a magnon "cooling" process 11,19. The SHE induced heating/cooling leads to the difference in linewidth broadening in the Idc=±5mA spin-pumping curves. Interestingly, the linewidth broadening difference is more obvious at the resonance high-field edge than the resonance low-field edge in Fig. 3(a), which could arise partially due to the different Oersted field generated by Idc=±5mA in the injector 23. Significantly, we observe reversed linewidth broadening difference for the Idc=±5mA spin- pumping curves in the negative resonance field regime, as plotted in Fig. 3(c), which is expected since the SHE induced heating/cooling effect switches sign when the YIG magnetization is reversed (Fig. 3(d)). The difference in the Idc=±5mA spin-pumping curves, after subtraction, gives rise to the dip features in Fig. 2(d). By comparing results in Fig. 2(b) and Fig. 2(d), it seems the dips appearing at the resonance field in the 1st harmonic nonlocal voltage V1w could be due to the electrical manipulation of the spin-pumping voltage through SHE induced heating/cooling and the Oersted field effect. More quantitatively, the dip in Fig. 2(b) versus the a.c. current/RF power ratio is, (cid:4666)𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:3050)/𝐼(cid:2911)(cid:2913)(cid:4667)/𝑃(cid:2902)(cid:2890) (cid:3404)25.3 µV∙A(cid:2879)(cid:2869)∙W(cid:2879)(cid:2869), while the dip in ΔVsp (Fig. 2(d)) versus the d.c. current/modulated RF power ratio reads, (cid:4666)Δ𝑉(cid:2929)(cid:2926)(cid:2914)(cid:2919)(cid:2926)/𝛥𝐼(cid:2914)(cid:2913)(cid:4667)/𝛥𝑃(cid:2902)(cid:2890) (cid:3404)29.5 µV∙ 6 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.A(cid:2879)(cid:2869)∙W(cid:2879)(cid:2869), suggesting their consistency. Here, 𝛥𝑃(cid:2902)(cid:2890) is the r.m.s value of the modulated RF power during the spin-pumping experiment shown in Fig. 2(c). Now, we will analyze the 2nd harmonic nonlocal voltage V2w. To check the universality of the microwave influence, this time we apply a 3GHz 20dBm microwave through the antenna. As shown in Fig. 4(a), in order to clearly measure the heating-induced magnon current flow from the injector to the detector, we apply a relatively large current, Iac=16mA (r.m.s.), through the injector. Both 2nd harmonic signals, measured with the constant RF microwave and with no RF microwave applied in the antenna, are presented in Fig. 4(a). It can be clearly seen that, in addition to the SSE-induced 2nd harmonic voltage step at around Hx=0Oe 1, when the constant microwave is applied, a dip (peak) is developed at the positive (negative) FMR resonance field position. The dip (peak) value increases with the Iac current and gets saturated when Iac≥12mA. Again, we carry out the amplitude modulation spin-pumping experiment in the presence of different d.c. current applied through the Pt injector, as shown in Fig. 4(b). We can see that when the d.c. current is large, Idc=±14mA, the spin-pumping voltage signal Vsp is greatly suppressed and the linewidth is significantly broadened. As we have discussed, this could be due to the heating-induced modification of the YIG property and/or the Pt/YIG interfacial property (such as the spin-mixing conductance) in the Pt detector area. Further measurement of the Pt detector resistance change indicates the local temperature can be as high as 97°C when Idc=±14mA in the injector. Since the 2nd harmonic nonlocal voltage detected in Fig. 4(a) is proportional to I2, it is natural to correlate it with the Vsp(Imax)-Vsp(I=0mA) data, as this difference in the spin-pumping voltage could in principle show up in the 2nd harmonic nonlocal voltage V2w. So, in Fig. 4(c) we plot the subtraction between the Vsp(Idc=+14mA) and the Vsp(Idc=0mA) data obtained from Fig. 4(b). Clearly, the ΔVsp data in Fig. 4(c) shows a close correlation with the data in Fig. 4(a), 7 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.suggesting that the heating-induced modulation of the spin-pumping voltage signal could show up in the 2nd harmonic nonlocal voltage. More quantitatively, the ΔVsp peak versus the d.c. current square/modulated RF power ratio in Fig. 4(c) reads, (cid:4666)𝛥𝑉(cid:2929)(cid:2926)(cid:2926)(cid:2915)(cid:2911)(cid:2921)/𝛥𝐼(cid:2914)(cid:2913)(cid:2870)(cid:4667)/𝛥𝑃(cid:2902)(cid:2890) (cid:3404)9.2 mV∙ A(cid:2879)(cid:2870)∙W(cid:2879)(cid:2869), while the V2w peak in Fig. 4(a) versus the a.c. current square/RF power ratio is, 𝑉(cid:2926)(cid:2915)(cid:2911)(cid:2921)(cid:2870)(cid:3050)/(cid:4666)𝐼(cid:2911)(cid:2913)(cid:2870)/√2(cid:4667)/𝑃(cid:2902)(cid:2890) (cid:3404)7.5 mV∙A(cid:2879)(cid:2870)∙W(cid:2879)(cid:2869). The consistency in numbers obtained from the two experiments again confirms the proposed scenario, while the small discrepancy could be due to the difference in d.c. and a.c. heating efficiency. In summary, in the nonlocal magnon transport configuration, with the application of RF microwave, the spin-pumping signal generated on the Pt detector can be modulated by the spin- Hall induced magnon heating/cooling, the Oersted field as well as the Joule heating generated by the a.c. current in the Pt injector, forcing the spin-pumping voltage to show up in the 1st and 2nd harmonic nonlocal voltage signals. It will be valuable to carefully calibrate such effect in the experiment during studying the interplay between the coherent and incoherent magnon transport, since this electrical- or heating- modulated spin pumping voltage may overwhelm the nonlocal incoherent magnon transport signals detected by the Pt detector. This problem originates in the fact that the ISHE in Pt detector cannot differentiate the spin-pumping voltage and the incoherent magnon spin signal, when they are both being modulated by the a.c. current in the injector. Optical method, such as the Brillouin light scattering (BLS) 26, might be a powerful tool to study the interplay between coherent and incoherent magnon transport, since BLS can probe magnons operating at different frequencies individually, and thus could potentially differentiate coherent and incoherent magnon signals in experiment. 8 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.See the supplementary material for several supplemental experiments that were conducted. Supplementary material section 1 presents estimation of the heating effect in the system and its influence on the spin-pumping. Section 2 shows the subtraction between two spin-pumping experiments measured under the same condition to rule out possible artifacts during data subtraction. Section 3 illustrates the current dependence of the 1st and 2nd harmonic nonlocal voltage dips under constant RF microwave. This research was partially supported by National Science Foundation through the Massachusetts Institute of Technology Materials Research Science and Engineering Center DMR -- 1419807 and AFOSR under award FA9550-19-1-0048. References Steven S. L. Zhang and Shufeng Zhang, Physical Review B 86 (21), 214424 (2012). L. J. Cornelissen, J. Liu, R. A. Duine, J. Ben Youssef, and B. J. van Wees, Nature Physics 11, 1022 (2015). Sebastian T. B. Goennenwein, Richard Schlitz, Matthias Pernpeintner, Kathrin Ganzhorn, Matthias Althammer, Rudolf Gross, and Hans Huebl, Applied Physics Letters 107 (17), 172405 (2015). Ken-ichi Uchida, Hiroto Adachi, Takeru Ota, Hiroyasu Nakayama, Sadamichi Maekawa, and Eiji Saitoh, Applied Physics Letters 97 (17), 172505 (2010); J. Shan, P. Bougiatioti, L. Liang, G. Reiss, T. Kuschel, and B. J. van Wees, Appl. Phys. Lett. 110 (13), 132406 (2017). Steven S. L. Zhang and Shufeng Zhang, Physical Review Letters 109 (9), 096603 (2012); Junxue Li, Yadong Xu, Mohammed Aldosary, Chi Tang, Zhisheng Lin, Shufeng Zhang, Roger Lake, and Jing Shi, Nat Commun 7, 10858 (2016); H. Wu, C. H. Wan, X. Zhang, Z. H. Yuan, Q. T. Zhang, J. Y. Qin, H. X. Wei, X. F. Han, and S. Zhang, Physical Review B 93 (6), 060403 (2016). Kathrin Ganzhorn, Tobias Wimmer, Joseph Barker, Gerrit E. W. Bauer, Zhiyong Qiu, Eiji Saitoh, Nynke Vlietstra, Stephan Geprägs, Rudolf Gross, Hans Huebl, and Sebastian T.B. Goennenwein, arXiv:1705.02871 (2017). Joel Cramer, Er-Jia Guo, Stephan Geprägs, Andreas Kehlberger, Yurii P. Ivanov, Kathrin Ganzhorn, Francesco Della Coletta, Matthias Althammer, Hans Huebl, Rudolf Gross, Jürgen Kosel, Mathias Kläui, and Sebastian T. B. Goennenwein, Nano Lett 17 (6), 3334 (2017). R. Lebrun, A. Ross, S. A. Bender, A. Qaiumzadeh, L. Baldrati, J. Cramer, A. Brataas, R. A. Duine, and M. Kläui, Nature 561 (7722), 222 (2018). L. J. Cornelissen, K. J. H. Peters, G. E. W. Bauer, R. A. Duine, and B. J. van Wees, Physical Review B 94 (1), 014412 (2016). Joseph Barker and Gerrit E. W Bauer, Physical Review Letters 117 (21), 217201 (2016). L. J Cornelissen, J. Liu, B. J van Wees, and R. A Duine, Physical Review Letters 120 (9), 097702 (2018). 9 1 2 3 4 5 6 7 8 9 10 11 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.N. Thiery, A. Draveny, V. V. Naletov, L. Vila, J. P. Kathrin Ganzhorn, Stefan Klingler, Tobias Wimmer, Stephan Geprägs, Rudolf Gross, Hans Huebl, and Sebastian T. B. Goennenwein, Applied Physics Letters 109 (2), 022405 (2016). Charles Kittel, Phys Rev 73 (2), 155 (1948); A. A. Serga, A. V. Chumak, and B. Hillebrands, Journal of Physics D: Applied Physics 43 (26), 264002 (2010). Alexander Khitun, Mingqiang Bao, and Kang L. Wang, Journal of Physics D: Applied Physics 43 (26), 264005 (2010); Nature Physics 11, 453 (2015). V. V. Kruglyak, S. O. Demokritov, and D. Grundler, Journal of Physics D: Applied Physics 43 (26), 264001 (2010). Andrii V. Chumak, Alexander A. Serga, and Burkard Hillebrands, Nat Commun 5, 4700 (2014). M. Collet, X. de Milly, O. d'Allivy Kelly, V. V. Naletov, R. Bernard, P. Bortolotti, J. Ben Youssef, V. E. Demidov, S. O. Demokritov, J. L. Prieto, M. Muñoz, V. Cros, A. Anane, G. de Loubens, and O. Klein, Nat Commun 7, 10377 (2016). Benedetta Flebus, Pramey Upadhyaya, Rembert A. Duine, and Yaroslav Tserkovnyak, Physical Review B 94 (21), 214428 (2016); Attané, C. Beigné, G. de Loubens, M. Viret, N. Beaulieu, J. Ben Youssef, V. E. Demidov, S. O. Demokritov, A. N. Slavin, V. S. Tiberkevich, A. Anane, P. Bortolotti, V. Cros, and O. Klein, Physical Review B 97 (6), 060409 (2018). Chunhui Du, Toeno van der Sar, Tony X. Zhou, Pramey Upadhyaya, Francesco Casola, Huiliang Zhang, Mehmet C. Onbasli, Caroline A. Ross, Ronald L. Walsworth, Yaroslav Tserkovnyak, and Amir Yacoby, Science 357 (6347), 195 (2017). Jing Liu, Frank Feringa, Benedetta Flebus, Ludo J. Cornelissen, Johannes C. Leutenantsmeyer, Rembert A. Duine, and Bart J. van Wees, Physical Review B 99, 054420 (2019). Tao Liu, Houchen Chang, Vincent Vlaminck, Yiyan Sun, Michael Kabatek, Axel Hoffmann, Longjiang Deng, and Mingzhong Wu, Journal of Applied Physics 115 (17), 17A501 (2014); Letters 5, 1 (2014). Luqiao Liu, Takahiro Moriyama, D. C. Ralph, and R. A. Buhrman, Physical Review Letters 106 (3), 036601 (2011). Yiyan Sun, Young-Yeal Song, Houchen Chang, Michael Kabatek, Michael Jantz, William Schneider, Mingzhong Wu, Helmut Schultheiss, and Axel Hoffmann, Applied Physics Letters 101 (15), 152405 (2012). M. B. Jungfleisch, A. V. Chumak, A. Kehlberger, V. Lauer, D. H. Kim, M. C. Onbasli, C. A. Ross, M. Kläui, and B. Hillebrands, Physical Review B 91 (13), 134407 (2015); Lihui Bai, P. Hyde, Y. S. Gui, C. M. Hu, V. Vlaminck, J. E. Pearson, S. D. Bader, and A. Hoffmann, Physical Review Letters 111 (21), 217602 (2013). Yaroslav Tserkovnyak, Arne Brataas, and Gerrit E. W. Bauer, Physical Review Letters 88 (11), 117601 (2002); Y. Kajiwara, K. Harii, S. Takahashi, J. Ohe, K. Uchida, M. Mizuguchi, H. Umezawa, H. Kawai, K. Ando, K. Takanashi, S. Maekawa, and E. Saitoh, Nature 464 (7286), 262 (2010); C. W. Sandweg, Y. Kajiwara, A. V. Chumak, A. A. Serga, V. I. Vasyuchka, M. B. Jungfleisch, E. Saitoh, and B. Hillebrands, Physical Review Letters 106 (21), 216601 (2011); H. L. Wang, C. H. Du, Y. Pu, R. Adur, P. C. Hammel, and F. Y. Yang, Physical Review B 88 (10), 100406 (2013). Kyongmo An, Kevin S. Olsson, Annie Weathers, Sean Sullivan, Xi Chen, Xiang Li, Luke G. Marshall, Xin Ma, Nikita Klimovich, Jianshi Zhou, Li Shi, and Xiaoqin Li, Physical Review Letters 117 (10), 107202 (2016); Vijay Venu, Maxim Tsoi, Jianshi Zhou, Li Shi, and Xiaoqin Li, Physical Review B 96 (2), 024448 (2017); Kevin S. Olsson, Kyongmo An, and Xiaoqin Li, Journal of Physics D: Applied Physics 51 (13), 133001 (2018). H. Chang, P. Li, W. Zhang, T. Liu, A. Hoffmann, L. Deng, and M. Wu, IEEE Magnetics A. V. Chumak, V. I Vasyuchka, A. A Serga, and B. Hillebrands, Kevin S. Olsson, Kyongmo An, Xin Ma, Sean Sullivan, 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. Figure 1. Material property measured by ferromagnetic resonance and non-local transport measurements. the GGG(substrate)/YIG(60nm) sample at 6GHz when sweeping the in-plane magnetic field. (b) The resonance field 𝐻(cid:2928)(cid:2915)(cid:2929) and the linewidth ∆𝐻 of the GGG/YIG(60nm) film at different microwave frequencies. Blue line is linear fit of ∆𝐻 versus 𝑓. (c) The nonlocal (a) Ferromagnetic resonance measurement on first and second harmonic transport measurement in the two Pt bars on top of the GGG/YIG(60nm) film, when the in-plane field (H=300Oe) rotates in the film plane. The a.c. current applied in the Pt injector is 16mA (r.m.s.) at 15.5Hz. The two Pt bars are 6µm wide, 100µm long and 10nm thick. They have a separation of 12µm (center to center). Measurement is carried out by a lock-in technique. 11 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. thick Pt films on injector and detector are made of 10nm Figure 2. (a) Schematic of the experimental setup and the measurement scheme. The antenna, top of GGG(substrate)/YIG(60nm) film. The injector and detector are both 6µm wide and 100µm long. The separation between them is 12µm (center to center). (b) The first harmonic nonlocal voltage measured by the Pt detector in the presence of 5mA (r.m.s.) 15.5Hz a.c. current injected in the Pt injector bar, when a constant 4GHz 25dBm microwave signal is sent into the antenna. (c) The spin pumping voltages measured by the detector Pt bar when different d.c. currents flow in the injector bar. The spin- pumping is measured by a lock-in technique with the microwave output power (17dBm at 4GHz) modulated by the lock-in via amplitude modulation (50% modulation). (d) The change in the Vsp signal by subtraction between the Vsp(Idc=+5mA) and the Vsp(Idc=-5mA) data measured in (c). 12 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. Figure 3. Zoom-in image of the spin pumping voltages measured by the detector Pt bar when different d.c. currents (±5mA) flow in the injector bar for the positive (a) and negative (c) external field resonance regime. The spin-pumping is measured by a lock- in technique with the microwave output power (17dBm at 4GHz) modulated by the lock- in via amplitude modulation (50% modulation). (b) and (d) show the SHE induced magnon heating and cooling effect when different d.c. currents flow in the injector bar under positive (b) and negative (d) external magnetic field, respectively. 13 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. Figure 4. (a) The second harmonic nonlocal voltage measured by the Pt detector in the presence of 16mA (r.m.s.) 15.5Hz a.c. current injected in the Pt injector bar, when a constant 3GHz 20dBm microwave signal is sent into the antenna. The second harmonic voltage measured with no RF microwave applied is also presented as a reference. (b) The spin pumping voltages measured by the detector Pt bar when different d.c. currents flow in the injector bar. The spin-pumping is measured by a lock-in technique with the microwave output power (17dBm at 3GHz) modulated by the lock-in via amplitude modulation (50% modulation). (c) The change in the Vsp signal by subtraction between the Vsp(Idc=+14mA) and the Vsp(Idc=0mA) data measured in (b). 14 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.Supplementary Material for Electrical manipulation of spin pumping signal through nonlocal thermal magnon transport Yabin Fan1, Justin T. Hou1, Joseph Finley1, Se Kwon Kim2, Yaroslav Tserkovnyak3 and Luqiao Liu1 Institute of Technology, Cambridge, 1Microsystems Technology Laboratories, Massachusetts Massachusetts 02139, USA 2Department of Physics and Astronomy, University of Missouri, Columbia, Missouri 65211, USA 3Department of Physics and Astronomy, University of California, Los Angeles, California 90095, USA Estimation of the heating effect in the system and its influence on the spin-pumping In order to check the heating effect, we have measured the heating in the system using the following method. First, we measure the resistance of the Pt detector bar, RDetector, as a function of the d.c. current applied in the Pt injector, IInjector. As shown in Fig.S1(a), the RDetector increases dramatically with IInjector, indicating a significant Joule heating effect by the current in the Pt injector bar. Next, we measure the RDetector as a function of the substrate temperature by heating the sample with an external heater. Here, no electrical current is applied through the Pt injector. As shown in Fig.S1(b), we get almost a linear dependence of RDetector versus temperature T. By comparing (a) and (b), we can estimate the temperature of the Pt detector bar area as a function of the d.c. current in the Pt injector (without external heater), as summarized in Fig.S1(c). In Fig.S1(d), we present the spin-pumping experiment detected by the Pt detector bar when different d.c. currents (0mA, +10mA) are applied in the Pt injector. From Fig.S1(c), we could get that the Pt detector area temperature is around 58 °C when IInjector=+10mA. To crosscheck the heating effect on the spin-pumping spectrum, we also performed spin-pumping experiment at the (IInjector=0mA, T=58 °C) condition by heating up the substrate via external heater, and the result is also plotted in Fig.S1(d). Compared with the Vsp(Idc=0mA) spectrum (with no external heater), the Vsp(Idc=0mA, T=58 °C) spin-pumping peaks shift towards higher field values, comparable with the Vsp(Idc=+10mA) spectrum. We also observed that the Vsp(Idc=0mA, T=58 °C) spin-pumping peak value is smaller than the Vsp(Idc=0mA) spin-pumping peak, and the linewidth of Vsp(Idc=0mA, T=58 °C) is a bit larger than that of the Vsp(Idc=0mA) spectrum, presumably because the substrate heating increases the damping factor of the YIG film. Interestingly, the linewidth broadening for the Vsp(Idc=+10mA) spin- pumping peak seems to be larger than the Vsp(Idc=0mA, T=58 °C) peak, which is most likely due to the inhomogeneous Joule heating of the Pt detector area by the d.c. current in the Pt injector for the Vsp(Idc=+10mA) case. In the Vsp(Idc=0mA, T=58 °C) experiment, the substrate has been uniformly heated up. 1 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. Figure S1. (a) The resistance of the Pt detector bar as a function of the current in the Pt injector bar. (b) The resistance of the Pt detector bar as a function of substrate temperature. (c) The extracted Pt detector bar area temperature versus the current in the Pt injector bar. (d) Spin- pumping voltage detected by the Pt detector bar when different d.c. currents (0mA, +10mA) are applied in the Pt injector. The spin-pumping signal detected with heated substrate (58 °C) by external heater is also shown as a comparison. Double-check of the spin-pumping experiment on the Pt detector in the presence of d.c. current in the Pt injector to rule out possible artifacts We have carried out the spin-pumping experiment twice on the Pt detector when 𝐼(cid:2914)(cid:2913)(cid:3404)(cid:3397)5mA is applied on the Pt injector, and the data subtraction is presented in the following figure, Fig.S2, to rule out possible artifacts that can be caused during data subtraction. In Fig.S2(a), we can see that the Vsp(+5mA, Test1) and Vsp(+5mA, Test2) data overlap with each other quite well. After subtraction, as shown in Fig.S2(b), we do not observe any peak or dip signals. So, the peaks and dips we have observed in Fig.2(d) and Fig.4(c) in the main text are real signals, and they are due to the modification of the spin-pumping spectrum by different currents in the Pt injector. 2 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link. Figure S2. (a) Spin-pumping voltage measured on the Pt detector in the presence of d.c. current (+5mA) in the Pt injector at two different times, Test1 and Test2. (b) The subtraction between Vsp (+5mA, Test1) and Vsp (+5mA, Test2). the maximum value when Iac=5mA in the injector. Further increase in the current would lead to decrease Current dependence of the first and second harmonic nonlocal voltage dips We use different d.c. current in Fig.4 and Fig.2 in the main text is because the 1st harmonic nonlocal voltage dip V1w measured in Fig. 2(b) has different current dependence compared with the 2nd harmonic nonlocal voltage dip V2w measured in Fig 4(a), at the resonance magnetic field position. In the following, we summarize the 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) and 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2870)(cid:2933) dependence on the current in Fig. S3(a-b). The 1st harmonic nonlocal voltage dip 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) shows a linear dependence on the current value in the small current regime, but it reaches of the 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) value. The reason is because 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) is due to the spin-Hall effect (SHE)-induced shifting of the effect caused by the SHE. This is the reason why at Iac=5mA, the 1st harmonic nonlocal voltage dip 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) For the 2nd harmonic nonlocal voltage dip 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2870)(cid:2933) at the positive resonance magnetic field position, as shown Fig.S3(b), the 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2870)(cid:2933) value increases with the Iac current applied through the Pt injector, and it reaches order to optimize the 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2870)(cid:2933) signal. spin-pumping spectrum, and larger current will result in larger SHE. But the current can also cause broadening of the spin-pumping peak through Joule heating effect. When the current is larger than 5mA in the Pt injector, the significant linewidth broadening of the Vsp peak will gradually smear out the shifting has reached the maximum value. Since the SHE-induced shifting of the spin-pumping spectrum is quite small, we have used higher RF power to optimize the signal-to-noise ratio in Fig.2(b) in the main text. in Fig.4(a) in the main text, it is primarily due to the suppression of the spin-pumping peak by the Joule heating effect. In other words, larger current will lead to more obvious effect. Indeed, as plotted in saturation value when Iac is above 12mA. So, we used quite large current value in Fig.4 in the main text in 3 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.Figure S3. (a) The 1st harmonic nonlocal voltage dip 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2869)(cid:2933) measured by the Pt detector at the 4GHz 25dBm microwave is sent into the antenna. (b) The 2nd harmonic nonlocal voltage dip 𝑉(cid:2914)(cid:2919)(cid:2926)(cid:2870)(cid:2933) resonance magnetic field versus the a.c. current applied through the Pt injector, when a constant measured by the Pt detector at the positive resonance magnetic field versus the a.c. current applied through the Pt injector, when a constant 3GHz 20dBm microwave is sent into the antenna. 4 The following article has been accepted by Applied Physics Letters. After it is published, it will be found at Link.
1810.02873	1	1810	2018-10-05T20:13:55	Perspective: Organic electronic materials and devices for neuromorphic engineering	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Neuromorphic computing and engineering has been the focus of intense research efforts that have been intensified recently by the mutation of Information and Communication Technologies (ICT). In fact, new computing solutions and new hardware platforms are expected to emerge to answer to the new needs and challenges of our societies. In this revolution, lots of candidates technologies are explored and will require leveraging of the pro and cons. In this perspective paper belonging to the special issue on neuromorphic engineering of Journal of Applied Physics, we focus on the current achievements in the field of organic electronics and the potentialities and specificities of this research field. We highlight how unique material features available through organic materials can be used to engineer useful and promising bioinspired devices and circuits. We also discuss about the opportunities that organic electronic are offering for future research directions in the neuromorphic engineering field.	physics.app-ph	physics	Perspective: Organic electronic materials and devices for neuromorphic engineering Sébastien Pecqueur a, Dominique Vuillaume a, Fabien Alibart a,b,* a: Institut of Electronic, Microelectronic and Nanotechnology, Centre National de la Recherche Scientifique, University of Lille, Villeneuve d'Ascq, France b: Laboratoire Nanotechnologie et Nanosystèmes, Centre National de la Recherche Scientifique, Sherbrooke, QC, Canada *: corresponding author, [email protected]‐lille1.fr Abstract: Neuromorphic computing and engineering has been the focus of intense research efforts that have been intensified recently by the mutation of Information and Communication Technologies (ICT). In fact, new computing solutions and new hardware platforms are expected to emerge to answer to the new needs and challenges of our societies. In this revolution, lots of candidates' technologies are explored and will require leveraging of the pro and cons. In this perspective paper belonging to the special issue on neuromorphic engineering of Journal of Applied Physics, we focus on the current achievements in the field of organic electronics and the potentialities and specificities of this research field. We highlight how unique material features available through organic materials can be used to engineer useful and promising bio‐inspired devices and circuits. We also discuss about the opportunities that organic electronic are offering for field. future neuromorphic engineering the research directions in 1 limitations 1‐ Introduction While the CMOS technologies are reaching physical in terms of performances, new solutions are expected to emerge in the coming years to sustain the development of information and communication technologies (ICT). This tendency represents a real opportunity to shift and explore new computing approaches that would provide not only better performances, but more adapted solutions to deal with the new needs of our societies. These new requirements should provide solutions to numerous big challenges among which energy limitations constraints, management of the high amount of heterogeneous data generated, massive parallelism and heterogeneity of the communication network and of the interconnected electronic devices (i.e. IoT) that are not coped easily by conventional machines. One of the direction that could potentially answer some of these challenges would be to reproduce concepts and features observed in the brain: this extremely low power computing engine is perfectly adapted to deal with heterogeneous data such as sound, vision, or other sensory‐like signals that we record and analyze intensively in our everyday life. Furthermore, its plasticity and ability to adapt and learn make this object extremely resilient to changing environments and operating conditions that would be very appealing for the management of the complex network of devices constituting ICT of tomorrow. This approach should not only be limited to reverse‐engineering of the brain or bio‐ mimetism (which corresponds to reproducing with the highest precision some biological features observed in biological systems by engineering), nor to map in hardware machine learning algorithms that have proven their efficiency for image 2 classification or other machine learning applications. It could be extended to a broader bio‐inspired approach that targets the exploration of features or concepts of interest for computing purposes and finding some material implementation with emerging technologies. Neuromorphic engineering and computing, a term coined by Carver Mead in the 80s, represents the foundation for this direction.1 This concept was initially based on the analogy between ions flux across the cell's membrane and the CMOS transistor's transconductance, and has been used for the implementation of the first CMOS neuron. One key aspect of neuromorphic engineering is to deeply rely on the device physics as a computational primitive to build complex computing systems. While neuromorphic engineering has been mostly implemented with standard CMOS technologies (so far leading to bio‐inspired sensors and neural circuits),2,3 the field has recently benefited from the emergence of new materials and devices that have provided new opportunities for neuromorphic engineers. Notably, pushed by the drastic requirements in terms of memory density, resistive memory ‐ or memristive devices ‐ have been envisioned for implementing the synaptic weight connection between neurons.4 More advanced utilization, closer to biological behavior, capitalized on the memory device physics to realize very efficiently and locally learning rules observed in biology such as, for instance, Spike Timing Dependent Plasticity (STDP).5‐7 More recently, nanoscale devices have been also proposed to implement neuron's building block.8‐10 Note that neuromorphic engineering with emerging technologies is now going beyond single neuromorphic devices and first neuromorphic circuits based on emerging nanoscale devices have been successfully demonstrated.11,12 3 issued Most of the current approaches share the basic principle of using electronic processes obtained by various devices engineering routes to mimic biological processes. Nevertheless, biological systems and circuits from microelectronic are intrinsically very different and it is not clear whether standard electronic platforms will be ideal candidates for neuromorphic implementations. In fact, devices and circuits issued from microelectronic have been optimized for serial data transmission and sequential logic applications where ON/OFF switching ratio, speed and reliability are major constraints while neuromorphic platform relax these requirements (i.e. biological computing engines have intrinsically low signal to noise ratio, slow signal propagation and are fault/variation tolerant). As an interesting alternative to silicon‐based technologies, this perspective paper proposes to emphasize on organic materials and devices as potential candidates for neuromorphic engineering. First, these materials and devices are recognized to be very versatile engineering platforms: the soft engineering and bottom‐up routes used to synthetize and fabricate them can lead to a very large panel of electronic properties relevant for neuromorphic engineering. Secondly, organic materials and devices can gather both electronic and ionic species as mobile charge carriers. In such systems, electronic and ionic processes can be coupled via numerous basic physical interactions, from electrostatic charge polarization to redox charge transfer. These fundamental properties make organic iono‐electronic systems very exciting candidates for implementing bio‐inspired concepts and could offer a new toolbox for neuromorphic engineering. We will present here few of the basic material properties that have been used for neuromorphic purposes. These properties will be discussed in the light of selected examples that have been proposed recently for neuromorphic implementation. Finally, we will discuss the very under‐looked direction 4 of material and device integration at the circuit and system level offered by organic material processing and will hypothesize on the perspectives that they are offering. 2‐ Organic Materials and Devices in Neuromorphic Engineering The aim of neuromorphic engineering at the material level is to find media where information carriers are transported in a way that they may enable similar mechanisms ruling the information transport processes in biological media. From a very general viewpoint, various biological features not available in standard electronics can be highlighted. (i) Duality: In the synaptic cleft between two neurons, neurotransmitter molecules and ions are both information carriers: neurotransmitters are a family of chemical carriers confined in the synapses, while ions are charged species delocalized across the neural network via the electrolytic medium and cytoplasm. (ii) Time Dependency: Transmission of signals between cells through synapses is largely dynamical. For example, neurotransmitters are responsible for important time dependency of the transmitted signals charging/discharging of the pre‐synaptic vesicles and post‐synaptic receptors with different kinetically‐controlled physicochemical processes. (iii) Chemical diversity: The distribution in physicochemical properties of the different anions, cations and molecules makes biological computing' nodes a cross‐ point of different vectors of information, enabling selective processes necessary to interface their rich environment: for sensing exogenous information or transducing it to the body. In that scope, organic electronic materials have shown over the past ten years great promises in emulating these properties and continue nowadays to bio‐inspire us. Following the example of synaptic plasticity, we illustrate how organic materials and devices have been used for neuromorphic engineering and how more could be expected in this direction. 5 2.1. Small‐molecule and allotropic‐carbon organic semiconductors materials. One great advantage of organic semiconductors is the possibility to add new functionalities by integrating different materials via soft processes without destroying their electronic transport properties. We present in this section examples of such material hybridization that have led to interesting neuromorphic applications when various electronic mechanisms are advantageously coupled together. In 2010, Alibart et al. demonstrated how to use charge trapping/detrapping to design an organic Synapstor (synapse transistor) mimicking the dynamic plasticity of a biological synapse (figure 1).13 This device is also called NOMFET (Nanoparticle Organic Memory Field Effect Transistor) and combines in a single structure both a memory effect (by charge carrier trapping in nanoparticles) and a transistor effect (as the channel conductance is field‐effect modulated). This device (which is memristor‐like) mimics short‐term plasticity (STP),13 and STDP:7 two functions at the basis of learning processes (Fig. 1). A compact model was developed,14 and an associative memory, which can be trained to present a Pavlovian response, was demonstrated.15 Although the presence of the gold nanoparticles affects the crystallinity of the pentacene semiconductor, the optimized mobility for the functional biomimetic devices can reproductively reach 10‐ 1 cm²/V‐1·s‐1 (higher than the ionic mobility in water).16 An electrolyte‐gated version of this device was developed for biocompatible applications (EGOS: electrolyte‐gated organic synapstor).17 STP with a useful relative amplitude (30‐50% of the average DC current) was demonstrated at spike voltages as low as 50 mV, with a dynamic response in the range of tens of ms in aqueous saline solution and cell culture medium (leading to an energy of ca. pJ/spike). These EGOSs working at low voltages (e.g. 50 mV) have 6 performances that open the potentiality to directly interface real neurons. Human neuroblastoma stem cells successfully adhered, proliferated and differentiated into neurons on top of the EGOS as monitored by optical microscopy.18 Figure 1: (a) STP (adapted from [19]) and (b) SDTP (adapted from [20]) of a biological synapse and (c‐d) corresponding behavior for the NOMFET (adapted from [7,13]). Carbon allotropes were also investigated as organic materials for neuromorphic engineering: OG‐CNTFET (optically‐gated carbon‐nanotube field‐effect transistor) are interesting since they are programmable (optically and electrically) with multiple memory states.21‐23 OG‐CNTFET are made of carbon nanotube FETs (single CNT or random network of CNTs) covered by a photo‐conducting polymer (e.g. P3OT, poly(3‐ 7 octylthiophene‐2,5‐diyl)) making them light‐sensitive and conferring their non‐volatile memory behavior. These carbon‐nanotube‐based memory elements can be used as artificial synapses and eight OG‐CNTFET, combined with conventional electronic neurons, have been trained to perform Boolean logic functions using a supervised learning algorithm.24 In a crossbar architecture, the OG‐CNTFET allowed an efficient individual addressing (reduced crosstalk) by virtue of the development of a "gate protection protocol" exploiting the specific electro‐optical behavior of these devices.25 Thin films of redox molecules can also be used for neuromorphic devices. Y.P. Lin et al. have developed a non‐volatile nanoscale organic memristor based on electrografted redox complexes (iron(II) tris -- bipyridine complex with diazonium grafting functions) on metal electrodes.26 These devices are programmable with a wide range of accessible intermediate conductivity states. These authors experimentally demonstrated a simple neural network combining four pairs of organic memristors as synapses (and neurons made of conventional electronics) capable of learning functions. 2.2. Electrolyte‐Gated Semiconducting Polymers Polymers are well established materials in microelectronic fabrication since they can be chemically tuned for their micro/nano‐patterning on silicon: they are therefore good candidates for technological hetero‐integration on CMOS. Also as organic electronic materials, they offer more flexibility than small molecules to interface solvents for solution processing and liquid electrolyte‐gating. The possibility to integrate electrolytes (as ion carriers) on top of organic semiconductors (as hole/electron carriers) without damaging the one or the other offers the possibility to benefit of both charge carrier processes in an all‐organic neuromorphic device. 8 To the best of our knowledge, the first report of an organic device for neuromorphic system was proposed in 2005 by Erokhin et al.27 The device was based on an HCl p‐ doped polyaniline (PANI) conducting polymer interfaced with a LiClO4/poly(ethylene oxide) (PEO) solid electrolyte.28 In this device, the kinetically limited process is attributed to a Resistive/Capacitive (RC) internal circuit, correlated to the bulk conductivity of the doped/dedoped‐polymer and the accumulation of ions at the PANI/PEO interface.29,30 The doping of PANI with stronger acids and bulkier counter‐ anions promotes the repeatability of the device performances while controlling its RC properties.31 More recently, Demin et al.32 used these devices as the synapse layer of a simple perceptron, which has learned using an error‐correction‐based algorithm proposed by Rosenblatt in its seminal paper on perceptron,33 to implement the NAND or NOR logic functions as simple examples of linearly separable tasks. Undoped hole transporting polymers have also been tested such as poly(3‐hexylthiophene) P3HT, showing paired‐pulse facilitation (PPF),34 and shows that the formation of water channels in the solid electrolyte plays an important role in the functioning.35 2.3. Iono‐electronic polymers: mix electronic and ionic conduction in organic materials A more intimate coupling between ions and electrons can be obtained when their interaction is not limited at the interface between two materials but could be realize in the bulk. This option has been advantageously deployed to transduce ionic into electronic signals (and reciprocally) in a variety of organic electronic materials. Iono‐electronic polymer can be intimate blends of charged polymers, such as the (PEDOT:PSS). well‐known poly(3,4‐ethylenedioxythiophene):poly(styrenesulfonate) The semiconducting PEDOT low oxidation potential polymer which thermodynamically undergo in its oxidized state in moist air.36 In PEDOT:PSS, the PSS‐ 9 is a negative charges are compensated by a positive charge (i.e. a hole) on the PEDOT aromatic molecule to respect electro‐neutrality. The negative charges on PSS‐ are fixed while the positive charge on the PEDOT+ can easily be delocalized and contribute to electronic transport. In addition, if some external mobile ions can penetrate into the bulk of the PEDOT:PSS, the local electronic conductivity will be tuned based on the same electro‐neutrality principle (i.e. adding one monovalent positive ion balances one hole extracted from the PEDOT polymer chains) . The abilities of these materials and devices to gradually change their electrical conductivity upon ion/electron exchanges made them promising materials for brain‐like circuitry that is in nature an "iontronics" system. Li et al. used PEDOT:PSS in a rectifying memristor structure to demonstrate STP, long‐term plasticity (LTP), STDP and spike‐rate‐dependent plasticity (SRDP).37 It has also been used with liquid‐electrolyte gating, showing time‐dependent paired‐pulse depression (PPD).38 By substituting the PSS to poly(tetrahydrofuran) (PTHF), the memory becomes less volatile,39 and promotes LTP,40 with PPD about ten times slower than PEDOT:PSS systems.38 Based on PEDOT:PSS, another device structure named ENODe (electrochemical neuromorphic organic device) was recently proposed as a low‐ voltage organic synapse.41 The device structure features two PEDOT:PSS electrodes (one partly reduced by a poly(ethylenimine) treatment separated by an electrolyte. Upon application of pulse voltage on one of the PEDOT:PSS electrode (used as pre‐synaptic input), cation exchange through the electrolyte modifies the conductivity of the PEDOT:PSS/PEI film used as the post‐synaptic output. Working at a low voltage (around 1mV) and a low energy (10 pJ), this device showed non‐volatility, long‐term potentiation and depression with 500 discrete states within the operating range. 2.4. Perspectives 10 On the first aspect about the information carriers' duality, electrolyte‐interfacing organic semiconductor systems can mimic biological synapses with electrons/holes emulating the ions and ions emulating the neurotransmitters (Figure 2a). The slow dynamics associated to ion/electron interaction are well adapted to reproduce volatile memory effects. This represents an advantage over other standard electronic technologies, in particular RRAM or OxRAM systems that have been optimized for non‐ volatile memory applications.42 While memristive devices have been mostly considered for their analog programmability potential (i.e. implementing advantageously synaptic weight), such volatile mechanisms are only obtained in diffusive memristors43,44 with only little possibilities to be adjusted (i.e. unstability in nanoscale filaments). Advantageously, volatility level appears to be largely tunable in organic system. In addition, studies performed with solid electrolytes such as PEO shows that these platforms can be downscaled for high‐density development,45,46 while studies performed with water as an electrolyte shows their potential application to interface biology.47‐49 The time‐dependency of signal transmission and propagation observed in biology can also be advantageously reproduced with organic systems. For instance, ionic and electronic properties such as charge mobility in electrolyte or metal, respectively, are strongly different and result in very different performances for signal propagation. Consequently, implementing at the device level bio‐realistic time constant observed in biology (such as membrane time constant in neural cells, or neurotransmitter dynamics at the synaptic cleft) requires large capacitances in the purely electronic medium and are a severe limitation in terms of circuit design. In particular, the interplay between ionic transient currents and electronic steady‐states allows the tuning of the signal propagation dynamics,50 which influences all the plasticity‐related elementary 11 the cellular membrane.55‐57 Developing mechanisms of the artificial synapse: the interfacing of the electrolyte with the semiconductor is therefore a key parameter. Iono‐electronic polymers have the advantage to be hydrophilic compared to neutral polymers and allow their swelling in the presence of water,51 promoting the ion charge/discharge of the bulk of the material rather than the top surface.52 The bulk capacitance of these electrolyte‐interfacing materials is non‐ideal and relates to constant‐phase‐elements impedances of porous systems, which can be emulated with infinite numbers of series/parallel RC elements (Figure 2b).53,54 Analogously, these non‐ideal impedances are also ruling the diffusion‐ control existing at further organic semiconductor promoting the ion transport at nanoscale in the bulk of the electrical material is attractive to emulate the non‐ideality of the impedance ruling the whole time‐dependency in the electrochemical signal response. Other ionomers are of recent interest in organic electrochemical transistors such as conjugated polyelectrolytes (either self‐doped or intrinsic)58‐60. Also several works are currently promoting the conception of neutral semiconducting polymer with hydrophilic properties, by the introduction of glycol chains.61,62 Finally, at the chemical diversity level, organic electronics could offer a well‐ adapted platform to couple various functionalities at the device level. Coupling neuromorphic sensing, transduction and computing at the hardware level is still not extensively investigated, although biomimetic sensing/transduction platforms are in need for neuromorphic data analysis of pattern‐recognition based applications.63,64 Especially for input/output layers, sensing/transduction of exogenous information (light, mechanical deformation, chemical) to/from ionic action potentials requires specific materials for coupling the neural information transport properties to the desired sensed/transduced physical information. Organic semiconductor materials can 12 show fluorescence and phosphorescence,65 promoting a direct coupling between the material electrical properties and a specific wavelength range of emitted light (as organic light‐emitted diodes)66,67, or absorbed one (as organic photodiodes).68‐70 The photo‐transduction associated to the modification of the rectification ratio of photodiodes by light can modulate the material electrical conduction under reversed bias, mimicking retinal photoreceptor cells. Organic semiconductors are also mechanically stable71, and semiconducting iono‐electronic polymers systems can also be electroactive and transduce by electrical stimuli as low as 2 V to morphological changes.72 Reversibly, they can also sense mechanical deformations (as cutaneous nerves or cochlea's hair cells), modifying the electrical properties of electronic skins sensors.73,74 Finally, electrical properties of organic semiconductors can also be molecularly modulated, gating the electronic state of the channel from non‐conducting (inhibition) to conducting (excitation). Molecular‐gating of the organic semiconductor conductivity can be done either directly in the material by doping with molecularly‐ specific strong electron acceptors (p‐type) or donors (n‐type),75,76 or indirectly by performances modification of an external gate with specific electroactive agents contained in the electrolyte.77,78 The integration of such molecular agents in the material (either the semiconductor or the electrolyte), modulating the conductivity accordingly to their chemistry, provides a versatile mean to incorporate both performance variability and selectivity necessary to integrate multiple processes at the network level using a single electronic technology. 13 Figure 2: (a) Analogy between a biological synapse and organic‐electrochemical synapse. (b) De Levie infinite transmission line impedance model for porous electrode with redox (Gerischer Impedance) or without redox (Warburg Impedance) reaction. 3‐ Circuits and systems based on organic processes 3.1‐ General considerations and recent achievements One of the important challenges that neuromorphic engineering is facing is to bring elementary devices (or basic building blocks) at the circuit and system level. Some important aspects of the problem are: (i) To reproduce the high parallelism of neural network in hardware. For example, conventional memory structures are mostly accessed sequentially when one would deeply benefit from a parallel and distributed memory architecture to emulate synaptic operations between cells. Potential solutions to this challenge have been considered based on the idea of passive crossbar arrays with resistive memories. But, scaling of this approach seems limited due to physical limitation in terms of energy cost and engineering limitations when one tries to move charges over long and passive metallic wires79. (ii) To reproduce the combination of both local and global effects observed in biology. Some processes are highly local such 14 as synaptic plasticity processes (i.e. Spike Timing Dependent Plasticity, for example) and would benefit from embedded plastic features as close as possible to the memory point. Some others such as reinforcement learning (i.e. associated to processes such as dopamine delivery in the brain) or homeostasis are global effects affecting large populations of cells and would benefit from a more global circuitry such as a central processing unit. This later aspect is somehow in opposition with the idea of distributed computing units and locality. (iii) To reproduce the complex wiring between populations of cells in hardware. In fact, if the issue of fan‐in / fan‐out could be solved, one remaining question is to implement in hardware the ability of cells to transmit information on different length scale. For instance, if one neuron projects on 10000 neurons on average, the target population of this projection varies deeply from cell to cell and involves both "hard‐wiring" (i.e. neural topology that pre‐exist before learning) and "plastic‐wiring" (i.e. wiring between cells resulting from different learning experiences). Since this complex wiring is in principle not known before learning, the conventional top‐down technological approach is to oversized the interconnection between cells (i.e. allow for the highest degree of interconnection between cells and restrict it after learning) or to make arbitrary decision on a reasonable degree of interconnection and a pre‐defined population of cells that will be allowed to communicate between each other. The first aspect is somehow illustrated by the concept of passive crossbar arrays and the second by the most advanced neuromorphic implementations such as True North80 or Spinnaker81 where Address Event Representation (AER) is used to emulate parallel communication between populations. 3.2. Organic materials for dendritic engineering 15 Passive crossbar arrays of resistive memory cross‐point are an elegant solution to reproduce the high level of parallelism between cells4. It corresponds to implementing with metallic wires the axono‐dendritic tree of biological neural cells and to implement the synaptic weight with memory cross‐point. In this approach, one challenge is to minimize wires' resistance in order to not hide the memory elements' resistance itself.79 Nevertheless, this approach neglects some important aspects at work in biological networks such as dendritic computing.82 In fact, transmission speed along axons and dendrites, localization of the synaptic cleft along the dendritic arbor and their respective timing are used as important computing ingredients by biological cells. These features can hardly be implemented with purely electronic conductors or at the expense of heavy overhead circuitry to implement the bio‐realistic time constant for example. Few interesting propositions have been recently published in the field of organic electronics that could offer new perspectives for this particular feature. Xu et al. proposed to implement the interconnecting wires (figure 3a) with conducting organic materials.83 This approach is of interest for the ease of implementation of the conductive wires by inkjet printing but also lead to interesting dendritic properties. So far, the authors demonstrated PPF and STDP on the organic nanowire/synapse system. Note that the material system is an organic/electrolyte device in nature and that the combination of both ionic and electronic dynamics results in bio‐realistic temporal features at the device level. An interesting perspective would be to extend the analogy with dendritic processes at work in biological cells. Along the idea of dendritic computing features, other works have reported the possibility to use organic materials for implementing dendritic connections. Work from Malliaras et al. proposed to implement dendrites with PEDOT:PSS materials.84 This iono‐ electronic polymer was used to demonstrate orientation selectivity along a long PEDOT 16 line with multiple gates localized along the wire in an electrolytic environment (figure 3c). Interestingly, time lag effect was demonstrated to depend on the gate location and a direct analogy with temporal integration along the dendrites can be seen in this experiment. Temporal features were the result of both the ionic dynamics in the electrolyte and the electronic properties of the PEDOT:PSS material. A third material system based on Indium Zinc Oxide (IZO) transistor electrolitically gated with a chitosan membrane demonstrates dendritic features (figure 3b).85 In this work the chitosan layer is used as a proton conductor material and can be thought as equivalent to an artificial dendrite. The organic ionic conductor (chitosan) is then used to implement temporal features observed in biological dendrites with multiple gates implementing the pre‐neuron input. Finally, toward the idea of mimicking biology, Yang et al. reported on the realization of an ionic cable based on polyacrylamide hydrogel (figure 3d).86 This purely ionic system based on ionic conductors separated by an insulator was used to implement an ionic propagation line. This system is a direct analogy to the neuron's membrane and to the way spikes propagate along the axono‐dendritic tree. We believe that organic materials and their ubiquitous ionic / electronic properties represent a real opportunity to implement such dendritic processes and that more research efforts in this direction could lead to very appealing technological solutions for neuromorphic engineers. 17 a) c) b) d) Electrode + + + _ _ _ Input port + + + x _ _ _ + + + _ _ _ i(x,t) i(x,t) Electrical double layer v(x,t) Ionic conductor + + + + + + _ _ _ _ _ _ Ionic conductor Dielectric + + + _ _ _ + + + _ _ _ Output port Z Figure 3: (a) Organic dendrites and synapses realized by inkjet printing of organic core‐ sheath nanowire (from [83]). (b) Emulation of dendritic propagation of signals with IZO transistor and chitosan ionic membrane (adapted from [85]). (c) Orientation selectivity with dendritic‐like organic electrochemical transistors (adapted from [84]). (d) Concept of ionic cable (adapted from [86]). 3.3. From local to global processes: mix ionic/electronic conductors with global electrolyte As mention previously, learning in biology can result from different spatial scales. From purely local event induced by chemical processes happening at the synaptic cleft, to extended contribution such as the tri‐partite synapse that involve the contribution of the glial cells into synaptic potentiation (note that glial cells can extend along multiple neurons and provide subsequently an additional interconnection between these cells), 18 to reinforcement learning where the overall synaptic learning can be influenced by a global marker such as dopamine concentration. These mechanisms are possible thanks to the multiple carriers of information that are used in biological engines (various ions and chemicals) and that are hardly reproduce with single data carriers in standard electronics. Gkoupidenis et al. published an interesting solution toward this goal recently.87 The proposed system was composed of a passive matrix of OECTs interconnected by metallic wires implementing the "electronic" part of the system and of a global electrolyte used to gate the OECT devices implementing the "ionic" part of the system (figure 4a and 4b). Local synaptic effects can potentially be implemented on single OECT (see previous sections) while global effects are implemented via the electrolyte and the ionic carriers of information. Integrating a grid of 4x4 OECTs gated by the same electrolyte, these authors demonstrated that they can globally control the behavior of the OECTs in the array, a behavior that resembles homeoplasticity phenomena of the neural environment. They also show that OECTs can communicate though the common electrolyte, i.e. an input signal sent to one OECT is also detected by a neighboring one in the network. Along this line, we propose recently a concept of spatial reservoir with OECT sensors.88 In this system (figure 4c‐e), the connectivity through the electrolyte was used to transmit the input signal from a global gate to a network of 12 OECTs. Thanks to the electropolymerization of the p‐type iono‐electronic polymer, each individual sensing element presented a many‐fold variability in time response, charge/discharge‐ symmetry, channel conductance and transconductance, affecting the output image of the gate input signal. We demonstrated in this work that non‐trivial binary classification tasks can be realized out of this system: the system has been able to discriminate two different frequency‐modulated voltage pulses patterns injected from the gate, with error 19 rates controlled by the number of training vectors, the number of OECTs and their variabilities. Showing that the more the better in terms of number of OECTs and training vectors, this study also unravels that an appropriate set of a restricted number of OECTs can systematically lead to an optimal recognition or not, depending on the individual device properties (the relevant properties responsible for pattern recognition performance of the network have not been quantitatively identified yet). Considering the relationship between electrolyte nature and the gate field‐effect efficiency75,89, more is expected in a study where a fixed and periodical pattern would be injected from the gate and the electrolyte medium would be modified as an input: more complex interaction between the electrolyte and OECT such as sensitivity to various ionic species, sensitivity to concentration could be added into the overall system. 20 Figure 4: (a‐b) Global gating through the electrolyte of organic electrochemical transistors showing both local and global tenability of the devices (adapted from [87]). (c‐ d) Concept of spatial reservoir implemented with electropolymerized OECTs for pattern classification (adapted from [88]) 21 3.4. From synaptic learning to network wiring. As pointed out by J. Hawkins at IEDM 2015,90 synaptic weighting with analog weight such as resistive memory is in essence different from the concept of wiring observed in biology. The former corresponds to continuous evolution of the synaptic weight between two values (the min and max conductance of the memory element) while the later corresponds to a continuous evolution of an "undisplayed" synaptic state. In other words, the potential synapse that will result from the growth of a dendritic path does not contribute to computing until the synaptic connection is established. Nevertheless, important learning and computing events can be attributed to this dendritic growth. One could expect that artificially reproducing the dendritic wiring through fixe network such as parallel crossbar topologies will miss important aspect of biological computing, notably in terms of energy consumption since all weights in the network will contribute to power consumption (even if there conductance is small) while undisplayed synaptic element not. In addition, aging consideration should consider that if fix topology network can be used to compensate to device aging (i.e. synaptic connections that become inefficient are replace by new ones) to some extend (until enough alternative pathway are available), this approach does not solve completely the important ability of biology to regenerate new connections. 3.5. Perspectives for neuromorphic Organic electronic materials offer also biomimetic perspectives at the network level by their fabrication processes (top‐down and bottom‐up), which could emulate the morphological evolution of the neural network while preserving the functionality of the elementary organic electronic elements. 22 On the nucleation, self‐assembling of conducting polymers micro/nano‐object could rapidly access to the systematic patterning of high density networks of functional conducting polymer units. Prior assembling, conducting polymers have been can be formulated into nanoparticles and micro‐spheres of different dimensions by many different template‐free versatile techniques such as droplet microfluidics (50 μm in diameter),91 delayed precipitation (diameter between 1 and 10 μm),92 sono‐ electrochemical polymerization (diameter between 1 and 4 μm),93 and by dispersion polymerization (diameter between 0.2 and 2 μm).94,95 The combination of these top‐ down approaches offers large perspectives of active‐material co‐integrations, anisotropic assembling variabilities and three‐dimensional interconnection for high density networking. On the growth and interconnection, electropolymerization is a voltage‐guided synthetic process which allows the nucleation and growing of conducting polymer objects, potentially providing an intimate coupling between device operation and network modification. As an example, Inagi et al. have demonstrated the possibility to synthesize dendritic PEDOT fibers using bipolar electrochemistry96‐98. In a polarized AC electric field, the conducting polymer fibers bridge neighboring gold wires and propagates from wire to wire along this field. The voltage‐guided propagation of the conducting polymer growth promotes the arborescence of a conducting network, while voltage also promotes the information conduction through the network. Therefore, conducting polymer technologies offer the perspective to couple both the physical evolution of the network with data processing. Combining the different soft‐techniques of town‐down assembling of conducting polymers and bottom‐up electropolymerized interconnection could provide an elegant technique to realize such functional bioinspired neural 23 network,99 but also to have a coupling between network morphological change and operation, necessary to mimic long‐term memory at the network level. a) b) c) 10 s E=30.3 V.cm-1 25 s E=30.3 V.cm-1 29 s E=30.3 V.cm-1 Figure 5: (a) Assembling of conducting polymer microspheres (adapted from [92]), scale bar is 5um. (b) Dendritic PEDOT:PSS interconnection (adapted from [97]). (c) Electrical interconnection of microspheres (adapted from [99]). 4‐ Conclusion In summary, the recent results reviewed in this paper have demonstrated that organic materials and devices are prone for the implementation of neuromorphic functions, not only at the single device level, but also at the circuit level. Some perspectives are discussed showing the possibility to improve these systems towards an increased complexity in terms of functions and circuit connectivity. From a very general viewpoint, table 1 points out the intrinsic properties of organic iono‐electronic materials and inorganic silicon‐based technologies. At the light of this table, the different perspectives 24 from materials, devices, to system level for neuromorphic engineering can be evaluated. Since neuromorphic engineering is still an emerging computing paradigm, the question of the most appropriate technology is still completely open. This last statement should be of course balanced by the very important criterion of technology maturity that will enable neuromorphic technologies to be transferred to the industrial level. Conventionnal computer: • Low impedance • Operating conditions: dry, high S/N ratio, fast • Communication: electronic • Computing: digital, serial • organization: Top‐down Material Device System Biological hardware: • High impedance • Operating conditions: wet, low signal/ noise ratio, slow • Communication: Ionic, chemical • Computing: analog, parallel • organization: Bottom‐up thickness Fabrication top‐down lithography top‐down lithography bottom‐up self assembling, Table 1: Comparison of the main intrinsic properties for both organic and inorganic technologies. These properties could be thought in the light of the main characteristics of biological computing systems. eletropolymerization Bottom‐up 25 Operational environment Information carrier Mobility /(V.s)) (cm 2 2 Capacity (μF/cm ) CMOS technology Dry (sealed packaged) electrons / holes 3 1‐10 (silicon) 1 (high‐k oxide with thickness of few nm) note: Capa. decreases with thickness Biology Wet Organic ion/electronic technology Air / water stable electrons / holes / ions / chemicals ions / chemicals ‐6 10 to 10 (electronic) ‐4 10 ‐3 to 10 (ionic) 500 (130 nm PEDOT film with volumic capacitance) note: Capa. increases with ‐3 10 (ionic) 1 (cell's membrane) Parts of the works reviewed here have been financially supported by the European Union projects NABAB (FP7‐FET‐FP7‐216777), SYMONE (FP7‐FET‐318597), I‐ONE (FP7‐NMP‐280772), RECORD‐IT (H2020‐FET‐664786) and the French Agency ANR project SYNAPTOR (12 BS03 010 01). Acknowledgements. 26 References: 1. C. Mead, Proceedings of the IEEE (1990), p. 1629. 2. C. Bartolozzi, R. Benosman, K. Boahen, G. Cauwenberghs, T. Delbrück, G. Indiveri, S.‐C. Liu, S. Furber, N. Imam, B. Linares‐Barranco, T. Serrano‐Gotarredona, K. Meier, C. Posch and M. Valle, Neuromorphic systems. (Wiley Encyclopedia of Electrical and Electronics Engineering, 2016). 3. G. Indiveri and S.‐C. Liu, Proceedings of the IEEE (2018), p. 1379‐. 4. D. B. Strukov, Nature 476, 403 (2011). 5. S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder and W. Lu, Nano Lett. 10, 1297 (2010). 6. D. Kuzum, R. G. Jeyasingh, B. Lee and H. S. P. Wong, Nano Lett. 12, 2179 (2011). 7. F. Alibart, S. Pleutin, O. Bichler, C. Gamrat, T. Serrano‐Gotarredona, B. Linares‐Barranco and D. Vuillaume, Adv. Funct. Mater. 22, 609 (2012). 8. J. Grollier, D. Querlioz and M. D. Stiles, Proceedings of the IEEE (2016), p. 2024. 9. J. Torrejon, M. Riou, F. A. Araujo, S. Tsunegi, G. Khalsa, D. Querlioz, P. Bortolotti, V. Cros, K. Yakushiji, A. Fukushima, H. Kubota, S. Yuasa, M. D. Stiles and J. Grollier, Nature 547, 428 (2017). 10. P. Stoliar, J. Tranchant, B. Corraze, E. Janod, M.‐P. Besland, F. Tesler, M. Rozenberg and L. Cario, Adv. Funct. Mater. 27, 164740 (2017). 11. M. Prezioso, F. Merrikh‐Bayat, B. D. Hoskins, G. C. Adam, K. K. Likharev and D. B. Strukov, Nature 521, 61 (2015). 12. G. W. Burr, R. M. Shelby, S. Sidler, C. Di Nolfo, J. Jang, I. Boybat, R. S. Shenoy; P. Narayanan, K. Viwani, E. U. Giacometti, B. N. Kürdi and H. Hwang, IEEE Trans. Electron Devices 62, 3498 (2015). 27 13. F. Alibart, S. Pleutin, D. Guerin, C. Novembre, S. Lenfant, K. Lmimouni, C. Gamrat, and D. Vuillaume, Adv. Funct. Mater. 20, 330 (2010). 14. O. Bichler, W. Zhao, F. Alibart, S. Pleutin, D. Vuillaume and C. Gamrat, IEEE Trans. Electron. Dev. Soc. 57, 3115 (2010). 15. O. Bichler, W. Zhao, F. Alibart, S. Pleutin, S. Lenfant, D. Vuillaume, and C. Gamrat, Neural Comput. 25, 549 (2013). 16. S. Desbief, A. Kyndiah, D. Guerin, D. Gentili, M. Murgia, S. Lenfant, F. Alibart, T. Cramer, F. Biscarini and Dominique Vuillaume, Org. Electron. 21, 47 (2015). 17. S. Desbief, M. di Lauro, S. Casalini, D. Guerin, S. Tortorella, M. Barbalinardo, A. Kyndiah, M. Murgia, T. Cramer, F. Biscarini and D. Vuillaume, Org. Electron. 38, 21 (2016). 18. M. di Lauro, M. Berto, M. Giordani, S. Benaglia, G. Schweicher, D. Vuillaume, C. A Bortolotti, Y. H Geerts and F. Biscarini, Adv. Electron. Mater. 3, 1700159 (2017). 19. J. A. Varela, K. Sen, J. Gibson, J. Fost, J. F. Abbott and J. B. Nelson, J. Neurosci. 17, 7926 (1997). 20. G. Q. Bi and M. M. Poo, J. Neurosci. 18, 10464 (1998). 21. J. Borghetti, V. Derycke, S. Lenfant, P. Chenevier, A. Filoramo, M. Goffman, D. Vuillaume and J.‐P. Bourgoin, Adv. Mater. 18, 2535 (2006). 22. C. Anghel, V. Derycke, A. Filoramo, S. Lenfant, B. Giffard, D. Vuillaume and J.‐P. Bourgoin, Nano Lett. 8, 3619 (2008). 23. G. Agnus, W. Zhao, V. Derycke, A. Filoramo, Y. Lhuillier, S. Lenfant, D. Vuillaume, C. Gamrat and J.‐P. Bourgoin, Adv. Mater. 22, 702 (2010). 24. K. Gacem, J.‐M. Retrouvey, D. Chabi, A.Filoramo, W. Zhao, J.‐O. Klein and V. Derycke, Nanotechnology 24, 384013 (2013). 28 25. W. S. Zhao, G. Agnus, V. Derycke, A. Filoramo, J.‐P. Bourgoin and C. Gamrat, Nanotechnology 21, 175202 (2010). 26. Y.‐P. Lin, C. H. Bennett, T. Cabaret, D. Vodenicarevic, D. Chabi, D. Querlioz, B. Jousselme, V. Derycke, and J.‐O. Klein, Sci. Rep. 6, 31932 (2016). 27. V. Erokhin, T. Berzina, and M. P. Fontana, J. Appl. Phys. 97, 064501 (2005). 28. V. V. Erokhin, T. S. Berzina and M. P. Fontana, Crystallog. Rep. 52, 159 (2007). 29. A Smerieri, T. Berzina, V. Erokhin and M. P. Fontana, J. Appl. Phys. 104, 114513. (2008) 30. T. Berzina, A. Smerieri, M. Bernabò, A. Pucci, G. Ruggeri, V. Erokhin and M. P. Fontana, J. Appl. Phys. 105, 124515 (2009). 31. T. Berzina, S. Erokhina, P. Camorani, O. Konovalov, V. Erokhin and M. Fontana, ACS Appl. Mater. Interfaces 1, 2115 (2009). 32. V. A. Demin, V. V. Erokhin, A. V. Emelyanov, S. Battistoni, G. Baldi, S. Iannotta and M. V. Kovalchuk, Org. Electron. 25, 16 (2015). 33. F. Rosenblatt, Psychol. Rev. 65, 386 (1958). 34. C. Quian, J. Sun, L. Kong, G. Gou, J. Yang, J. He, Y. Gao and G. Wan, ACS Appl. Mater. Interfaces 8, 26169 (2016). 35. S. Lu, F. Zeng, W. Dong, A. Liu, X. Li and J. Luo, Nano‐Micro Lett. 7, 121 (2015). 36. E. Mitraka, M. J. Jafari, M. Vagin, X. Liu, M. Fahlman, T. Ederth, M. Berggren, M. P. Jonsson and X. Crispin, J. Mater. Chem. A 5, 4404 (2017). 37. S. Li, F. Zeng, C. Chen, H. Liu, G. Tang, S. Gao, C. Song, Y. Lin, F. Pan and D. Guo, J. Mater. Chem. C 1, 5292 (2013). 38. P. Gkoupidenis, N. Schaefer, B. Garlan and G. G. Malliaras, Adv. Mater. 27, 7176 (2015). 29 39. B. Winther‐Jensen, B. Kolodziejczyk and O. Winther‐Jensen, APL Materials 3, 014903 (2015). 40. P. Gkoupidenis, N. Schaefer, X. Strakosas, J. A. Fairfield and G. G. Malliaras, Appl. Phys. Lett. 107, 263302 (2016). 41. Y. Van de Burgt, E. Lubberman, E. J. Fuller, S. T. Keene, G. C. Faria, S. Agarwal, M. J. Marinella, A. A. Talin and A. Salleo, Nat. Mater. 16, 414 (2017). 42. J. J. Yang, D. B. Strukov and D. R. Stewart, Nat. Nanotechnol. 8, 13. (2013). 43. Z. Wang, S. Joshi, S. E. Savel'ev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan, Z. Li, Q. Wu, M. Barnell, G.‐L. Li, X. L. Xin, R. S. Stanley Williams, Q. Xia and J. Yang, Nat. Mater. 16, 101 (2017). 44. T. Ohno, T. Hasegawa, T. Tsuruoka, K. Terabe, J. K. Gimzewski and M. Aono, Nat. Mater. 10, 591 (2011). 45. M. Tada, K. Okamoto, T. Sakamoto, M. Miyamura, N. Banno and H. Hada, IEEE Trans. Electron Devices 58, 4398 (2011). 46. D. J. Carad, A. M. Burje, R. W. Lyttleton, H. J. Joyce, H. H. Tan, C. Jagadish, K. Storm, H. Linke, L. Samuelson and A. P. Micolich, Nano Lett. 14, 94 (2014). 47. V. Benfenati, S. Toffanin, S. Bonetti, G. Turatti, A. Pistone, M. Chiappalone, A. Sagnella, A. Stefani, G. Generali, G. Ruani, D. Saguatti, R. Zamboni and M. Muccini, Nat. Mater. 12, 672 (2013). 48. D. Khodagholy, T. Doublet, P. Quilichini, M. Gurfinkel, P. Leleux, A. Ghestem, E. Ismailova, T. Hervé, S. Sanaur, C. Bernard and G. G. Malliaras, Nat. Commun. 4, 1575 (2013). 49. L. J. Juarez‐Hernandez, N. Cornella, L. Pasquardini, S. Battistoni, L. Vidalino, L. Vanzetti, S. Caponi, M. D. Serra, S. Ianotta, C. Pederzolli, P. Macchi and C. Musio, Biophys. Chem. 208, 40 (2016). 30 50. S. Pecqueur, D. Guérin, D. Vuillaume and F. Alibart, Org. Electron. 57, 232 (2018). 51. E. Stavrinidou, P. Leleux, H. Rajaona, D. Khodagholy, J. Rivnay, M. Lindau, S. Sanaur and G. G. Malliaras, Adv. Mater. 25, 4488 (2013). 52. J. Rivnay, P. Leleux, M. Ferro, M. Sessolo, A. Williamson, D. A. Koutsouras, D. Khodagholy, M. Ramuz, X. Strakosas, R. M. Owens, C. Benar, J.‐M. Badier, C. Bernard and G. G. Malliaras, Sci. Adv. 1, e1400251 (2015). 53. O. E. Barcia, E. D'Elia, I. Frateur, O. R. Mattos, N. Pébère, and B. Tribollet, Electrochim. Acta 47, 2109 (2002). 54. B. A. Boukamp, and H. J. Bouwmeester, Solid State Ionics 157, 29 (2003). 55. C. Bédard, S. Rodrigues, N. Roy, D. Contreras and A. Destexhe, J. Comput. Neurosci. 29, 389 (2010). 56. C. Bédard, H. Kröger and A. Destexhe, Biophys. J. 86, 1829 (2004). 57. J.‐M. Gomes, C.Bédard, S. Valtcheva, M. Nelson, V. Khokholova, P. Pouget, L. Venance, T. Bal and A. Destexhe, Biophys. J. 110, 234 (2016). 58. S. Inal, J. Rivnay, P. Leleux, M. Ferro, M. Ramuz, J. C. Brendel, M. M. Schmidt, M. Thelakkat and G. G. Malliaras, Adv. Mater. 26, 7450 (2014). 59. E. Zeglio, M. Vagin, C. Musumeci, F. N. Ajjan, R. Gabrielsson, X. T. Trinh, N. T. Son, A. Maziz, N. Solin and O. Inganäs, Chem. Mater. 27, 6385 (2015). 60. E. Zeglio, J. Eriksson, R. Gabrielsson, N. Solin and O. Inganäs, Adv. Mater. 29, 1605787 (2017). 61. C. B. Nielsen, A. Giovannitti, D.‐T. Sbircea, E. Bandiello, M. R. Niazi, D. A. Hanifi, M. Sessolo, A. Amassian, G. G. Malliaras, J. Rivnay and I. McCulloch, J. Am. Chem. Soc. 138, 10252 (2016). 62. A. Giovannitti, D. T. Sbircea, S. Inal, C. B. Nielsen, E. Bandiello, D. A. Hanifi, J. Rivnay, Proc. Natl. Acad. Sci. U. S. A. 113, 12017 (2016). 31 63. M. M. Shulaker, G. Hills, R. S. Park, R. T. Howe, K. Saraswat, H. S. P. Wong and S. Mitra, Nature 547, 74 (2017). 64. P. O'Connor, D. Neil, S. C. Liu, T. Delbruck and M. Pfeiffer, Front. Neurosci. 7, 178 (2013). 65. A. Kholer and H. Bässler, Electronic Processes in Organic Semiconductors: An Introduction. (Wiley‐VCH Verlag GmbH & Co. KGaA, 2015). 66. C. W. Tang and S. A. VanSlyke, Appl. Phys. Lett. 51, 913 (1987). 67. F. Guo, A. Karl, Q. F. Xue, K. C. Tam, K. Forberich and C. J. Brabec, Light: Sci. Appl. 6, e17094 (2017). 68. K. Kudo and T. Morrizumi, Appl. Phys. Lett. 39, 609 (1981). 69. F. Arca, S. F. Tedde, M. Sramek, J. Rauh, P. Lugli and O. Hayden, Sci. Rep. 3, 1324 (2013). 70. R. D. Jansen‐van‐Vuuren, A. Armin, A. K. Pandey, P. L. Burn and P. Meredith, Adv. Mater. 28, 4766 (2016). 71. S. E. Root, S. Savagatrup, A. D. Printz, D. Rodriquez and D. J. Lipomi, Chem. Rev. 117, 6467 (2017). 72. A. Khaldi, C. Plesse, C. Soyer, E. Cattan, F. Vidal, C. Legrand and D. Teyssié, Appl. Phys. Lett. 98, 164101 (2011). 73. C. Yeon, G. Kim, J. W. Lim and S. J. Yun, RSC Adv. 7, 5888 (2017). 74. Y. Wang, C. Zhu, R. Pfattner, H. Yan, L. Jin, S. Chen, F. Molina‐Lopez, F. Lissel, J. Liu, N. I. Rabiah, Z. Chen, J. W. Chung, C. Linder, M. F. Toney, B. Murmann and Z. Bao, Sci. Adv. 3, e1602076 (2017). 75. B. Lüssem, M. Riede and K. Leo, Phys. Status Solidi A 210, 9 (2013). 32 76. S. Pecqueur, A. Maltenberger, M. A. Petrukhina, M. Halik, A. Jaeger, D. Pentlehner, G. Schmid, Angew. Chem. Int. Ed. 55, 10493 (2016), Angew. Chem. 128, 10649 (2016). 77. F. Cicoira, M. Sessolo, O. Yaghmazadeh, J. A. DeFranco, S. Y. Yang and G. G. Malliaras, Adv. Mater. 22, 1012 (2010). 78. S. Pecqueur, S. Lenfant, D. Guérin, F. Alibart and D. Vuillaume, Sensors 17, 570 (2017). 79. J. Liang and H. S. P. Wong, IEEE Trans. Electron Devices 57, 2531 (2010). 80. P. A. Merolla, J. V. Arthur, R. Alvarez‐Icaza, A. S. Cassidy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y. Nakamura, B. Brezzo, I. Vo, S. K. Esser, R. Appuswamy, B. Taba, A. Amir, M. D. Flickner, W. P. Risk, R. Manohar and D. S. Modha, Science 345, 668 (2014). 81. S. B. Furber, F. Galluppi, S. Temple and L. A. Plana, Proceedings of the IEEE (2014), p. 652. 82. M. Häusser, N. Spruston, and G. J. Stuart, Science 290, 739 (2000). 83. W. Xu, S. Y. Min, H. Hwang, H. and T. W. Lee, Sci. Adv. 2, e1501326 (2016). 84. P. Gkoupidenis, D. A. Koutsouras, T. Lonjaret, J. A. Fairfield and G. G. Malliaras, Sci. Rep. 6, 27007 (2016). 85. Y. H. Liu, L. Q. Zhu, P. Feng, Y. Shi and Q. Wan, Adv. Mater. 27, 5599. (2015) 86. C. H. Yang, B. Chen, J. J. Lu, J. H. Yang, J. Zhou, Y. M. Chen, Z. Suo, Extreme Mech. Lett. 3, 59 (2015). 87. P. Gkoupidenis, D. A. Koutsouras and G. G. Malliaras, Nat. Commun. 8, 15448 (2017). 33 88. S. Pecqueur, M. Mastropasqua‐Talamo, D. Guérin, P. Blanchard, J. Roncali, D. Vuillaume and F. Alibart, Adv. Electron. Mater., In press (doi: 10.1002/aelm.201800166). 89. S. Pecqueur, D. Guérin, D. Vuillaume and F. Alibart, Org. Electron. 57, 232 (2018). 90. J. Hawkins, Proceeding of International Electron Device Meeting (2015), p. 95. 91. S. N. Beesabathuni, J. G. Stockham, J. H. Kim, H. B. Lee, J. H. Chung and A. Q. Shen, RSC Adv. 3, 24423 (2013). 92. K. Tabata, D. Braam, S. Kushida, L. Tong, J. Kuwabara, T. Kanbara, A. Beckel, A. Lorke and Y. Yamamoto, Sci. Rep. 4, 5902 (2014). 93. M. Atobe, K. Ishikawa, R. Asami and T. Fuchigami, Angew. Chem. Int. Ed. 48, 6069 (2009). 94. A. J. C. Kuehne, M. C. Gather, and J. Sprakel, Nat. Commun. 3, 1088 (2012). 95. S. Ciftci and A. J. C. Kuehne, Macromolecules 48, 8389 (2015). 96. M. Ohira, Y. Koizumi, H. Nishiyama, I. Tomita and S. Inagi, Polymer J. 49, 163 (2017). 97. Y. Koizumi, N. Shida, M. Ohira, H. Nishiyama, I. Tomita, S. Inagi, Nat. Commun. 7, 10404 (2016). 98. T. Watanabe, M. Ohira, Y. Koizumi, H. Nishiyama, I. Tomita, S. Inagi, ACS. Macro. Lett. 7, 551 (2018). 99. J. C. Bradley, H. M. Chen, J. Crawford, J. Eckert, K. Ernazarova, T. Kurzeja, M. Lin, M. McGee, W. Nadler and S. G. Stephens, Nature 389, 268 (1997). 34 a) c) b) d) Electrode + + + _ _ _ Input port + + + x _ _ _ + + + _ _ _ i(x,t) i(x,t) Electrical double layer v(x,t) Ionic conductor + + + + + + _ _ _ _ _ _ Ionic conductor Dielectric + + + _ _ _ + + + _ _ _ Output port Z a) b) c) 10 s E=30.3 V.cm-1 25 s E=30.3 V.cm-1 29 s E=30.3 V.cm-1 Conventionnal computer: •  Low impedance •  Operating conditions: dry, high S/N •  Communication: electronic •  Computing: digital, serial •  organization: Top-down ratio, fast Material Device System Biological hardware: noise ratio, slow •  High impedance •  Operating conditions: wet, low signal/ •  Communication: Ionic, chemical •  Computing: analog, parallel •  organization: Bottom-up
1802.08967	2	1802	2018-03-02T12:58:10	Shape Control for Experimental Continuation	[ "physics.app-ph", "cond-mat.soft" ]	An experimental method has been developed to locate unstable equilibria of nonlinear structures quasi-statically. The technique involves loading a structure by application of either a force or a displacement at a main actuation point, while simultaneously controlling the overall shape using additional probe points. The method is applied to a shallow arch, and unstable segments of its equilibrium path are identified experimentally for the first time. Shape control is a fundamental building block for the experimental---as opposed to numerical---continuation of nonlinear structures, which will significantly expand our ability to measure their mechanical response.	physics.app-ph	physics	Shape Control for Experimental Continuation Robin M. Neville, Rainer M. J. Groh, Alberto Pirrera,∗ and Mark Schenk Bristol Composites Institute (ACCIS), Department of Aerospace Engineering, University of Bristol, Bristol, BS8 1TR, United Kingdom An experimental method has been developed to locate unstable equilibria of nonlinear structures quasi-statically. The technique involves loading a structure by application of either a force or a displacement at a main actuation point, while simultaneously controlling the overall shape using additional probe points. The method is applied to a shallow arch, and unstable segments of its equilibrium path are identified experimentally for the first time. Shape control is a fundamental building block for the experimental -- as opposed to numerical -- continuation of nonlinear structures, which will significantly expand our ability to measure their mechanical response. I. BACKGROUND The force-displacement response of nonlinear struc- tures can be complex and even chaotic. Limit and branch points can partition equilibrium manifolds into stable and unstable segments. In particular, displacement and force limit points change the stability of a structure and inject unstable eigenmodes into the deformation shape, thereby rendering certain segments of the ensuing force- displacement manifolds inaccessible experimentally. This kind of behaviour is observed even in simple structures, such as the shallow arch studied in this paper. Figure 1 shows how force limit points cause force- controlled structures to snap to the next available stable equilibrium. Similarly, displacement limit points cause displacement-controlled structures to snap while conserv- ing the displacement at the point(s) of actuation. To trace an equilibrium manifold like that shown in Figure 1, a means of controlling both the forces acting on the struc- ture and its global deformation is required. This combi- nation is readily implemented in a numerical setting be- cause force and deformation can be independently con- trolled via a third parameter, namely the arc-length [1]. However, tracing similar equilibrium manifolds experi- mentally remains an open challenge. The challenge is that force and displacement at the actuation point(s) are not independent, but rather inherently linked through elasticity. A force applied at a specific control point re- sults in a displacement at that point; similarly, an applied displacement induces a reaction force. This differentiates quasi-static from dynamic problems [2] where the input vibration frequency and amplitude are decoupled. In this paper, we present a general method to explore the unstable equilibria of a nonlinear structure quasi- statically. The key to this technique is decoupling force and displacement at a specific control point by introduc- ing a third control variable: the shape of the structure. Consider a shallow arch loaded transversely at its mid- point as shown in Figure 2A, which is a known bench- mark for numerical arc-length solvers. For given combi- nations of geometry and material parameters, the arch ∗ [email protected] FIG. 1. Limit points cause displacement and load control to snap to different parts of the equilibrium curve, resulting in an experimentally inaccessible region. features particularly pronounced nonlinear behaviour [3] with many unstable loops in force-displacement space (Figure 3A). These loops give rise to the problem -- accessing unstable equilibria and tracing experimentally inaccessible segments of equilibrium manifolds -- but also the insight for a solution. For loops to exist, the struc- ture must exhibit multiple different values of midpoint force for one midpoint displacement, and vice versa. By examination of the various arch shapes in Figure 3C and corresponding force values, it is evident that each force corresponds to a unique arch shape. Hence, each com- bination of midpoint force, midpoint displacement and deformation shape corresponds to a unique equilibrium. Control over the deformation shape is the key ingredi- ent for decoupling force and displacement at the control point. For example, for a given midpoint displacement, the reaction force at the same point can be controlled indirectly by changing the deformation shape. The idea of separating force and displacement at a con- trol point through shape control provides the first fun- damental building block towards experimental continu- ation. The method for shape control is shown in Fig- force controlforce controldisp. controldisp. controlforce limitforce limitdisp. limitdisp. limitdisplacementforceinaccessiblecurrently2 II. EXPERIMENTAL METHODS A. Experimental Setup and Equipment Figure 4 gives a detailed explanation of the experi- mental setup, including implementation of the loads and boundary conditions. An Instron 8872 hydraulic test machine with an In- stron Dynacell ±250 N tension/compression dynamic load cell (load cell 1) was used for all displacement- Load cell 2 was a ±500 N ten- controlled tests. sion/compression load cell manufactured by Applied Measurements Ltd. A Gefran PZ34-A-250 linear trans- ducer was used to measure the height of the midpoint platform. LabView (version 14.0) was used to log exper- imental data to ensure that synchronised readings were obtained from the various sensors. An Imetrum Video Gauge camera system was used to record the tests. B. Arch Specimens The geometry of the arches tested is shown in Figure 2, with dimensions L = 205 mm, h = 20 mm, t = 1.57 mm, and depth D = 4.68 mm (into the page). Ten speci- mens were manufactured using a Trotec Speedy 100 laser engraver to cut the arches from sheets of acrylic (sup- plied by F.R. Warren Ltd). Mechanical coupon test- ing was performed to find the Young's modulus E = 3200 ± 70 MPa and Poisson's ratio ν = 0.38 ± 0.02 for use as inputs to Finite Element (FE) models. C. Test Sequence Two types of tests were performed -- midpoint scans and probe scans. Midpoint scans represent the standard displacement-controlled experimental approach. The test machine is connected to the midpoint clamp, and no probes are used. The midpoint is moved down and back up under displacement control. This produces a load- displacement curve similar to the solid blue lines in Fig- ure 3B. The two segments of the equilibrium curve cor- respond to the "downwards" and "upwards" parts of the test. At limit points L1 and L2, the arch snaps to the other blue segment. A midpoint scan was performed for each of the 10 arch specimens. In a probe scan, the configuration in Figure 4B is used. The midpoint is fixed at a given displacement, and the probes are moved down and back up under displacement control. During this test, the arch passes through both stable equilibrium segments, and one or more unstable segments, as depicted in Figure 5. By repeating the probe scans at different midpoint displacements, the location of additional unstable equi- librium segments can be "mapped out" without path- following. For each arch specimen, a probe scan was performed for δm = {8, 12, 16, 20, 24, 28, 32} mm. FIG. 2. (A) The shallow arch structure studied in this paper. The edges of the arch are pinned. A vertical displacement (δm) or force (Pm) is applied to the midpoint. Rotations and lateral translations are constrained at the midpoint to pre- serve symmetry. (B) Additional control points provide shape control. A displacement (δp) or force (Pp) is applied sym- metrically to the "probes" halfway between the midpoint and edges. The probes allow rotations and lateral displacements in order to prevent reaction moments and horizontal reaction forces that would force the structure into a different equilib- rium. ure 2B. Namely, we add two (to enforce symmetry) probe points, which allow us to manipulate the arch shape using another displacement-controlled input. Experimental continuation requires feedback control interlinking loading and shape. Herein, we propose a simpler experiment as an initial step towards the full ca- pability. Rather than moving the midpoint and probes simultaneously, we fix the midpoint at a given displace- ment and move the probes to scan for other equilibria. When the force on the probes reads zero, an equilib- rium state of the system is found. With this method, which was recently used to determine localised solutions of the axially compressed cylinder [4, 5], we find unsta- ble equilibria which have never before been pinpointed quasi-statically. This concept of obtaining a zero-force reading on the probe to determine equilibria is analo- gous to the minimisation of virtual work in response to a probing virtual displacement, the vanishing of the resid- ual in Newton's method at a converged state, and zero control in dynamic experimental continuation [2]. In previous work, the existence of unstable static equi- libria has been intuited dynamically in the transients in- duced by large perturbations [6]. Our approach differs in that the experimental setup stabilises otherwise unsta- ble equilibria using additional control points. Additional control points have been used to constrain [7] or probe [5] nonlinear structures in one direction, but are here rigidly fixed to the structure to allow the push/pull control re- quired to scan for additional equilibria. 3 (A) The entirety of the FIG. 3. FE prediction of the highly nonlinear load-displacement behaviour of the shallow arch. symmetric response. The arch shapes corresponding to the first and final equilibria are shown. (B) A subset of the response, with the solid blue lines indicating the segments a displacement-controlled experiment would obtain; dashed lines indicate equilibrium configuration currently inaccessible experimentally. At the limit points L1 and L2, displacement control snaps to the opposite blue segment of the equilibrium curve. At δm = 5 there exist multiple load values for a given midpoint displacement (points 1 -- 5). (C) The arch shapes at points 1 -- 5 for a fixed midpoint displacement. FIG. 4. Experimental setup. (A) The test fixture supporting the arch is bolted to the base of the test machine. The dashed rectangle shows the camera's field of view. (B) Physical implementation of the arch loads and boundary conditions. The idealised model is shown above for reference. The test machine controls the probe displacement δp and load cell 1 measures the probe force Pp. A moveable platform sets the midpoint displacement δm. A linear transducer measures the position of the midpoint platform. Load cell 2 measures the midpoint force Pm. (C) Wedge-shaped blocks restrain translations and allow rotations of the arch ends. (D) A linear guide rail allows the probes to move in the X direction, while controlling the displacement along Y. The probes connect to the arch with pins which allow rotations. (E) A clamp restrains all translations and rotations at the arch midpoint to maintain symmetry. The clamped area is 5 mm wide. 010203040-20-10010201234512345ABCL2PmPm(N)L14 controlled structure. The pairs of data points where Pp crossed zero were identified, and the δp values cor- responding to Pp = 0 were found by linear interpola- tion. The gradient of the load-displacement curve and the shape of the arch at the equilibrium point revealed the segment to which the equilibrium belonged. Two unstable segments were detected, in addition to the two known stable segments (Figure 6A). In all probe scans, one unstable equilibrium was found in the first ("downwards") part of the test, and the other was found in the second ("upwards") part of the test. Consequently, equilibria corresponding to the downwards/upwards part of the test are labelled with a 1 and a 2, respectively. At each δm location, the equilibria detected by the probe scans were grouped by type, and the mean and standard deviation of δm and Pm were calculated for each group. The aggregate data for each δm location were then connected to give the mean ± standard deviation plot shown in Figure 6A. The results of the probe scans match the results of the midpoint scans fairly well, de- spite the fact that the two types of tests are performed separately, and with different equipment set-ups. Con- sequently, it is assumed that the unstable segments have been located with a similar degree of accuracy. Discrepancies between the probe and midpoint scan results may be due to relaxation of the specimen mate- rial. Upon removing the arch from the fixture after the final probe scan, it was noted that most specimens did not immediately spring back to their undeformed shape. Imperfections in the experimental set-up (especially the boundary conditions) may also have contributed to the differences. An FE analysis of the arch was performed using non- linear beam elements and idealised pinned boundary con- ditions. Figure 6B shows the FE prediction, with the seg- ments colour-coded to match their experimental counter- parts in Figure 6A. This shows that we have located the "next" two segments of the arch response -- i.e. the seg- ments beyond limit points L1 and L2. These segments correspond to arch shapes which are not stable under midpoint control only (shapes 2 and 4 in Figure 3C), but are stable when supported by the probes. Further seg- ments will correspond to more complex shapes (e.g. shape 3 in Figure 3C), which will require additional probes for support. Despite the sensitivity to initial conditions in nonlin- ear systems, there is excellent qualitative and quantita- tive agreement between the experimental and theoreti- cal results. This provides confidence that the unstable equilibria have been correctly identified by the testing method. FIG. 5. A schematic midpoint scan is plotted in black on the Pm-δm plane. The probe scan takes place on the green Pp-Pm plane, which intersects the midpoint scan at a fixed value of δm. Starting at shape A, the probes move down to shape B. The probe reaction force, Pp, is plotted on the green plane. For Pp = 0, an equilibrium configuration is detected. The probe scan detects the segments of the equilibrium curve which are stable with midpoint control only (A and B), and also detects an unstable segment (C). It is important to note that this method does not require us to follow or balance on an unstable equi- librium path; the structure is simply pushed through an unstable equilibrium and its location on the mid- point load-displacement curve is measured. Conse- quently, the probe scan experiment can be performed by a displacement-controlled test machine. In future exper- iments, the probes will be controlled (via a more sophis- ticated feedback-control approach) to seek zero reaction force, and follow an unstable equilibrium segment while moving the midpoint. III. RESULTS A. Midpoint Scans The midpoint scan δm, Pm data from 10 specimens were split into the "downwards" and "upwards" portions of the test, to prevent the loops in the plot affecting the following calculations. The data were then separated into 1 mm wide bins, and the mean and standard deviation of δm and Pm were found for each bin. Figure 6A shows the results in purple and blue. The width of the filled area indicates the confidence interval of the measurements, based on the standard deviation. B. Probe Scans IV. CONCLUSIONS AND OUTLOOK The probe scan δp, Pp data were analysed to find all the zero crossings of Pp -- i.e. the equilibria of the midpoint- We have presented an experimental method to detect and identify unstable equilibria of nonlinear structures quasi-statically, using probes to control the shape of the ACB=05 FIG. 6. Comparison of experimental and FE results. (A) Midpoint scan and probe scan results. Midpoint data after the snaps have been omitted. The white lines show the mean of the data, and the coloured areas represent ± one standard deviation of Pm. The midpoint scans are indicated by an "M" in the legend; the probe scans by a "P". The vertical white bars show the δm locations of the probe scans. (B) FE prediction of the midpoint symmetric load-displacement response, truncated to the "first" few segments as in Figure 3B. The segments are colour-coded to match their counterpart in the experimental results. structure. Using this method, we have, for the first time experimentally, shown the location of unstable equilibria of a shallow arch which would not be accessible using tra- ditional quasi-static testing techniques. These equilibria correspond to structural shapes that have zero reaction force at the probe points. The shape control provides independent, albeit indirect, control over force and dis- placement at the point of actuation, which are otherwise intrinsically linked. The natural extension of this work is to exploit the probing technique as a means to trace equilibrium paths of nonlinear structures. This may be achieved through a concerted variation in actuation point force or displace- ment, and shape control via the probes. The result is an experimental continuation technique, which enables the quasi-static nonlinear response of a structure to be explored systematically. In future, the addition of multi- ple, independent probes would provide more refined con- trol over the structural shape, thereby enabling more seg- ments of the equilibrium manifold to be identified exper- imentally. The development of continuation techniques using shape control will enable the experimental validation of the response of nonlinear structures. In turn, this will help encourage the exploitation of nonlinear structures in engineering applications, for example in morphing struc- tures and compliant mechanisms. V. ACKNOWLEDGMENTS This work was funded by the EPSRC through grant numbers EP/N509619/1, EP/M013170/1 and EP/P511298/1. EPSRC's support is gratefully acknowl- edged. The authors also wish to acknowledge the help of the University of Bristol technicians: G. Pearn, A. Kraft, R. Billingham, R. Bragg, and R. Hooper. [1] E. Riks, Journal of Solids and Structures 15, 529 (1979). [2] J. Sieber, A. Gonzalez-Buelga, S. Neild, D. Wagg, and B. Krauskopf, Physical Review Letters 100, 244101 (2008). [4] J. Thompson, J. Hutchinson, and J. Sieber, International Journal of Bifurcation and Chaos 27, 1730048 (2017). [5] E. Virot, T. Kreilos, T. M. Schneider, and S. M. Rubin- stein, Physical Review Letters 119, 224101 (2017). [3] H. Harrison, Proceedings of the Institution of Civil Engi- [6] R. Wiebe and L. N. Virgin, Proceedings of the Royal So- neers 65, 283 (1978). ciety A 472, 20160172 (2016). [7] P. Harvey and L. Virgin, International Journal of Solids and Structures 54, 1 (2015). 010203040-8-6-4-202468stable 1 (M)stable 1 (P)stable 2 (M)stable 2 (P)unstable 1 (P)unstable 2 (P)snapsnapA-8-6-4-202468010203040B(mm)(N)(N)(mm)L2L1
1806.06934	1	1806	2018-06-02T19:49:41	Digitized metamaterial absorber-based compressive reflector antenna for high sensing capacity imaging	[ "physics.app-ph", "physics.class-ph" ]	Conventional multistatic radar systems using microwave and millimeter-wave (mm-wave) frequencies seek to reconstruct the target in the imaging domain, employing many transmitting and receiving antenna elements. These systems are suboptimal, in that they do not take into consideration the large mutual information existing between the measurements. This work reports a new mm-wave radar system for high sensing capacity applications. The system is composed of a Compressive Reflector Antenna (CRA), whose surface is specially tailored by digitized Metamaterial Absorbers (MMAs). The MMA elements are designed to have a highly frequency-dispersive response in the operating band of the radar. This enables the CRA to create highly uncorrelated spatial and spectral codes in the imaging region. A semi-analytic method based on Drude-Lorentz model is used to approximate the reflection response of the MMAs. The performance of the developed radar system is evaluated in active mm-wave sensing systems by imaging PEC scatterers and an extended human-size model in the near-field of the radar. A computational method based on physical optics is established for solving the numerical examples. For reconstructing the image using compressive sensing techniques, a norm-1 regularized iterative algorithm based on the Alternating Direction Method of Multipliers (ADMM) and a Nesterov-based algorithm (NESTA) were applied.	physics.app-ph	physics	Digitized metamaterial absorber-based compressive reflector antenna for high sensing capacity imaging Ali Molaei, Juan Heredia-Juesas, Galia Ghazi, James Vlahakis and Jose Martinez-Lorenzo 1 8 1 0 2 n u J 2 ] h p - p p a . s c i s y h p [ 1 v 4 3 9 6 0 . 6 0 8 1 : v i X r a Abstract-Conventional multistatic radar systems using mi- crowave and millimeter-wave (mm-wave) frequencies seek to reconstruct the target in the imaging domain, employing many transmitting and receiving antenna elements. These systems are suboptimal, in that they do not take into consideration the large mutual information existing between the measurements. This work reports a new mm-wave radar system for high sensing capacity applications. The system is composed of a Compressive Reflector Antenna (CRA), whose surface is specially tailored by digitized Metamaterial Absorbers (MMAs). The MMA elements are designed to have a highly frequency-dispersive response in the operating band of the radar. This enables the CRA to create highly uncorrelated spatial and spectral codes in the imaging region. A semi-analytic method based on Drude-Lorentz model is used to approximate the reflection response of the MMAs. The performance of the developed radar system is evaluated in active mm-wave sensing systems by imaging PEC scatterers and an extended human-size model in the near-field of the radar. A computational method based on physical optics is established for solving the numerical examples. For reconstructing the image using compressive sensing techniques, a norm-1 regularized iterative algorithm based on the Alternating Direction Method of Multipliers (ADMM) and a Nesterov-based algorithm (NESTA) were applied. Index Terms-Compressive reflector antenna, millimeter-wave imaging, metamaterial absorber, coded aperture. I. INTRODUCTION M ILLIMETER-WAVE imaging is one of the few sensing techniques that can penetrate through walls, dielectric materials, and clothing, making it a good candidate for appli- cations such as security screening, surveillance, autonomous driving, non-destructive testing, and elsewhere [1]–[8]. Elec- tromagnetic waves are non-ionizing in the mm-wave frequency band, which makes the sensing harmless for radiation-sensitive targets. Unfortunately, the current sensing techniques, such as Synthetic Aperture Radar (SAR) [9], [10], require a mechan- ical raster motion to sweep the T x/Rx antennas along an ef- fective aperture. The SAR method needs an extensive amount of time to scan the imaging domain, making real-time imaging impractical. The development of a fast, fully electronic system does not require a mechanical scanning approach; however, implementing such system would be highly expensive due to the large number of T x/Rx radar modules required [2]. A. Molaei and G. Ghazi are with the Department of Electrical and Computer Engineering, Northeastern University, Boston, MA, 02115 USA. J. Vlahakis is with the Department of Mechanical Engineering, Tufts University, Medford, MA, USA. J. Heredia, and J. Martinez are with the Department of Electrical and Computer Engineering and Department of Mechanical and Industrial Engi- neering, Northeastern University, Boston, MA, 02115 USA e-mail: j.martinez- [email protected]. Recently, several fully electronic-based imaging systems have been proposed, aiming to perform the sensing faster (in real-time) and more efficiently. A frequency diverse one- dimensional radar, which is based on an active metamaterial aperture, is reported in [11]. The imaging system operates in the microwave frequencies (17.5-21.1 GHz). The metamaterial elements are dynamically controlled by voltage-controlled ac- tive diodes, producing a series of distinct radiation patterns that illuminate the imaging region. Due to the increased size and bandwidth of the active aperture, the obtainable information is increased and real-time imaging is achieved. As another example, in [12] a multistatic mm-wave imaging system with planar 2D-arrays is presented. A digital beamforming technique is used to reconstruct the images. Compared to the monostatic setup, the proposed array configurations (grid array and plus array) provide a considerably better dynamic range. In this work, a specially coded CRA [13]–[22], which is based on digitized MMAs, is proposed to be used for mm- wave sensing systems. The CRA has the ability to illuminate the imaging region with spatial and spectral coded patterns. The codes that will enable high bandwidth communications will be dynamically assigned while maximizing the overall performance of the network. The proposed mm-wave sensing system can be regarded as a coded aperture that modifies the effective Singular Value (SV) spectrum seen by the transmit- ting and receiving arrays. Contrary to the systems described above, the proposed CRA-based system can be reconfigured to operate in the far-field, thus enabling standoff detection of security threats. The paper is organized as follows: Section 2 describes the framework of the system and its underlying parameters. This is done through the following steps: (a) defining a linear equation that expresses the physics of the system; (b) describing the sensing capacity as a parameter to quantify the amount of information received by a sensing system. Section 3 is dedicated to the design and characterization of the digitized binary MMA: first, the polarization-independent meander-line MMA as a unit-cell element is introduced; next, the digitized MMAs built from the unit-cell meander- line elements, along with their frequency dispersive response, are presented; finally, the digitized MMAs are characterized based on a semi-analytical formulation derived from a multi- resonance Drude-Lorentz model. In Section 4, the performance of the system is validated through a numerical example for sensing metallic scatterers. The results of this testing are reported to demonstrate the performance of the system in terms of sensing capacity, radiation pattern, beam focusing, and image reconstruction. Section 5 describes a CRA-based array, composed of eight compressive reflectors, to image an extended human-size region. The surface of each CRA is tailored by digitized MMAs to enhance the sensing capacity of the array imaging system. Finally, the paper is concluded in Section 6 by discussing some extensions to the presented imaging systems. II. METAMATERIAL ABSORBER-BASED COMPRESSIVE REFLECTOR ANTENNA 2 The proposed MMA-based CRA system, as illustrated in Fig. 1, is composed of a custom designed reflector antenna with a 2D array of conical horns as feeding elements. The conical horns are configured as a plus shape along the focal plane of the reflector with the following arrangement: NT x transmitting and NRx receiving elements are positioned equi- distantly along the x-axis and y-axis, respectively; the antenna elements feed the reflector by a linear polarization in the y- axis. The MMA-based CRA consists of a Traditional Reflector Antenna (TRA), whose surface is covered with digitized MMAs. In a general case, an n-bit binary code ci = a1a2...an, i ∈ {1, ..., 2n}, is associated to each of the MMA de- signs. Each digit of the binary code aj ∈ {0, 1}, where j ∈ {1, 2, ..., n}, is associated with a resonance frequency fj within the operating frequency band. Depending on the binary value '0' or '1' associated to each digit aj, the MMA elements are designed to reflect or absorb the incident field at the resonance frequency fj. Figure 2, for example, shows the surface of a TRA divided into 16 domains and each domain is coated with a unique 4-bit MMA. The rest of this section is dedicated to describe the frame- work of the system and its underlying parameters. In sub- section II-A, a linear equation that expresses the physics of the system is defined. Then, in subsection II-B, the sensing capacity as a parameter to quantify the amount of information received by the sensing system is described. Fig. 1. Illustration of the proposed imaging setup: feeding array located on the focal plane induces currents on the surface of the CRA, which ultimately creates the spectral codes in the imaging region. As an example of these spectral codes, magnitude of electric field created by the CRA when illuminated by T x2 is shown as a function of frequency. A. Sensing Matrix Under the configuration mentioned previously, each receiver collects the signal from each transmitter for Nf different Fig. 2. Front view of the proposed CRA: surface of the reflector is divided into 16 domains and each domain is coated with a unique 4-bit MMA. frequencies, for a total number of Nm = NT x · NRx · Nf measurements. The image reconstruction domain is made up of Np pixels in the Region Of Interest (ROI) located zT 0 meters away from the focal plane of the reflector. Under this configuration, the vector of complex field measured data g ∈ CNm is related to the unknown complex vector of reflectivity u ∈ CNp, defined in each pixel, through the sensing . The vector hi is the i-th row of H, matrix hi = ET i are, respectively, the fields from the Tx and Rx ET i on the i-th measurement, and the symbol ◦ is the Hadamard product. and ERi ◦ ERi x x x x This relationship can be expressed in a matrix form as follows: (1) where w ∈ CNm represents the noise collected at each measurement. g = Hu + w, B. Sensing Capacity The maximum information rate has been studied by the field of information theory. It was pioneered by Shannon, almost seventy years ago [23]. A similar concept of Sensing Capacity with the aim to quantify the amount of information received by a sensing system has been studied in [13], and it is defined as follows: (cid:32) M(cid:89) m=1 (cid:33) m · em σH ≈ log2 CM M(cid:88) m=1 (cid:19) (cid:18) σH m · em = log2 [bits] (2) where σH m is the m-th Singular Value (SV) of the sensing matrix H, em is power of the signal received in the m-th orthogonal channel of H, and M is the number of SVs that are above the uncertainty level . According to Eq. ((2)), the sensing capacity of a system operating under an uncertainty level and a total transmitted power e can be increased in two ways: 1) by reducing the dispersion of the SVs, their spectrum becomes flatter, so more SVs can be measured above the noise level; and 2) by increas- m=1 em = e · ηr, where ηr ≤ 1 refers to the fraction of the transmitted power that is measured by the receiving array. ing the amount of received power er = (cid:80)M III. DESIGN AND CHARACTERIZATION OF DIGITIZED BINARY METAMATERIAL ABSORBERS The goal in this section is to present the procedure for the design of the digitized MMA. Subsection III-A begins by developing a polarization-independent unit-cell, followed by a 4-bit MMA design. Subsection III-B focuses on the design of 16 different MMA configurations. These MMAs are later used for building a diverse set of frequency dispersive codes, by being used as a coating on the surface of the reflector. A. Quasi Polarization-Independent Meander-Line Metamate- rial with Near-Unity Absorption In the work presented here, a previously developed MMA called meander-line [24] is utilized and modified to build the unit-cell. Fabrication of the meander-line MMA presented in [24] is performed using gold sputtering followed by a lift- off process [Fig. 3(a)]. Three different meander-line MMAs having resonance frequencies in the range of 70-77 GHz are designed and fabricated. As an example, simulation and measurement comparison of the magnitude of the S11, corre- sponding to the meander-line MMA resonating at 70 Ghz, is shown in Fig. 3(b). For the purpose of this work, the meander- line unit-cell is modified to design a new quasi polarization- independent MMA. The MMA unit-cell is simulated and designed using HFSS, a commercial software based on Finite Element Analysis (FEA). Master/slave periodic boundary conditions are defined for the xz- and yz-planes surrounding the unit-cell. Figure 4 shows the unit-cell structure of the presented polarization- independent meander-line MMA. A general incident plane wave with an arbitrary elevation (θ) and azimuth (φ) angles is defined to excite the unit-cell [Fig. 4(b)]. The geometric parameters of the MMA, which is designed to resonate at 71 GHz, are given by ax = ay = 1300, l = 398, w = 43 and g = 43, all in µm. Rogers RO3003 with a thickness d of 127 µm is used as the substrate, whose permittivity and dielectric loss tangent are 3 and 0.0013, respectively. The meander-line patterns on the top layer and the metal coating on the bottom layer are composed of copper with a thickness of 17 µm. The unit-cell benefits from two unique features: (i) it is almost insensitive to the polarization of the incident plane wave, i.e. for a given elevation incident angle (θ in Fig. 4(b)), is not significantly altered the reflection coefficient for different azimuth incident angles (φ in Fig. 4(b)), (ii) it has a near-unity absorption rate for wave excitations with normal incident angles, i.e. for the elevation incident angle of θ = 0, the absorption rate value, which is defined as A = 1 − S112 − S212, is near unity. It is noteworthy to mention that we only need the S11 term to calculate the absorption rate, since the bottom layer of the MMA is covered by copper [Fig. 4(b)] and there would be no transmission through the MMA array, therefore S21 = 0. A set of simulations was carried out to justify the afore- mentioned features for the proposed unit-cell. Figs. 5(a) and 5(b) illustrate the absorption rate and S11 plots for TE and TM excitations, respectively. The plots are for a fixed value of 3 (a) (b) Fig. 3. (a) Fabricated meander-line MMA on a polyimide substrate: For fabricating the MMA array, gold sputtering followed by a lift-off process is used to pattern the top layer. For the bottom layer, only a gold sputtering process is performed. Based on the simulation 20/200nm of Cr/Au was targeted, but based on the post-deposition measurement, the actual deposition is estimated to be slightly less (18/185nm Cr/Au). (b) Simulation and measurement comparison of the magnitude of the S11. (a) (b) Fig. 4. (a) Top view, and (b) 3-D view of the polarization-independent meander-line MMA unit cell. The dimensions in [µm] are: ax = ay = 1300, l = 398, w = 43, g = 43. elevation (θ = 30◦) and different azimuth (φ = 0◦, 30◦, 60◦ 4 (a) (b) Fig. 5. Absorption rate (solid lines) and S11 (dashed lines) of the quasi polarization-independent MMA at various incident angles for the (a) TE, and (b) TM incidence. and 90◦ ) incident angles. It is apparent from the absorption rate plots that, for different azimuth incident angles, there is a small shift in the frequency response (less than 0.3%), and the peak at resonance frequencies is close to unity (larger than 0.99). The absorption bandwidth, defined as the frequency range in which absorption is above half its maximum value, is 1.64 GHz. A scaling factor to the original MMA design can be applied, in order to obtain different designs resonating at different frequencies. In this regard, the original design, whose center resonating frequency is 71 GHz, is assumed to have the base dimensions with scaling factor S equal to 1. Three additional designs with center resonance frequencies of 73 GHz, 75 GHz, and 77 GHz are configured by adjusting the scaling factor to be S = 0.97, S = 0.95, and S = 0.93, respectively. For the new designs, the periodicities of the unit-cell (ax and ay) are kept the same and the scaling is applied only on the geometrical parameters of the meander-line pattern (l, w and g). Figure 6 depicts the absorption rate and S11 for the four MMA designs, which are excited by a TEM-mode plane wave. B. Design of a 4-bit Metamaterial Absorber The unit-cell design presented in subsection III-A is ex- tended to a 2 × 2 configuration, as illustrated in Fig. 7(a), to design a digitized 4-bit MMA (n = 4). Each digit aj of the Fig. 6. Absorption rate (solid lines) and S11 (dashed lines) of the MMAs with different scaling factors, for the TEM excitation: (black) S = 1.00, (blue) S = 0.97, (red) S = 0.95, and (green) S = 0.93. 4-bit MMA, is associated with one of the four basic MMA unit-cell designs. Based on the '1' or '0' values associated with each digit of the 4-bit binary code ci, the MMA may or may not absorb the impinged electromagnetic power at four different resonance frequencies, respectively. The scaling factor S for the a1, a2, a3, and a4 digits corresponds to reso- nances at f1 = 71 GHz, f2 = 73 GHz, f3 = 75 GHz, and f4 = 77 GHz, respectively. For the '0' value corresponding to a digit, the scaling factor for that digit is selected to have an out-of-band resonance (S = 1.1). Simulations show that the bandwidth for the single resonance MMAs is about 1.56 GHz, equivalent to an approximate quality factor of 46. Having a higher quality factor, more resonances can be included in the same operating frequency range, therefore, more spectral codes can be added to the imaging system. Here, based on the achieved quality factor for the single resonance MMAs, up to four resonances were used in the 70 − 77GHz frequency range. Figs. 7(b)-7(f) illustrate the absorption rate, magnitude, and phase of S11 plots for some examples of the presented 4- bit MMA. A TEM-mode plane wave excitation is applied for these examples. It is apparent from the plots that there is some frequency shift in the peak value of the absorption rate with respect to the expected resonance frequency. This is due to the mutual coupling between the adjacent meander-line elements. However, the scaling factors for each digit could be customized to compensate the unwanted mutual coupling and have the absorption rate peak at the desired frequencies. In order to further understand the behavior of the 4-bit MMA, volume loss density for one unit-cell of the array has also been studied. Figure 8 depicts the volume loss density on the top substrate layer of the 4-bit MMA unit cell with binary code number c16 ='1111', excited by a TEM-mode plane wave and an azimuth angle of φ = 0◦ [Fig. 8(b)], φ = 45◦ [Fig. 8(c)], and φ = 90◦ [Fig. 8(d)]. As illustrated in Fig. 8(b), the E-field has a component only in the x-axis and, consequently, only the meander-line elements aligned in the y-axis have absorbed the incident power. However, in Fig. 8(c), the E-field has components in both the x-axis and y- axis; hence, both meander-line elements aligned in the x-axis 5 and y-axis have absorbed the incident power. Finally, in Fig. 8(d), the E-field has a component only in the y-axis and, as a result, only the meander-line elements aligned in the x-axis have absorbed the incident power. Moreover, it can be seen from Fig. 8 that, as expected, at 71 GHz, 73 GHz, 75 GHz, and 77 GHz, only the meander-line elements associated with digits a1, a2, a3, and a4 have absorbed the incident power, respectively. Keeping the periodicity (ax and ay) constant might break the symmetry; however, doing so will not restrict the per- formance of the design for the following reasons: (a) Other geometrical parameters (l, g, and w) can be optimized to shift the resonance frequency; and (b) the change in the scaling factor (S) is very small for different designs, so it can be assumed that there is not a large discontinuity in the symmetry related to the periodicity of the unit-cells. The main reason that the periodicity has been kept constant is to be able to easily place the MMA digits side-by-side, to build a larger cell with a well-defined square shape. In this way, digitized MMA apertures could be engineered on a 2D surface without having a gap between the MMA elements. C. Characterization of the MMA, Based on Multi-Resonance Drude-Lorentz Model A semi-analytical model is used to characterize the electro- magnetic response of the meander-line MMAs. The character- ization is performed in terms of the total reflection coefficient of the meander-line MMAs. Each MMA is modeled as a bulk magneto-dielectric medium [25] and characterized by a generalized multi-resonance Drude-Lorentz model [26]. This generalized model allows any number of resonances to be accounted for with a simple summation: εr(ω) = εinf + ω2 p,e µr(ω) = µinf + ω2 p,m N(cid:88) N(cid:88) k=1 k=1 fk,e 0,k,e − ω2 − iγk,eω ω2 fk,m 0,k,m − ω2 − iγk,mω ω2 (3a) (3b) in which εinf and µinf are the static permittivity and perme- ability at infinite frequency, respectively; N is the total number of resonances; ωp,e and ωp,m are the plasma frequencies; ω0,k,e and ω0,k,m are the resonant frequencies of the kth resonance; γk,e and γk,m are the damping frequencies of the kth resonance; and fk,e and fk,m are the oscillator strengths of the kth resonance. The semi-analytical model is developed from optimizing the generalized Drude-Lorentz parameters in Eq. (3) of a three-layer magneto-dielectric media, with the goal of having a reflection coefficient equal to that computed using the commercial FEA method software HFSS. The total reflection coefficient from the stratified three-layer magneto-dielectric medium is given by [15]: Γ = Γ12 + T12Γ23T21e−j2Φtrans 1 − Γ23Γ21e−j2Φtrans (4) in which Γ is the total reflection coefficient; Γij and Tij are the reflection and transmission coefficients, respectively, linked to (a) (b) (c) (d) (e) (f) Fig. 7. Examples of absorption rate and S11 of the (a) 4-bit MMA unit cell with binary digits of (b) c3='0011', (c) c5='0101', (d) c7='0111', (e) c9='1001', (f) c15='1111'. (a) (b) (c) (d) Fig. 8. (a) TEM excitation of the 4-bit MMA unit cell with binary digit c16 ='1111'. Volume loss density on top layer of the MMA unit cell with the configuration shown in (a), for (b) φ = 0◦, (c) φ = 45◦, and (d) φ = 90◦. 6 the interface between medium i and medium j; and Φtrans is a complex value that accounts for the phase delay and amplitude attenuation associated with the wave traveling from the first interface into the second one or vice versa. The semi- analytical model is able to account for the multiple resonances of the MMAs. Moreover, it is valid for TEM, TE and TM polarizations and for a wide range of incidence angles. It is noteworthy mentioning that the effective medium ap- proximation is a valid model to be regarded for the 4-bit MMA structure, as the MMAs can be considered locally periodic. The unit-cell elements of the digitized MMA are designed to resonate at four discrete frequencies in the range of 71 GHz to 77 GHz, which is equivalent to a relatively low bandwidth (8.1%). The scaling factor for the smallest and biggest unit- cells are S = 0.93 and S = 1, respectively, which means the largest geometry variation between the adjacent cells is only 7%. Under these assumptions, it is safe to consider that the variation between the adjacent cells is smooth; therefore, local periodicity approach can be considered to establish an effective medium approximation model to predict the behavior of the digitized MMAs [27], [28]. IV. IMAGING METALLIC SCATTERERS USING A 4-BIT MMA-BASED CRA A 4-bit MMA-based CRA is used in an active mm-wave radar system to image metallic scatterers. The radar is com- posed of a CRA that is excited by an array of feeding elements. The parameters of the antenna elements in the feeding array, such as their quantity, spacing, and amplitude tapering through the surface of the reflector, are essential aspects of the imaging system. These parameters relate to the effective aperture of the feeding array, which defines the array factor in the far-field domain. The array factor is created by the physical apertures of the transmitters and receivers, and is governed by the double-convolution operation [29], [30]. It can be inferred from this operation that for having an effective aperture area, the whole area with radiating elements. Instead, by engineering the placement of the radiating elements, one can generate an effective aperture in the desired area. The feeding array parameters should be selected in a way that the effective aperture is able to cover the whole imaging region. Here, conical horn antennas are employed as the feeding elements. The transmitters and receivers are uniformly spaced in the x- axis and y-axis, respectively, forming a plus configuration, as illustrated in Fig. 1. The physical and electrical parameters of the imaging system are described in Table I. is not necessary to fill it 2BW and Rcr = λcR 2Def f Range and cross-range resolutions are the most important parameters determining imaging performance of any sensing system. Their limits for far-field imaging are, respectively, (both in [m]) [11], [31]; in Rr = c which c is the velocity of light in [m/s], BW is the bandwidth of the radar in [Hz], λc is the wavelength in [m] at the center frequency, R is the distance between the physical aperture and the target in [m], and Def f is the effective length of the physical aperture in [m] in the direction where the beam-width is to be measured in [m]. However, for near-field imaging, the above-mentioned equations provide an estimation for the range and cross-range resolution and can be used to approximate the resolution performance of the system. The measurements were collected through the 7 GHz fre- quency span with a center frequency of 73.5 GHz. The range distance between the focal plane of the CRA and the center of the imaging domain was 84 cm, and Def f in both x-axis and y-axis was 30 cm. Following the above-mentioned values, the range and cross-range resolution limits of the imaging system is calculated to be Rr = 21.4 mm and Rcr = 2.3 mm, respectively. Accordingly, the imaging domain was meshed to comply with the estimated range and cross-range resolution. An in-house numerical solver (MECA) was used for per- forming the simulations. The solver is based on physical optics and uses a modified equivalent current approximation [32] to calculate the equivalent currents based on the oblique incidence of the plane wave on the interface. The solver discretizes the interfaces into planar triangular facets, on which induced currents are assumed to have constant magnitude and linear phase variation. DESIGN PARAMETERS OF THE 4-BIT MMA-BASED IMAGING SYSTEM. TABLE I PARAMETER Frequency band No. of frequencies (Nf ) Reflector diameter (D) Focal length (f) No. of T x. No. of Rx. VALUE 70 − 77 GHz 15 30 cm 30 cm 4 4 Triangle size on the reflector Number of measurements Number of pixels (Np) Range distance (z0 T ) 5λc = 20.5 mm NT xNRxNf req = 240 80, 000 84 cm A. Sensing Capacity In order to obtain the SV distribution of the system, the sensing matrix H was calculated using MECA. Figure 9(a) shows the improved SV distribution of the CRA when com- pared to that of the TRA, and Fig. 9(b) shows how the sensing capacity of the CRA is enhanced when the Signal to Noise Ratio (SNR) is increased. It is clear from Fig. 9(b) that by adding the MMAs, the sensing capacity of the imaging system improves. For example, if the sensing channel has an SNR level equal to 40 dB, the CRA would have 97 more number of effective SVs above the noise floor, than the TRA. Ultimately, the sensing capacity of the CRA is improved, which can be inferred as an enhancement in the imaging performance. B. Radiation Pattern By plotting the fields in the imaging region, one can study the radiation patterns as a function of frequency. Figure 10 shows the y-component of the electric field magnitude of the MMA-based reflector calculated on the xy-plane, at a distance z = 85 cm from the focal plane. These radiation patterns correspond to the case where Rx2 is feeding the reflector and 7 (a) (b) Fig. 9. between the TRA and the 4-bit CRA. (a) Normalized SV distribution and (b) sensing capacity comparison they are plotted for 70.5 GHz [Fig. 10(a)], 72.5 GHz [Fig. 10(b)], 74.5 GHz [Fig. 10(c)], and 76.5 GHz [Fig. 10(d)]. As expected, the radiation pattern for each of the antennas varies as frequency changes. This behavior is due to the peculiar frequency dispersive response of the 4-bit MMAs. C. Beam Focusing To have a better understanding of the CRA performance, the Beam Focusing (BF) capability of the system is examined. The BF can be evaluated by applying a phase compensation method [12]. To accomplish this, for each measurement mode, a phase correction term is introduced to the Tx and Rx elements, to produce a phase equal to zero at the desired focusing point. Therefore, adding all the measurements results in a construc- tive summation at the focusing point. Such a study in the imaging region visualizes the resolution, side-lobes, grating lobes, and artifacts of the MMA-based reflector and helps to further understand the behavior of the imaging system. In this regard, focusing in two points ([0, 0, 85]cm and [5, 5, 85]cm) of the observation domain is studied. Figure 11 illustrates the obtained images of the BF, calculated using the phase compensation method. Figs. 11(a) and 11(b) show the BF of the receiver and transmitter array, respectively, and Fig. 11(c) shows the product of the receiver and the transmitter array BFs. It is important to note here that the norm-1 regularized imaging algorithm described below is capable of overcoming the degrading effect of the grating lobes. 8 (a) (b) (c) Fig. 11. Beam focusing at (left) [x, y, z] = [0, 0, 85]cm, and (right) [x, y, z] = [5, 5, 85]cm, by the (a) receiver array, (b) transmitter array, (c) and the product of the receiver and the transmitter array BFs. N where INp and I Nm indicate the identity matrices of sizes Np N , respectively. Since the consensus-based ADMM al- and Nm gorithm is efficiently parallelizable, it can be solved by running a MATLAB code using a GPU, reducing the computational time of the imaging process. 2) Imaging results: In this example, 2D planar PEC objects located at different ranges are considered as the targets. The objects are three vertically and three horizontally oriented rectangles, one square, and one circle. The width and height of the rectangles are equal to 12 mm and 61 mm, respectively. The side of the square, as well as the diameter of the circle, is 24 mm. The center of the imaging domain is considered to be 84 cm far from from the focal plane of the reflector and the PEC scatterers are located in four different range positions, separated by 21 mm. The reconstruction for imaging using the TRA and CRA are shown in Figs. 12(a) and 12(b), respectively. The imaging has been carried out through the ADMM algorithm. The regularization parameters of ADMM, which were optimized for each imaging case separately, are as follows: ρ = 0.1, λ = 5, and N = 48 for the TRA; and ρ = 1, λ = 100, and N = 12 for the CRA. It is evident that (a) (c) (b) (d) Fig. 10. Electric field radiation pattern of the CRA system excited with Rx2, plotted at the range of 85 cm, at (a) 70.5 GHz, (b) 72.5 GHz, (c) 74.5 GHz, and (d) 76.5 GHz. D. Image Reconstruction 1) ADMM algorithm: In order to accelerate the imaging process, the sensing matrix H in Eq. 1 may be split into N sub- matrices Hi by rows, and the vector g into N sub-vectors gi. Additionally, N different variables ui may be defined in order to solve N independent sub-problems. By adding a constraint that enforces the consensus of all solutions, the problem to optimize becomes the following: minimize 1 2 s.t. (cid:107)Hiui − gi(cid:107)2 2 + λ(cid:107)v(cid:107)1 i=1 ui = v, ∀i = 1, ..., N. (5) N(cid:80) The variable v serves as a consensus variable, imposing the agreement between all the variables ui. See for example [33]– [38]. This problem can be solved by the following iterative scheme: −1(cid:0)H∗ (cid:0)¯uk+1 + ¯sk(cid:1) , i Hi + ρI) i gi + ρ(cid:0)vk − sk i (cid:1)(cid:1) , ρN (6) (7) i + uk+1 i − vk+1, = (H∗ uk+1 i vk+1 = S λ = sk sk+1 i where si is the dual variable for the i-th constraint, ρ is the augmented parameter, and Sκ (·) is the soft thresholding operator [39] interpreted elementwise. Moreover, ¯u and ¯s are the mean of ui and si, respectively, for all i. The term −1 requires the inversion of an Np × Np matrix, (H∗ the matrix which is computationally expensive. However, inversion lemma [40] can be applied in order to perform N N × Nm N , as equation inversions of matrices of reduced size Nm (9) shows: i Hi + ρI) (8) (cid:18) (cid:19)−1 (cid:0)H∗ i Hi + ρINp (cid:1)−1 INp ρ −H∗ i ρ2 = HiH∗ ρ i + I Nm N Hi, (9) the CRA has an improved imaging, compared to that of the TRA. generated by 8-bit MMAs, reflector. tailored on the surface of the 9 Each CRA is illuminated with two orthogonal transmitting [Fig. 14(b)] and receiving [Fig. 14(a)] arrays located on the focal plane of the reflector. The electromagnetic cross-coupling between adjacent CRAs was used to perform the imaging. Given the location of the target with respect to the array, only the electromagnetic cross-coupling between CRA-l and CRA- k: (l = 1, k = 2), (l = 3, k = 4), (l = 5, k = 6), (l = 7, k = 8), (l = 1, k = 4), (l = 3, k = 6) and (l = 5, k = 8) is considered. (a) (b) Fig. 12. TRA and (b) the 4-bit CRA. Reconstructed image using ADMM iterative method for (a) the V. IMAGING A HUMAN MODEL USING AN ARRAY OF CRAS As another example of digitized MMAs, a reflector-based array made of eight CRAs is used to image an extended human-size region, as shown in Fig. 13(a). The 3D human model was projected into a 2D plane located 2 m away from the focal plane. The coding mechanism of the CRAs relies on two principles: 1) spatial coding of the electromagnetic field, generated by introducing discrete PEC scatterers on the surface of a TRA–see [13] for a detailed explanation on spatially coded CRAs–and 2) spectral coding of electromagnetic fields, (a) (b) (a) CRA array for imaging a human model; (b) Left: a single CRA Fig. 13. in offset mode. Right: the 8-bit MMAs are tailored on the surface of the CRA. The design parameters for each one of the reflectors are shown in Table II–see [13] for a detailed explanation on the meaning of these parameters. Both the vertical receiving array [Fig. 14(a)] and the horizontal transmitting array [Fig. 14(b)] of each CRA consist of 8 uniformly distributed conical horn antennas. With a similar procedure explained in Section III, an 8-bit binary code ci = a1...a8 (i ∈ {1, ..., 256}) is associated to each MMA design. Each digit of the binary code aj ∈ {0, 1} (j ∈ {1, 2, ..., 8}) is associated with a resonance frequency fj. The eight resonant frequencies are uniformly selected within the 67.375 GHz to 79.625 GHz frequency band. The reflector's surface is divided into 28 = 256 sub-surfaces and is randomly coated with the 256 different MMA configurations. In Fig. 13(b), each gray-scale color (rgb(i/2, i/2, i/2)) on the surface of the reflector represents a unique MMA configuration operating with the code ci. In Fig. 10 15, the reflection coefficient and absorption value for eight different binary codes are plotted. Based on the resonance frequencies of the 8-bit MMAs, 66.5 − 80.5 GHz is selected to be the operational band of the radar and Nf = 87 regularly sampled frequencies are used to perform the imaging. In this numerical example, the 8-bit MMAs are established in the frequency range of 66.5 GHz to 80.5 GHz, equivalent to a bandwidth of 19%, which is more than twice the bandwidth of the 4-bit MMA example presented in the previous example. Nevertheless, there have been several efforts on extending the local periodicity approach [41], which aim at expanding application scope of the method to the cases where the variations between the elements are not smooth. As it can be seen from the results in [41], in such cases the deviations and inaccuracies predominantly occur at the side lobes while the predictions of local periodicity for the main lobe remains accurate. In the case of this example, there is not much concern about the side lobes and they do not play a major role in the near-field sensing, which makes effective medium approach valid, even if the variations between the adjacent elements are not very smooth. To justify the improvement of the MMA-based CRA com- pared to that of the CRA without the MMAs, the imaging accuracy for both configurations are quantitatively measured and then compared to each other. Fig. 18(a) shows the original target mask that needs to be reconstructed. In Figs. 18(b) and 18(c), the reconstructed images above a threshold level of -16 dB are represented for the CRA and MMA-based CRA configurations, respectively. In Figs. 18(d) and 18(e), the wrong reconstructed areas (red color) and correct recon- structed areas (light blue color) are represented for the CRA and MMA-based CRA configurations. The accuracy of the imaging is calculated as the number of correct reconstruction pixels divided by the total number of pixels in the imaging domain. The wrong reconstructions could be either a point in the target domain that is detected with a reflectivity level smaller than -16 dB or a point outside the target domain that has been detected with a reflectivity level higher than -16 dB. The accuracy for the CRA configuration and MMA-based CRA configurations are calculated to be 88.74% and 92.46%, respectively, which shows the effectiveness of the established MMAs in the imaging system. Finally, this improvement will be even more evident for a target that is less sparse than that used in Fig. 17 under a Total Variation Norm metric. Specifically, as shown in Fig. 16, for a SNR=50 dB, the MMA-based CRA has about 30% more singular values above the noise level than the simple CRA; so considering a typical value on the minimum number of measurements "m" needed to reconstruct an "s"-sparse signal to be m = 4s, then the MMA-based CRA should be able to reconstruct signals of sparsity sCRA/MM A = 1000, while the CRA will only be able to reconstruct signals with sparsity sCRA = 750. (a) (b) Fig. 14. of transmitting horn antennas feeding the CRA-1. (a) Array of receiving horn antennas feeding the CRA-2 (b) array DESIGN PARAMETERS FOR A SINGLE CRA. TABLE II PARAMETER Frequency band VALUE 66.5 − 80.5 GHz No. of frequencies (Nf ) No. of T x. No. of Rx. Aperture size (Dx = Dy) Focal length (f) Offset height (hof f ) Size of the random facets Max. Distortion of facets 87 8 8 50 cm 50 cm 35 cm 4λ 3◦ Figure 16(a) shows the improved SV distribution of the single MMA-based CRA, when compared to that of the CRA without MMAs and the TRA. Also, Fig. 16(b) shows how the sensing capacity of the CRA is enhanced for different SNR levels. The image reconstruction is performed using the MATLAB toolbox NESTA [42] imposing norm-1 regulariza- tion priors. The imaging result of the CRA with and without the 8-bit MMAs are plotted in Fig. 17. Fig. 15. Magnitude of reflection (S11) and absorption (1 − S112) of the MMA array for different binary codes: (a) c16 = 00010000, (b) c32 = 00100100, (c) c73 = 01001001, (d) c85 = 01010101, (e) c107 = 01101011, (f) c231 = 11100111, (g) c223 = 11011111, and (h) c255 = 11111111. 11 Fig. 18. (a) Original target. Reconstructed image above the -16 dB threshold for the (b) CRA and (c) MMA-based CRA configurations; Image recon- struction error (red color) for the (d) CRA and (e) MMA-based CRA configurations. VI. CONCLUSION A new radar system for high sensing capacity imaging applications, using a metamaterial-based CRA is presented. By adopting the frequency dispersive MMAs coated on the surface of the reflector antenna, the new imaging system is able to increase the sensing capacity and create highly uncorrelated spatial and spectral patterns in the near-field of the system. Polarization-independent meander-line metamate- rials are utilized as the unit-cells for designing 16 different configurations of a 4-bit MMA. Using a norm-1 regularized iterative algorithm based on ADMM, the performance of the system is evaluated in an active mm-wave imaging application. The system is characterized in terms of its radiation pattern and beam focusing capabilities. Promising results are achieved for the sensing capacity and image reconstruction quality of the 4-bit MMA-based reflector antenna, compared with that of the TRA. As a second example, an array of CRAs equipped by 8-bit MMAs was utilized to image an extended human-size region. One of the main features of the proposed CRA-based system is that it can be easily reconfigured to operate in the far-field, thus enabling standoff detection of security threats. Although this paper only considered the metamaterials for tailoring the surface of the reflector, the use of all-dielectric materials or other MMA configurations on the surface of the reflector opens up several exciting areas of further research. In particular, utilizing an MMA with a higher quality factor allows one to fit more resonances in the frequency range of the radar. This is expected to result in a higher dispersion in the frequency response of the MMAs, which ultimately increases the sensing capacity and the imaging capabilities of the system even further. ACKNOWLEDGMENT This work is funded by the U.S. Department of Homeland Security, Award No. 2013-ST-061-ED0001. REFERENCES [1] D. Sheen, D. McMakin, and T. Hall, "Three-dimensional millimeter- wave imaging for concealed weapon detection," Microwave Theory and Techniques, IEEE Transactions on, vol. 49, no. 9, pp. 1581–1592, 2001. (a) (b) Fig. 16. Comparison of (a) the normalized SV distribution and (b) the sensing capacity of a single CRA and TRA. (a) (b) Fig. 17. Reconstructed image using iterative compressive sensing algorithm (NESTA) for the (a) CRA and (b) MMA-based CRA arrays. 12 [23] C. Shannon, "A mathematical theory of communication, bell system technical journal 27: 379-423 and 623–656," Mathematical Reviews (MathSciNet): MR10, 133e, 1948. [24] C. Watts, "Metamaterials and their applications towards novel imaging technologies," Ph.D. dissertation, PhD Thesis, 2015. [25] C. M. Watts, X. Liu, and W. J. Padilla, "Metamaterial electromagnetic wave absorbers," Advanced materials, vol. 24, no. 23, pp. OP98–OP120, 2012. [26] A. Ugawa, A. G. Rinzler, and D. Tanner, "Far-infrared gaps in single- wall carbon nanotubes," Physical Review B, vol. 60, no. 16, p. R11305, 1999. [27] J. A. Encinar, L. S. Datashvili, J. A. Zornoza, M. Arrebola, M. Sierra- Castaner, J. L. Besada-Sanmartin, H. Baier, and H. Legay, "Dual- polarization dual-coverage reflectarray for space applications," IEEE Transactions on Antennas and Propagation, vol. 54, no. 10, pp. 2827– 2837, 2006. [28] J. A. Encinar, "Recent advances in reflectarray antennas," in Antennas and Propagation (EuCAP), 2010 Proceedings of the Fourth European Conference on. IEEE, 2010, pp. 1–6. [29] G. Lockwood and F. S. Foster, "Optimizing sparse two-dimensional transducer arrays using an effective aperture approach," in Ultrasonics Symposium, 1994. Proceedings., 1994 IEEE, vol. 3. IEEE, 1994, pp. 1497–1501. [30] S. S. Ahmed, A. Schiess, and L.-P. Schmidt, "Near field mm-wave imaging with multistatic sparse 2d-arrays," in 2009 European Radar Conference (EuRAD), 2009. [31] E. Byron, "Radar principles, technology, applications," Englewood [32] J. Meana, J. Cliffs, New Jersey: PTR Prentice hall, pp. 618–627, 1993. ´A. Mart´ınez-Lorenzo, F. Las-Heras, and C. Rappaport, "Wave scattering by dielectric and lossy materials using the modified equivalent current approximation (meca)," Antennas and Propagation, IEEE Transactions on, vol. 58, no. 11, pp. 3757–3761, 2010. [33] J. Heredia Juesas, G. Allan, A. Molaei, L. Tirado, W. Blackwell, and J. A. Martinez Lorenzo, "Consensus-based imaging using admm for a compressive reflector antenna," in Antennas and Propagation Symposium, July 2015. [34] M. H. DeGroot, "Reaching a consensus," Journal of the American Statistical Association, vol. 69, no. 345, pp. 118–121, 1974. [35] T. Erseghe, D. Zennaro, E. Dall'Anese, and L. Vangelista, "Fast consen- sus by the alternating direction multipliers method," Signal Processing, IEEE Transactions on, vol. 59, no. 11, pp. 5523–5537, November 2011. [36] J. F. Mota, J. Xavier, P. M. Aguiar, and M. Puschel, "Distributed basis pursuit," Signal Processing, IEEE Transactions on, vol. 60, no. 4, pp. 1942–1956, April 2012. [37] J. F. Mota, J. M. Xavier, P. M. Aguiar, and M. Puschel, "D-admm: A communication-efficient distributed algorithm for separable optimiza- tion," Signal Processing, IEEE Transactions on, vol. 61, no. 10, pp. 2718–2723, May 2013. [38] P. A. Forero, A. Cano, and G. B. Giannakis, "Consensus-based dis- tributed support vector machines," The Journal of Machine Learning Research, vol. 11, pp. 1663–1707, 2010. [39] K. Bredies and D. A. Lorenz, "Linear convergence of iterative soft- thresholding," Journal of Fourier Analysis and Applications, vol. 14, no. 5-6, pp. 813–837, October 2008. [40] M. A. Woodbury, "Inverting modified matrices," Memorandum report, vol. 42, p. 106, 1950. [41] M. Zhou, S. B. Sørensen, E. Jørgensen, P. Meincke, O. S. Kim, and O. Breinbjerg, "Analysis of printed reflectarrays using extended local periodicity," in Antennas and Propagation (EUCAP), Proceedings of the 5th European Conference on. IEEE, 2011, pp. 1408–1412. [42] S. Becker, J. Bobin, and E. J. Candes, "Nesta: a fast and accurate first- order method for sparse recovery," SIAM Journal on Imaging Sciences, vol. 4, no. 1, pp. 1–39, 2011. [2] J. A. Martinez-Lorenzo, F. Quivira, and C. M. Rappaport, "Sar imaging of suicide bombers wearing concealed explosive threats," Progress In Electromagnetics Research, vol. 125, pp. 255–272, 2012. [3] S. S. Ahmed, A. Schiessl, and L.-P. Schmidt, "Novel fully electronic active real-time millimeter-wave imaging system based on a planar multistatic sparse array," in Microwave Symposium Digest (MTT), 2011 IEEE MTT-S International. IEEE, 2011, pp. 1–4. [4] B. N. Lyons, E. Entchev, and M. K. Crowley, "Reflect-array based mm- Interna- wave people screening system," in SPIE Security+ Defence. tional Society for Optics and Photonics, 2013, pp. 890 002–890 002. [5] J. A. Martinez-Lorenzo, Y. Rodriguez-Vaqueiro, C. M. Rappaport, O. R. Lopez, and A. G. Pino, "A compressed sensing approach for detection of explosive threats at standoff distances using a passive array of scatters," in Homeland Security (HST), 2012 IEEE Conference on Technologies for. IEEE, 2012, pp. 134–139. [6] A. Jam and K. Sarabandi, "A horizontally polarized beam-steerable antenna for sub-millimeter-wave polarimetrie imaging and collision avoidance radars," in Antennas and Propagation (APSURSI), 2016 IEEE International Symposium on. IEEE, 2016, pp. 789–790. [7] S. Kharkovsky and R. Zoughi, "Microwave and millimeter wave non- destructive testing and evaluation-overview and recent advances," IEEE Instrumentation & Measurement Magazine, vol. 10, no. 2, pp. 26–38, 2007. [8] D. Ferris and N. C. Currie, "Microwave and millimeter-wave systems for wall penetration," in Proc. sPIE, vol. 3375, 1998, pp. 269–279. [9] M. Dehmollaian and K. Sarabandi, "Refocusing through building walls using synthetic aperture radar," IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 6, pp. 1589–1599, 2008. [10] M. Dehmollaian, M. Thiel, and K. Sarabandi, "Through-the-wall imag- ing using differential sar," IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 5, pp. 1289–1296, 2009. [11] T. Sleasman, M. Boyarsk, M. F. Imani, J. N. Gollub, and D. R. Smith, "Design considerations for a dynamic metamaterial aperture for computational imaging at microwave frequencies," JOSA B, vol. 33, no. 6, pp. 1098–1111, 2016. [12] S. S. Ahmed, A. Schiessl, and L.-P. Schmidt, "Multistatic mm-wave imaging with planar 2d-arrays," in 2009 German Microwave Conference. IEEE, 2009, pp. 1–4. [13] J. Martinez-Lorenzo, J. Heredia Juesas, and W. Blackwell, "A single- transceiver compressive reflector antenna for high-sensing-capacity imaging," Antennas and Wireless Propagation Letters, IEEE, vol. PP, no. 99, pp. 1–1, 2015. [14] J. Martinez Lorenzo, J. Juesas, and W. Blackwell, "Single-transceiver compressive antenna for high-capacity sensing and imaging applica- tions," in Antennas and Propagation (EuCAP), 2015 9th European Conference on, April 2015, pp. 1–3. [15] A. Molaei, J. H. Juesas, W. Blackwell, and J. A. M. Lorenzo, "Inter- ferometric sounding using a metamaterial-based compressive reflector antenna," IEEE Transactions on Antennas and Propagation, 2018. [16] J. H. Juesas, G. Allan, A. Molaei, L. Tirado, W. Blackwell, and J. A. M. Lorenzo, "Consensus-based imaging using admm for a compressive reflector antenna," in 2015 IEEE International Symposium on Antennas and Propagation & USNC/URSI National Radio Science Meeting. IEEE, 2015, pp. 1304–1305. [17] A. Molaei, G. Allan, J. Heredia, W. Blackwell, and J. Martinez-Lorenzo, "Interferometric sounding using a compressive reflector antenna," in 2016 10th European Conference on Antennas and Propagation (Eu- CAP). IEEE, 2016, pp. 1–4. [18] A. Molaei, J. H. Juesas, G. Allan, and J. Martinez-Lorenzo, "Active imaging using a metamaterial-based compressive reflector antenna," in Antennas and Propagation (APSURSI), 2016 IEEE International Symposium on. IEEE, 2016, pp. 1933–1934. [20] A. Molaei, G. Ghazi, [19] A. Molaei, J. Heredia-Juesas, and J. Martinez-Lorenzo, "A 2-bit and 3-bit metamaterial absorber-based compressive reflector antenna for high sensing capacity imaging," in Technologies for Homeland Security (HST), 2017 IEEE International Symposium on. IEEE, 2017, pp. 1–6. J. Heredia-Juesas, H. Gomez-Sousa, and J. Martinez-Lorenzo, "High capacity imaging using an array of com- pressive reflector antennas," in Antennas and Propagation (EUCAP), 2017 11th European Conference on. IEEE, 2017, pp. 1731–1734. [21] J. Heredia-Juesas, A. Molaei, L. Tirado, W. Blackwell, and J. ´A. Mart´ınez-Lorenzo, "Norm-1 regularized consensus-based admm for imaging with a compressive antenna," IEEE Antennas and Wireless Propagation Letters, vol. 16, pp. 2362–2365, 2017. [22] A. Molaei, J. H. Juesas, and J. A. M. Lorenzo, "Compressive reflector InTech, antenna phased array," in Antenna Arrays and Beam-formation. 2017.
1902.07371	1	1902	2019-02-20T01:59:03	Low-threshold Stimulated Emission of Hybrid Perovskites at Room Temperature through Defect-Mediated Bound Excitons	[ "physics.app-ph" ]	Excitonic stimulated emission provides a promising mechanism and route to achieve low-threshold semiconductor lasers for micro-nano optoelectronic integrations. However, excitonic stimulated emission from quantum structure-free semiconductors has rarely been realised at room temperature due to the phase transition between excitonic and electron-hole plasma states. Herein, we show that through trap-state and band-edge engineering, bound exciton states can be stabilised within the hybrid lead bromide perovskite. Under modest pumping conditions, these states enable stimulated emission behaviour that exhibits a low threshold carrier density of only 1.6*1017 cm-3, as well as a high peak gain coefficient of ~1300 cm-1. This is the first time that bound exciton stimulated emission has been realised at room temperature from a quantum structure-free semiconductor. Not only does this open up new research horizons for perovskite materials, but also it has important implications for semiconductor excitonic physics and the development of next-generation optoelectronic applications.	physics.app-ph	physics	Low-threshold Stimulated Emission of Hybrid Perovskites at Room Temperature through Defect-Mediated Bound Excitons Jiangjian Shi,1 Yiming Li,1,2 Jionghua Wu,1,2 Huijue Wu,1 Yanhong Luo,1,2 Dongmei Li,1,2 Jacek J. Jasieniak,3* Qingbo Meng1,4,* 1. Key Laboratory for Renewable Energy, Chinese Academy of Sciences, Beijing Key Laboratory for New Energy Materials and Devices, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, P. R. China. 2. Department of Physics Science, University of Chinese Academy of Sciences, Beijing 100049, P. R. China 3. ARC Centre of Excellence in Exciton Science, Department of Materials Science and Engineering, Monash University, Clayton, Victoria 3800, Australia. 4. Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing 100049, China Corresponding authors: [email protected], [email protected] Abstract: Excitonic stimulated emission provides a promising mechanism and route to achieve low-threshold semiconductor lasers for micro-nano optoelectronic integrations. However, excitonic stimulated emission from quantum structure-free semiconductors has rarely been realised at room temperature due to the phase transition between excitonic and electron-hole plasma states. Herein, we show that through trap-state and band-edge engineering, bound exciton states can be stabilised within the hybrid lead bromide perovskite. Under modest pumping conditions, these states enable stimulated emission behaviour that exhibits a low threshold carrier density of only 1.6×1017 cm-3, as well as a high peak gain coefficient of ~1300 cm-1. This is the first time that bound exciton stimulated emission has been realised at room temperature from a quantum structure-free semiconductor. Not only does this open up new research horizons for perovskite materials, but also it has important implications for semiconductor excitonic physics and the development of next-generation optoelectronic applications. Keywords: Hybrid perovskite, Excitonic stimulated emission, Bound exciton, Trap state Since the discovery of the ruby laser in 1960, lasers have literally changed the world by enabling global communication through satellites and the internet, advancing manufacturing, and unravelling the wonders of almost all the scientific worlds. 1-2 As optoelectronic components continue to reduce in size and increase in speed, the need for integrated microscale laser systems that operate at room temperature will follow.3-5 Semiconductor thin- film lasers are a promising photonic source for such micro/nano optoelectronic applications, owing to their tunable wavelength, compact size and ease of optical or electronic coupling.6-8 For satisfying on-chip integration, realising laser devices with low energy consumption is critical. Numerous materials, device structures and physical mechanisms have been investigated to achieve this goal, mainly targeting reduced lasing threshold and increased optical gain.9-17 Lasers operate by amplifying stimulated emission (STE) arising from an optically or electrically pumped semiconductor gain medium that is under population inversion.18 Compared to the conventional degenerate electron-hole plasma (EHP) mechanism, the nature of exciton states and their photon emission processes provide more possibilities to realize population inversion, which can be exploited to significantly reduce the STE thresholds, especially at low temperatures.18 However, except for in quantum structures,11-13 room- temperature (RT) STE of exciton states within inorganic semiconductors has rarely been ) of such materials, for which (i) realised.18 This is attributed to the low Rydberg energy (Ry the thermal energy at RT readily overcomes the necessary attractive exciton Coulomb interactions and (ii) the high carrier pumping required for STE usually induces the excitonic- EHP phase transition.18 Owing to the high energy bandgap (Eg) and Ry * (~60 meV), ZnO is a standout case that exhibits violet STE at RT with a high threshold of >1018 cm-3.10 However, the high Eg of ZnO makes it difficult to be optically or electrically pumped and controversy still exists on whether its STE arises from an excitonic process.18-19 As such, developing new excitonic material systems that can achieve low-threshold RT STE across the visible spectrum from bulk semiconductor continues to be a major challenge. Hybrid lead halide perovskites have emerged as lucrative optoelectronic materials for light- emitting and photovoltaic applications.6,20 They have also been used as optically pumped lasers, exploiting their EHP states to achieve modest STE thresholds of ~1018 cm-3.8, 21-26 The population inversion of perovskite EHP originates from the carrier occupation of continuum bands. Achieving EHP within perovskites is a double-edged sword. On the one hand, the high effective density-of-states close to the band-edge of perovskite (~1018 cm-3) contribute to its high light absorption coefficient (104 ～ 105 cm-1); however, they also necessitate large pumping thresholds of >1018 cm-3 for realising a population inversion.21 Exploiting the excitonic characteristics of hybrid perovskites presents a major opportunity towards reducing these thresholds. In this article, we describe how bound exciton complex (BE) states that form by the Coulomb binding between an exciton and a trap centre can be accessed within solution-processed CH3NH3PbBr3 (MAPbBr3) polycrystalline films through precise engineering of the perovskite crystallization, traps and exciton state dynamics. It is shown that these BE states enable RT STE with a nearly 10-fold reduction in threshold fluence compared to previous reports harnessing EHP inversion. This is the first work to report BE STE at RT from a cavity-free bulk semiconductor. Moreover, it presents a lucrative excitonic mechanism and route for optoelectronic applications, helping to broaden the research and application horizons of perovskite materials. (a) Energy Continuum * y R . m r o f . s s d i E B b E d n u o b d n u o b n u Tail state Free exciton Bound exciton Eg Trap state ) D E ( A E - g E (b) ) g E ( l e v e l y g r e n e i m r e F (c) ) 3 - m c ( y t i s n e D 1.0 0.8 0.6 0.4 0.2 0.0 1018 1017 1016 1015 Threshold this work Eg i n o s r e v n i n o i t l a u p o P P H E f i o e m g e r BE nth 3 - m c 6 1 0 1 × 2 . 7 ~ ) I ( M n Efn Efp NT 0.5NT 3 - m c 7 1 0 1 × 5 . 4 ~ ) r B ( M n Free carrier Free exciton Bound exciton (Ry=40 meV) Bound exciton (Ry=15 meV) 1018 1019 Momentum (K) 1014 1016 1017 Injected carrier density (cm-3) Figure 1. Charge occupation and population inversion within different energy states of the perovskite. (a) Schematic diagram of energy states (including continuum band, free and bound excitons, and trap state) and internal charge transfer processes within a perovskite semiconductor. (b) Theoretical calculations of the quasi-Fermi energy level (electron: Efn, hole: Efp) of the perovskite under carrier injections. The EHP shows a population inversion when the Δμ (i.e. Efn-Efp) is larger than the energy bandgap (Eg), with a threshold of ~1.7×1018 cm-3. (c) Theoretical calculation for the densities of free carriers, free excitons and bound excitons as determined through the Saha-Langmuir equation and the thermal equilibrium between free and bound excitons. The density of trap centers (NT) that could bind free excitons was experimentally determined to be ~2×1017 cm-3. Moreover, experimental * at ~40 meV and the binding energy of an exciton bound to a trap center values of the Ry BE) at ~100 meV were also used. When the occupation probability of trap centers is higher (Eb BE) is than 1/2, a population inversion of the BE state will occur. The inversion threshold (nth determined to be ~2.7×1017 cm-3. The Mott densities (nM) of iodide and bromide perovskites are illustrated through dashed green lines crossing (b) and (c). For comparison, the STE thresholds of perovskites reported in the literate (open red squares)7,8,22-26 and those obtained here (solid circles) are included in (b). The BE STE is depicted in red, while EHP in blue. Theoretical estimation of perovskite STE: EHP vs. BE While STE has been confirmed in perovskites, its mechanism has not been clearly understood, with most reported cases showing high thresholds of >1018 cm-3.7-8, 20-28 As such, compared to conventional semiconductors, the advantages of these material system in terms of STE performance have not been fully realised. Towards addressing this, we firstly give a theoretical description of the energy states and charge-carrier properties in perovskites, and then estimate their population inversion limitations for EHP and excitonic processes. As depicted from photoexcitation or electric injection populate within a variety of energy states, including the continuum band (i.e. conduction and valence bands) and its tail states, free exciton (FE), BE and trap states.18 The balance between the continuum band carriers and excitons is described by the Saha-Langmuir equation.29 Meanwhile, the equilibrium between the FE and BE states can be considered analogous to that of detrapped and trapped charges at trap centres, respectively (Supplementary Note 1).30 in Figure 1(a), charge-carriers generated within a semiconductor The quasi-Fermi energy level splitting (Δμ) of MAPbBr3 is calculated from the Fermi-Dirac distribution to estimate the population inversion of EHP (Figure 1(b)). Under higher carrier density, Δμ becomes progressively larger. When it exceeds the bandgap energy of MAPbBr3 (Eg~2.4 eV), a population inversion is achieved. STE can be realised under these conditions within a suitable device configuration. The threshold of the EHP STE for MAPbBr3 is found to be ~1.7×1018 cm-3, which is similar to other theoretical predictions for MAPbI3.21 As is shown in the top graph of Figure 1(b), this value is comparable to both previously reported STE values and our experiments without energy-state engineering (i.e. ~1018 cm-3).8,23-26 Notably, the EHP mechanism underlying these experimental STE observations can be readily identified from pump fluence-dependent emission characteristics, which show a progressive shift of the STE peak due to bandgap renormalization (red-shift) or carrier filling of continuum bands (blue-shift).7,18,22,23,28 To evaluate the population inversion behaviour within excitonic systems, we have determined the occupancy density within the continuum band, as well as FE and BE states, under thermal equilibrium (Figure 1(c), Supplementary Note 1). In these systems, when excitons occupy more than half of the trap centres, a population inversion of the BE state can occur under.11 In our previous work we proved the existence of the BE state within MAPbBr3.31 Additionally, BE ~ 100 meV), we derived experimental binding energies of the FE (Ry as well as the trap densities responsible for the BE (~2×1017 cm-3). Using these values, we can determine the threshold for BE STE at ~2.7×1017 cm-3. This value is more than 6 times lower than that for STE arising from EHP. It confirms that excitonic processes are more lucrative towards achieving STE than EHP.18 Moreover, this threshold value of MAPbBr3 is lower than its theoretically calculated Mott density (~4.5×1017 cm-3),18 while the opposite is found for MAPbI3 (nM ~ 7.2×1016 cm-3). Noting that the Mott density provides an estimation of the upper limit for the stable regime of the excitonic phase, these values confirm that the bromide perovskite system is a promising and uncommon candidate to realise low-threshold BE STE. ~ 40 meV) and BE (Eb Stabilisation of perovskite exciton states Until recently, BEs within bulk perovskites have not been widely observed or studied.31 This has arisen because these exciton states can be easily screened by local electric fields stemming from shallow defect (ST) or tail state induced band-edge fluctuation (BF).18 The BE is formed by the Coulomb attraction between a trap centre and an electron-hole pair (i.e. exciton),18 as schematically shown in Figure 2(a). According to Haynes rule,18, 32 an exciton that is bound to a neutral acceptor (i.e. A0X) is (i) more stable than a free exciton and (ii) BE when it is associated with a trap that has a larger ionisation energy. exhibits a higher Eb Thus, to stabilise exciton states suitable for realising novel optoelectronic properties, such as perovskite RT STE, shallow defects need to be eliminated (to suppress charge screening), while deeper acceptor centres need to be introduced. Figure 2. Energy state engineering and exciton stabilisation of MAPbBr3. (a) Schematic diagram of the bound exciton complex, which is formed by the Coulomb attraction between a trap centre and an exciton. (b) Trap state distribution of the MAPbBr3 film deposited by an anti-solvent method, where the time of chlorobenzene dropping is adjusted. (c) Temperature- dependent steady-state PL of the MAPbBr3 with different deposition conditions (6 and w/o, excited at 450 nm, 5 nJ cm-2). The emission peaks are systematically fitted and distinguished, as indicated by the black dashed lines (FE: free exciton, BE: bound exciton, ST: shallow trap, DTs: deep traps, BF: band fluctuation). For the 6-MAPbBr3, FE and BE emission bands are observed across the entire temperature region. (d) Peaks fittings of emission spectra at 70K and 300K, respectively. MAPbBr3 is a ternary semiconductor, whose trap characteristics are largely dictated through self-doping mechanisms, e.g. vacancies and interstitials.33 These can be modified through precisely tailored processing conditions. In this work, we have adopted an anti-solvent (chlorobenzene), solution-processed method to adjust the crystallization and growth dynamics of MAPbBr3 thin films.34 For simplicity, we have adopted a naming convention for light absorption and transient PL measurements the samples that relates the time at which the anti-solvent was cast onto the samples during spin-coating of the precursor solution (Supplementary Figure S2), i.e. the sample derived from chlorobenzene dropped after 6 seconds is denoted as 6-MAPbBr3. Samples that did not have an anti-solvent dropped are denoted as w/o-MAPbBr3. Variation in the processing conditions resulted in modified optical and structural characteristics, as verified optically, and through x-ray diffraction, (see Supplementary Figure S3-S7). These imply that the crystallization process has indeed been modified. Moreover, their trap state properties have been characterized using thermal admittance spectroscopy within an FTO/compact TiO2/MAPbBr3/Spiro-OMeTAD/Au device architecture at temperatures ranging from 150K to 330K (Supplementary Figure S8-S10).30 All the samples possess an intermediate-high-frequency capacitance response, which arises from shallow trap states. However, the energy level of this shallow trap state is dependent on the sample (Figure 2(b)). For the w/o-MAPbBr3, the trap state is located at ~0.05 eV above the valence band and possesses a broad Gaussian distribution of ~0.05 eV. Meanwhile, for 6- MAPbBr3, this shallow defect is observed at ~0.18 eV and has a much narrower distribution of ~0.03 eV. These energy levels and their varying densities (Supplementary Figure 11) imply that the 6-MAPbBr3 exhibits the smallest band-edge fluctuation. In addition to these shallow traps, the low-frequency capacitance response indicates that deeper trap states are also presented in the 5- and 6-MAPbBr3 samples at ~0.25 eV and 0.38 eV, respectively. As discussed above, these deep traps are likely to be favourable towards stabilising BE states. To further probe their impact on the photo emission properties of the samples, temperature- dependent photoluminescence (PL) measurements were carried out from 10K to 300 K. The results of the 6- and w/o-MAPbBr3 samples are presented in Figure 2(c), with the additional samples presented in Supplementary Figure S12 and S13. To distinguish the fine structure of the energy states, the PL spectra across all the samples were de-convoluted (Figure 2(d), Supplementary Figure S14). The emission position, broadening and intensity of each contribution were mapped across the studied temperature range (Supplementary Figures S15- S17), which helps to distinguish the mechanism for each emission band. For each sample, up to five emission bands could be observed at temperatures less than 100K. Based on the above energy-state discussions, these have been assigned to FE, BE (or BF), ST and deep defects (DTs). As expected, the emission from DTs disappeared at temperatures higher than ~130K due to the orthorhombic-tetragonal phase transition.31 The FE and BE emission positions were independent of the pump intensity, while the ST and DT emissions showed characteristic state-filling behaviour (Supplementary Figure S15). For the w/o-MAPbBr3, its FE emission disappeared above 200K, while the BF and ST emission remained across the studied temperature range. This result shows that the BF emission is correlated to a high density of STs in the w/o-MAPbBr3, which could screen the excitonic nature of this material. Meanwhile, the deeper ST and DT in the 6-MAPbBr3 help to stabilise the exciton states. These observations are consistent with the above considerations of excitonic phenomena. (a) ) s p ( y a e D l 103 102 101 (b) ) s p ( y a e D l 100 103 102 101 100 (d) ) . m r o n ( A Δ − 1.0 0.5 0.0 (c) 103 ) s p ( y a e D l 102 101 w/o-MAPbBr3 2.5 μJ cm-2 10 ) D O m ( A Δ 6-MAPbBr3 0.5 μJ cm-2 -10 PB 6-MAPbBr3 2.5 μJ cm-2 40 ) D O m ( A Δ -40 100 2.6 PA 2.2 2.3 2.4 2.5 Probe Energy (eV) 6-MAPbBr3 2.5 μJ cm-2 2.2 2.3 2.5 Probe Energy (eV) 2.4 2 ps 500 ps 2500 ps 2 [email protected] μJ cm-2 2.2 2.3 Probe Energy (eV) 2.4 2.5 2.6 2.6 Figure 3. Femtosecond transient absorption of the MAPbBr3. Two-dimensional pseudocolor plots (ΔA) of transient absorption for the MAPbBr3 films at 300K as functions of probe energy and time delay under different pump fluences (a. w/o: 2.5 μJ cm-2, b. 6: 0.5 μJ cm-2, (c) 6: 2.5 μJ cm-2, pumped at 2.92 eV). The photobleaching (PB) peaks are illustrated by dashed black lines. (d) Photo-induced absorption spectra of the 6-MAPbBr3 at different time delays and pump fluences. Gaussian-shaped dashed black lines are guides for the eye to illustrate the PB bands composed of FE and BE signals. Femtosecond transient absorption (TA) has been used to further probe the temporal dynamics of the exciton states (Figure 3). When the MAPbBr3 samples were pumped at 2.92 eV (425 nm), distinct photobleaching (PB) and photoinduced absorption (PA) bands could be observed. The PB peak of the w/o-MAPbBr3 at ~2.4 eV exhibited a broad, high-energy distribution, with a progressive bathochromic-shift in time (Figure 3(a)). This should be a clear signature of carrier occupation and relaxation within the continuum band and bandedge states. Meanwhile, for the 6-MAPbBr3, the spectral position of the main PB peak at ~2.35 eV was independent of the time delay and pump fluence (Figure 3(b) and (c)). These trends are characteristic signatures of excitonic processes,11 which in this case can be ascribed to FEs. Together with the PL and TA, we can reasonably conclude that excitons and not free carriers dominate the emission properties of the 6-MAPbBr3. We further note that for the 6-MAPbBr3, a small energy difference of ~28 meV exists between the FE PB and PL peaks. This energy difference is similar to the longitudinal optical phonon (LO) energy of MAPbBr3 (Supplementary Figure 16), implying an FE-LO process within the FE emission. At a higher pump fluence of 2.5 μJ cm-2, a PB contribution centred at ~2.25 eV becomes observable, consistent with the BE emission in the PL spectra. It has a rise time of ~1 ps and a decay lifetime of ~60 ps. This behaviour is consistent with an ultrafast exciton trapping process and subsequent fast spontaneous emission relaxation of the formed BE excited state. The BE state may simultaneously form when the FE relaxes within its Rydberg states. The bleaching characteristics of the identified BE state indicate that it possesses a much lower density-of-states than the FE contribution at ~2.35 eV. More detailed analysis on the high- energy PA35 (Supplementary Figure 18-20) illustrate that no bandgap renormalization occurs in our studied pump fluence range. Therefore, we can confidently conclude that the FE and BE states of the MAPbBr3 have been stabilised by energy state engineering. Perovskite excitonic STE To determine how the BE states influence the intrinsic STE behaviour of the MAPbBr3, we have performed steady-state PL measurements (reflection configuration) with varying pump fluence under cavity free conditions. For the 6-MAPbBr3, the FE and BE emission contributions can be observed across the entire studied fluence range without any observable change in the peak position. At ~2 μJ cm-2, a much narrower emission peak centred at ~2.245 eV (552 nm) appears and becomes more prominent at higher fluences (Figure 4(a)). These are clear signatures of optically amplified spontaneous emission (ASE).6 According to the pump-dependent emission intensity, the threshold fluence of the ASE for the 6-MAPbBr3 is ~1.7 μJ cm-2 (Figure 4(c) and Supplementary Figure 21). The corresponding threshold carrier density is calculated to be ~1.6×1017 cm-3. This value is almost one order of magnitude lower than previous results and the theoretical EHP threshold.21 For reference, the 5-MAPbBr3 also exhibited ASE with a similar spectral feature, albeit with a slightly higher threshold of ~2.4×1017 cm-3. As no signature of bandgap renormalization or excitonic-EHP transition was observed from the femtosecond TA, and the peak position overlays the BE emission, this optical amplification is attributed to BE STE. Notably, the ASE signals possessed no polarisation dependence, which is consistent with the samples being cavity-free and the excitons being randomly oriented within the polycrystalline film. (a) ) . u . a ( y t i s n e n t I L P 2.322 eV 2.245 eV FE BE (b) 6-MAPbBr3 11.1 μJ cm-2 7.2 μJ cm-2 ) . u . a ( y t i s n e n t I L P 3.8 μJ cm-2 2.1 μJ cm-2 0.8 μJ cm-2 ~ 5 nJ cm-2 2.288 eV 2.263 eV BF ST w/o-MAPbBr3 60.0 μJ cm-2 31.2 μJ cm-2 Total intensity FE BE-SPE Pth~1.7 μJ cm-2 6-MAPbBr3 (c) ) . u . a ( y t i s n e t n I L P 6 5 4 3 2 1 7 Pump fluence (μJ cm-2) Total intensity BF ST-SPE w/o-MAPbBr3 21.3 μJ cm-2 0 (d) 14.1 μJ cm-2 6.6 μJ cm-2 ~ 5 nJ cm-2 ) . u . a ( y t i s n e t n I L P 2.4 2.3 2.2 Energy (eV) 2.1 2.4 2.1 0 ) m n ( M H W F 35 30 25 20 15 10 25 ) 20 m n ( M H W F 15 10 2.3 2.2 Energy (eV) 10 20 30 40 50 60 70 Pump fluence (μJ cm-2) Figure 4. Photo-pumped stimulated emission behaviour of MAPbBr3 films. (a and b) Time-integrated PL emission spectra of MAPbBr3 films photo-pumped by 450 nm, 35 fs and 1 kHz pulses (a. 6-MAPbBr3, b. w/o-MAPbBr3). All the spectra are well fitted by considering the FE and BE spontaneous emission (BF and ST for the w/o-sample) and BE stimulated emission (ST for the w/o-sample). (c and d) FWHM and intensity of the emission peaks as a function of the pump fluence for (c) 6-MAPbBr3 and (d) w/o-MAPbBr3. The black dashed lines denote the threshold of stimulated emission. SPE: spontaneous emission. Pump fluence dependent PL measurements of the w/o-MAPbBr3 also demonstrated ASE. However, in this case, the band overlays the ST emission at 2.263 eV (548 nm) and possessed a much higher threshold fluence of ~15 μJ cm-2 (Figure 4(b)). The peak position of this ASE is comparable to that previously reported,8 indicating a similar EHP mechanism. The blue- shift of this ASE peak under higher pump fluences further supports this scenario. Due to the dominance of Auger recombination under these pumping conditions,36 the optical amplification of the emission is weak (Figure 4(d)). Thus, although widely reported, the EHP STE has a greater reliance on the gain medium and cavity structure to fabricate laser performances with modest power consumption. Figure 5. Composition control to improve the charge transfer and the STE performance. (a) Configuration coordinate diagram of total energy (electronic + elastic energies) of the perovskite to illustrate the charge transfer processes between the FE and BE states. Charge transfer occurs through the state coupling point between FE and BE to sustain energy conservation. The energy barrier (ΔEa) is the energy difference between the state coupling point and the bottom of the FE. Shifting of BE level in the energy axis results in a change in the ΔEa and the charge transfer velocity. (b) STE spectra of the MAPbBr3 and MA1.05PbBr3.05, which show a blue-shift of the BE energy level when the perovskite composition is adjusted from 1 to 1.05. (c) Arrhenius curves of deep trap states of the MAPbBr3 and MA1.05PbBr3.05 samples to derive their attempt-to-escape frequency (ν0). (d) PL intensity of the MA0.95PbBr2.95, MAPbBr3 and MA1.05PbBr3.05 as a function of pump fluence. The MA1.05PbBr3.05 shows an obvious enhancement in the BE STE and saturation in the FE emission under high pump fluences. The gain/absorption spectrum is included in the inset. (e) Photo of the STE under an edge-coupled optical configuration, in which the film was pumped by an optical stripe and emission was detected from the film edge. (f) Stripe length (z) dependent emission intensity (solid circles: experimental data; solid lines: theoretical fittings to derive the gain coefficient (g)). The low threshold STE within the 6-MAPbBr3 is inherent to the BE optical gain mechanism; however, its super-radiance is not prominent from the pump-dependent PL curves. This arises because the emitted light only experiences a short gain length of up to ~200 nm within a reflection configuration and a large proportion of charges exist within the FE states. The linear relationship between the FE emission intensity and pump fluence indicate that the FE states are weakly correlated to the BE STE process. To overcome this intrinsic factor, enhancement in the charge transfer between the FE and BE states within the STE duration is required. This is a non-radiative, multi-phonon assisted process, whose rate is limited by an energy barrier (ΔEa, Figure 5(a)) that is determined by state coupling and energy conservation rules. 37-38 Based on these, at a given temperature, reducing the energy difference of the initial and final states (i.e. FE-BE) will promote the charge transfer. Compared to a pure crystallization approach, element composition enables a more direct route to control the self-doping and thus the trap state distribution within the resulting perovskites.39-40 As such, three samples with MABr/PbBr2 ratios of 0.95, 1.0 and 1.05, respectively, were optimised. By using carrier density profiling measurement (Supplementary Figure 22), a much more concentrated carrier distribution was observed in the surface region of MA0.95PbBr2.95 film due to the MA or Br deficiency, while the MA1.05PbBr3.05 exhibited a similar carrier distribution to that of MAPbBr3. The FE and BE emission bands were observed for all three samples, with the FE emission peaks all located at a similar energy position of ~2.32 eV. The high carrier density at the surface of the MA0.95PbBr2.95 led to a reduced STE performance, presumably due to charge screening effects. Meanwhile, the BE emission peak of the MA1.05PbBr3.05 was observed at a slightly higher energy of ~2.265 eV. This feature is also presented in the STE spectra shown in Figure 5(b), implying a non- negligible difference exists in the energy level of the BE state. To better understand the underlying traps associated with the BE states across these samples, we have performed temperature-dependent capacitance measurements to extract their energy level and attempt- to-escape frequencies (ν0).30 As shown in Figure 5(c), the MA1.05PbBr3.05 exhibits a slightly lower trap energy compared to the stoichiometric sample, but with a 7-fold increase in ν0. These modifications in the energy states are favourable for enhancing the charge transfer between FE and BE states within the sample, thus resulting in a more optimised three energy level system for achieving STE. The enhanced exciton transfer within the BE STE process of MA1.05PbBr3.05 is indicated by the femtosecond TA (Supplementary Figure 23-24), in which the FE PB shows a much faster decay in the early 10 ps and the initial depopulation of the BE PB is obviously slowed down. The impact is finally confirmed with a 3-fold enhancement in the optical amplification of the MA1.05PbBr3.05 sample (Figure 5(d)). Under an edge-coupled optical configuration, green STE emission can be clearly observed from the sample edge (Figure 5(e)). Notably, without a designed cavity for optical resonance, the STE light here did not show good directionality. A key material parameter that reflects this amplification process is the optical gain coefficient. By using a variable stripe length (VSL) method8,28, the net gain coefficient of MAPbBr3 and MA1.05PbBr3.05 was determined to be 196 cm-1 and 385 cm-1, respectively, under a pump density of 9×1017 cm-3. The gain spectrum of the MA1.05PbBr3.05 was further determined using pump-probe measurements taken at a 2 ps time delay, and a peak gain coefficient of up to ~1300 cm-1 was derived (inset of Figure 5(d) and Supplementary Figure 25). Compared to VSL method, the pump-probe method minimizes the optical losses, which results in a more realistic measurement of a material's intrinsic gain.41 The gain coefficients obtained here are among the highest for reported perovskites at such low pump density (Supplementary Table S3), and are even comparable to that of conventional quantum-well structures.13, 41-42 Discussions In this work we have demonstrated room temperature STE arising from a bound exciton state for the first time within bulk perovskite MAPbBr3 films through judicial energy state engineering. Its existence enables the material to possess a high gain coefficient of ~1300 cm- 1 and for optical amplification to be observed at a remarkably low threshold of ~1.6×1017 cm- 3. These results highlight the tremendous promise that bulk metal halide perovskites offer towards laser application. Until now, only several works have reported RT excitonic STE from bulk semiconductor ~35 meV) perovskite films, films. Within inorganic CsPbCl3 (Ry ~64 meV) and CsPbBr3 (Ry optical gain states were achieved at very high threshold powers of ~100 kW cm-2.43-44 These were considered to originate from inelastic exciton-exciton scattering. Meanwhile, exciton- cavity polaritons were reported within CsPbBr3 and MAPbBr3 perovskite nanowires, resulting in STE with reduced threshold fluences of ~15 μJ cm-2.45-46 For metal halides, such of >100 meV, RT excitonic as CuBr and CuCl, which are well known to possess higher Ry STE has not been observed, even under very high pump powers.47 RT STE has been reported * ~60 meV) with a threshold power of 240 kW cm-2, although the underlying for ZnO (Ry mechanism is still under debate.10, 19 These studies demonstrate that RT STE within bulk semiconductors has been achieved; however, due to a fundamental difference in the exciton emission mechanisms, they usually have much larger STE thresholds compared to what has been reported here. For quantum systems that have an enhanced exciton effect, such as CsPbBr3 quantum dots,25,28,48 comparable STE thresholds of several uJ cm-2 have been demonstrated, but questions about their excitonic or EHP STE mechanisms remain unanswered. The excellent STE characteristics of the studied hybrid perovskites arise from their distinct * of ~ 40 meV for MAPbBr3 is higher than that of inorganic excitonic properties. Firstly, a Ry semiconductors with a similar Eg, such as ZnTe (~12 meV) and CdS (~25 meV), as well as * leads to a wider Eg candidates, such as GaN (~25 meV) and ZnS (~28 meV).18 This large Ry high Mott density, which helps to stabilise the RT excitonic phase at high carrier densities, BE of enabling the population inversion of the BE state. Secondly, to our knowledge, an Eb ~100 meV is among the largest reported for bulk semiconductors,31 which has approached the upper limit of Haynes rule.18 It is evident that what enables such rich optoelectronic properties to emerge within perovskites is the underlying presence of tunable defects. The relatively wide and stable phase regime of this ternary semiconductor provides a large parameter space for self-doping and trap-state engineering using simple solution processes.33 In this work we have focused on intrinsically derived modifications to the energy landscape. However, one can envisage the possibilities that further extrinsic dopants could provide, particularly by modifying their excitonic properties. With exciton states being sensitive to external fields, such as electric or magnetic, high sensitive electrical-optical and magnetic-optical coupled devices can be considered. Moreover, the spin properties of the BE state provides implications for quantum * of the hybrid perovskite, and manipulation and communication.49-50 The relatively large Ry especially the stabilised bound exciton state, helps to make these applications practical at high temperatures, possibly even at room temperature. Each of these exciting prospects requires a greater level of understanding into the nature of the carrier and exciton states within perovskite materials. References 1. Maiman, T. H. Stimulated Optical Radiation in Ruby. Nature 187, 493-494 (1960). 2. Shulika, O. V., Sukhoivanov, I. V. Advanced Lasers. Springer Netherlands, 2015. 3. Rickman, A. The commercialization of silicon photonics. Nat. Photon. 8, 579-582 (2014). 4. Heck, M. J. R., Bowers, J. E. Energy efficient and energy proportional optical interconnects for multi-core processors: driving the need for on-chip sources. IEEE J. Sel. Top Quant. 20, 8201012 (2014). 5. Liang, D., Bowers, J. E. Recent progress in lasers on silicon. Nat Photonics 4, 511-517 (2010). 6. a) Sutherland, B. R. and Sargent, E. H. Perovskite photonic sources. Nat. Photon. 10, 295- 302 (2016); b) Stranks, S. D. and Snaith, H. J. Metal-halide perovskites for photovoltaic and light-emitting devices. Nat. Nanotech. 10, 391-402 (2015). 7. Zhu, H., Fu, Y., Meng, F., Wu. X., Gong, Z., Ding, Q., Gustafsson, M. V., Trinh, M. T., Jin, S. and Zhu, X. Lead halide perovskite nanowire lasers with low lasing thresholds and high quality factors. Nat. Mater. 14, 636-642 (2015). 8. Xing, G., Mathews, N., Lim, S. S., Yantara, N., Liu, X., Sabba, D., Grätzel, M., Mhaisalkar, S., and Sum, T. C. Low-temperature solution-processed wavelength-tunable perovskites for lasing. Nat. Mater. 13, 476-480 (2014). 9. Newbury, P. R., Shahzad, K. and Cammack, D. A. Stimulated emission via inelastic exciton-exciton scattering in ZnSe epilayers. Appl. Phys. Lett. 58, 1065-1067 (1991). 10. Bagnall, D. M., Chen, Y. F., Zhu, Z., Yao, T., Koyama, S., Shen, M. Y. and Goto, T. Optically pumped lasing of ZnO at room temperature. Appl. Phys. Lett. 70, 2230-2232 (1997). 11. Ding, J., Jeon, H., Ishihara, T., Hagerott, M., Nurmikko, A. V., Luo, H., Samarth, N. and Furdyna, J. Excitonic Gain and Laser Emission in ZnSe-Based Quantum Wells. Phys. Rev. Lett. 69, 1707-1710 (1992). 12. Wegscheider, W., Pfeiffer, L. N., Dignam, M. M., Pinczuk, A., West, K. W., McCall, S. L. and Hull, R. Lasing from Excitons in Quantum Wires. Phys. Rev. Lett. 71, 4071-4074 (1993). 13. Puls, J., Mikhailov, G. V., Henneberger, F., Yakovlev, D. R., Waag, A. and Faschinger, W. Laser Action of Trions in a Semiconductor QuantumWell. Phys. Rev. Lett. 89, 287402 (2002). 14. Huang, M. H., Mao, S., Feick, H., Yan, H., Wu, Y., Kind, H., Weber, E., Russo, R. and Yang, P. Room-Temperature Ultraviolet Nanowire Nanolasers. Science 291, 1897-1899 (2001). 15. Klimov, V. I., Mikhailovsky, A. A., Xu, S., Malko, A., Hollingsworth, J. A., Leatherdale, C. A., Eisler, H. J. and Bawendi, M. G. Optical Gain and Stimulated Emission in Nanocrystal Quantum Dots. Science 290, 314-317 (2000). 16. Yan, R., Gargas, D. and Yang, P. Nanowire photonics. Nat. Photon. 3, 569-576 (2009). 17. Jasieniak, J. J., Fortunati, I., Gardin, S., Signorini, R., Bozio, R., Martucci, A. and Mulvaney, P. Highly Efficient Amplified Stimulated Emission from CdSe-CdS-ZnS Quantum Dot Doped Waveguides with Two-Photon Infrared Optical Pumping. Adv. Mater. 20, 69-73 (2008). 18. Klingshirn, C. F. Semiconductor optics (4th). Springer, 2012. 19. Klingshirn, C., Hauschild, R., Fallert, J. and Kalt, H. Room-temperature stimulated emission of ZnO: Alternatives to excitonic lasing. Phys. Rev. B 75, 115203 (2007). 20. Veldhuis, S. A., Boix, P. P., Yantara, N., Li, M., Sum, T. C., Mathews, N. and Mhaisalkar, S. G. Perovskite Materials for Light‐Emitting Diodes and Lasers. Adv. Mater. 28, 6804- 6834 (2016). 21. Saba., M., et al. Correlated electron-hole plasma in organometal perovskites. Nat. Commun. 5, 5049 (2014). 22. Liao, Q., Hu, K., Zhang, H., Wang, X., Yao, J. and Fu, H. Perovskite Microdisk Microlasers Self-Assembled from Solution. Adv. Mater. 27, 3405-3410 (2015). 23. Eaton, S. W., Lai, M., Gibson, N. A., Wong, A. B., Dou, L., Ma, J., Wang, L. Leone, S. R. and Yang, P. Lasing in robust cesium lead halide perovskite nanowires. PNAS 113, 1993- 1998 (2016). 24. Geiregat, P., Maes, J., Chen, K., Drijvers, E., De Roo, J., Hodgkiss, J. M. and Hens, Z. Using Bulk-like Nanocrystals To Probe Intrinsic Optical Gain Characteristics of Inorganic Lead Halide Perovskites. ACS Nano 12, 10178-10188 (2018). 25. Wang, Y., Li, X., Song, J., Xiao, L., Zeng, H., Sun, H. All-Inorganic Colloidal Perovskite Quantum Dots: A New Class of Lasing Materials with Favorable Characteristics. Adv. Mater. 25, 7101-7108 (2015). 26. Fu, Y., Zhu, H., Schrader, A. W., Liang, D., Ding, Q., Joshi, P., Hwang, L., Zhu, X. and Jin, S. Nanowire Lasers of Formamidinium Lead Halide Perovskites and Their Stabilized Alloys with Improved Stability. Nano Lett. 16, 1000-1008 (2016). 27. Deschler, F., et al. High Photoluminescence Efficiency and Optically Pumped Lasing in Solution-Processed Mixed Halide Perovskite Semiconductors. J. Phys. Chem. Lett. 5, 1421- 1426 (2014). 28. Yakunin, S., Protesescu, L., Krieg, F., Bodnarchuk, M. I., Nedelcu, G., Humer, M., De Luca, G., Fiebig, M., Heiss, W., Kovalenko, M. V. Low-threshold amplified spontaneous emission and lasing from colloidal nanocrystals of caesium lead halide perovskites. Nat. Commun. 6, 8056 (2015). 29. D'Innocenzo, V., Grancini, G., Alcocer, M. J. P., Kandada, A. R. S., Stranks, S. D., Lee, M. M., Lanzani, G., Snaith, H. J., Petrozza, A. Excitons versus free charges in organo-lead tri-halide perovskites. Nat. Commun. 5, 3586 (2014). 30. Walter, T., Herberholz, R., Muller, C., Schock, H. W. Determination of defect distributions from admittance measurements and application to Cu(In,Ga)Se2 based heterojunctions. J. Appl. Phys. 80, 4411-4420 (1996). 31. Shi, J., Zhang, H., Li, Y., Jasieniak, J., Li, Y., Wu, H., Luo, Y., Li, D. and Meng Q. Identification of high-temperature exciton states and their phase-dependent trapping behaviour in lead halide perovskites. Energy Environ. Sci. 11, 1460-1469 (2018). 32. Haynes, J. R. Experimental proof of the existence of a new electronic complex in silicon. Phys. Rev. Lett. 4, 361-363 (1960). 33. Shi, T., Yin, W., Hong, F., Zhu, K. and Yan, Y. Unipolar self-doping behavior in perovskite CH3NH3PbBr3. Appl. Phys. Lett. 106, 103902 (2015). 34. Xiao, M., Huang, F., Huang, W., Dkhissi, Y., Zhu, Y., Etheridge, J., Gray-Weale, A., Bach, U., Cheng, Y. and Spiccia, L. A Fast Deposition-Crystallization Procedure for Highly Efficient Lead Iodide Perovskite Thin-Film Solar Cells. Angew. Chem. Int. Ed. 53, 9898- 9903 (2014). 35. Yang, Y., Ostrowski, D. P., France, R. M., Zhu, K., Lagemaat, J. van de, Luther, J. M., Beard, M. C. Observation of a hot-phonon bottleneck in lead-iodide perovskites. Nat. Photon. 10, 53-59 (2016). 36. Trinh, M. T., Wu, X., Niesner, D., Zhu, X. -Y. Many-body interactions in photo-excited lead iodide perovskite. J. Mater. Chem. A 3, 9285-9290 (2015). 37. Henry, C. H. and Lang, C. V. Nonradiative capture and recombination by multiphonon emission in GaAs and GaP. Phys. Rev. B 15, 989-1016 (1977). 38. Huang, K. and Rhys, A. Theory of light absorption and non-radiative transitions in F- centres. Proc. R. Soc. A 204, 406-423 (1950). 39. Wang, Q., Shao, Y., Xie, H., Lyu, L., Liu, X., Gao, Y. and Huang, J. Qualifying composition dependent p and n self-doping in CH3NH3PbI3. Appl. Phys. Lett. 105, 163508 (2014). 40. Shi, J., Wei, H., Lv, S., Xu, X., Wu, H., Luo, Y., Li, D. and Meng, Q. Control of Charge Transport in the Perovskite CH3NH3PbI3 Thin Film. ChemPhysChem 16, 842-847 (2015). 41. Kreller, F., Lowisch, M., Puls, J. and Henneberger, F. Role of Biexcitons in the Stimulated Emission of Wide-Gap II-VI Quantum Wells. Phys. Rev. Lett. 75, 2420-2423 (1995). 42. Ding, J., Kozlov, V., Kelkar, P., Salokatve, A. and Nurmikko, A. V. Physics of Gain and Stimulated Emission in ZnSe-Based Quantum Well Lasers. Phys. Stat. Sol. (b) 188, 153-163 (1995). 43. Kondo, S., Ohsawa, H., Saito, T., Asada, H. and Nakagawa, H. Room-temperature stimulated emission from microcrystalline CsPbCl3 films. Appl. Phys. Lett. 87, 131912 (2005). 44. Kondo, S., Takahashi, K., Nakanish, T., Saito, T., Asada, H. and Nakagawa, H. High intensity photoluminescence of microcrystalline CsPbBr3 films: Evidence for enhanced stimulated emission at room temperature. Curr. Appl. Phys. 7, 1-5 (2007). 45. Evans, T. J. S., Schlaus, A., Fu, Y., Zhong, X., Atallash, T. L., Spencer, M. S., Brus, L. E., Jin, S. and Zhu, X. Continuous-Wave Lasing in Cesium Lead Bromide Perovskite Nanowires. Adv. Opt. Mater. 6, 1700982 (2018). 46. Zhang, S., et al. Strong Exciton-Photon Coupling in Hybrid Inorganic-Organic Perovskite Micro/Nanowires. Adv. Opt. Mater. 6, 1701032 (2018). 47. Kondo, S. and Saito, T. Dramatic improvement of excitonic photoluminescence in metal halide films. J. Lumin. 130, 191-205 (2010). 48. Wang, Y., Zhi, M., Chang, Y., Zhang, J. and Chan, Y. Stable, Ultralow Threshold Amplified Spontaneous Emission from CsPbBr3 Nanoparticles Exhibiting Trion Gain. Nano Lett. 18, 4976-4984 (2018). 49. Zrenner, A., Beham, E., Stufler, S., Findeis, F., Bichler, M. and Abstreiter, G. Coherent properties of a two-level system based on a quantum-dot photodiode. Nature 418, 612-614 (2002). 50. Biolatti, E., Lotti, R. C., Zanardi, P. and Rossi, F. Quantum Information Processing with Semiconductor Macroatoms. Phys. Rev. Lett. 85, 5647-5650 (2000). Acknowledgements This work was supported by Natural Science Foundation of China (No. 11874402, 91733301, 51761145042, 51421002, 51627803, 51872321) and the International Partnership Program of Chinese Academy of Sciences (No. 112111KYSB20170089). JJ would like to acknowledge funding through the ARC Centre of Excellence in Exciton Science (CE170100026). Author contributions J. Shi designed this study, and made experimental measurements, theoretical calculations and analysis; Y. Li prepared the sample and made photoluminescence measurements; J. Wu assisted in capacitance and transient absorption measurements and data extraction; H. Wu, Y. Luo and D. Li participated in organizing the measurement setups and assisted in the sample preparation; J. Jasieniak conceived and supported this project and participated in experimental and theoretical analysis; Q. Meng conceived and supported this project. All of authors contributed to the discussion. The manuscript was written by J. Shi, corrected by J. Jasieniak and Q. Meng and approved by all contributions. Additional information Supplementary information is available in the online version of the paper. Competing financial interests The authors declare no competing financial interests. Methods Perovskite film deposition and characterisations. The MAPbBr3 and MAPbI3 samples were deposited by using an anti-solvent method, where chlorobenzene is used as the anti-solvent. Briefly, a 1M MAPbBr3 (or MAPbI3) precursor solution in dimethylformamide (DMF) was firstly spread onto a quartz glass/compact Al2O3 substrate. A few of certain seconds after the beginning of spin coating at 4000 rpm, the chlorobenzene was dropping onto the film to produce a fast crystallisation of the perovskite. To adjust the film deposition process, the addition time for applying the anti-solvent was controlled as described in the main text. After 30 s of spinning, the film was transferred to a hot plate at 100 °C and annealed for 10 min. A 50 mg/ml PMMA solution was finally spin coated onto the perovskite film for protection. To adjust the composition, the precursors with different ratios of MABr/PbBr2 were used. For each composition, the dropping time of anti-solvent has been optimised. The phase of the final perovskite was characterised by a Panalytical X-ray diffractometer. The averaged thickness of each film is measured by using a step profiler. Optical characterization. The light transmission and reflectance of the perovskite film were measured with a UV-vis spectrometer (Shimadzu 3600). The light absorption is calculated from the transmission and reflectance. The steady-state and transient PL of the perovskite film were measured using an Edinburgh Fluorescence Spectrometer (FLS 920). A 445 nm pulsed diode laser (EPL-445, ~5 nJ cm-2, 62 ps) was used as the excitation source. A circular adjustable neutral density filter was adopted to adjust the excitation intensity. The PL was collected in the reflectance mode, where a PMT together with a TCSPC module was applied to detect the time-resolved or time-integrated PL. For temperature dependent measurements, an ARS liquid helium cryostat was employed as the sample chamber under vacuum, where the temperature was controlled using a Lakeshore controller. The perovskite film on the glass substrate was directly fixed to the optical sample holder by screws. During the temperature adjustment, the system was firstly stabilised for at least 5 min at each temperature. Before the PL measurements, the perovskite sample was always kept in the dark without any bias light or laser illumination to avoid the possible production of charge accumulation and a local electric field. For the STE measurement, a Coherent Astrella femtosecond amplifier (1 KHz, 35 fs, 7 mJ/pulse) together with an optical parameter amplifier (OPA, Opera-solo, 4 mJ/pulse split pulses for the OPA) was used as the pump source. The pump pulse was guided toward the sample vertically by planar mirrors and final a convex mirror to obtain a light spot diameter of ~12 mm. The laser power was measured by a coherent power meter. The averaged carrier density under photo-injection was calculated by considering the reflection and transmission. The emission from the sample was coupled into an optical fibre, placed at an angle relative to the incident laser of ~45°, through a long-pass filter and a convex lens. The spectrum was measured using an Avantes spectrometer (AvaSpec-ULS2048CL-EVO) with a CMOS detector operating at an integration time of 50 ms. The final spectrum were obtained by averaging 200 spectra with a total recording time of 10 s. For polarisation measurements, a linear polariser was added between the film and the optical fibre. For different polariser angles, the emission spectra were collected, and their intensity was compared directly. No difference in the STE intensity could be observed. For the femtosecond transient absorption measurement, a high energy pulse generated from the femtosecond amplifier and the OPA were used as the pump light, and a white light generated by focusing the 800 nm (35 fs) pulse onto a 2 mm-thick sapphire and filtered by a short-pass filter (775 nm, Sigmakoki) was used as the probe light. The probe light was vertically focused onto the sample through a series of achromatic lens, with a final probe spot diameter of <1 mm. The pump light was focused onto the same centre by a concave mirror, with a spot diameter of ~5 mm. There was an angle of ~20° between the pump and probe light, that is, non-degenerate configuration. The time delay between these two pulses was controlled by using an Electric stage (Zolix) together with a hollow retroreflector (Newport). The transmitted probe light was detected by a CMOS optical fibre spectrometer (Avantes, AvaSpec-ULS2048CL-EVO) with a single-pulse mode (integrating time 30 μs). A pinhole 10 cm away from the sample is put in this light path to eliminate the light scattering from pump light and PL. With multiple spectra recording and calculating, the noise of the ΔA is lower than 0.1 mOD. The chirp of the probe pulse was calibrated by measuring the Kerr effect of CS2. The VSL measurement was performed based on another CCD array optical fibre spectrometer (Ocean optics QE pro). Electrical characterisation. The trap states of the MAPbBr3 film were measured by using thermal admittance spectroscopy (TAS). The frequency-dependent capacitances of the FTO/compact TiO2/MAPbBr3/Spiro/Au cells were measured by using an electrochemical station (Princeton). The cell was placed inside a vacuum chamber of a low-temperature probe station (Lakeshore TTPX) with liquid nitrogen as the cooling source. The bandwidth of the electrical probe was 1 GHz. For capacitance measurement, the cell was always measured under dark conditions with no bias voltage being applied to the cell. The data analysis was performed according to published literature describing TAS. The trap state energy levels were derived from the Arrhenius curves. Mott-Schottky curves were also measured to help obtain trap density approximations.
1909.06046	1	1909	2019-09-13T05:57:09	Verification of shielding effect predictions for large area field emitters	[ "physics.app-ph", "cond-mat.mes-hall", "physics.plasm-ph" ]	A recent analytical model for large area field emitters, based on the line charge model (LCM), provides a simple approximate formula for the field enhancement on hemi-ellipsoidal emitter tips in terms of the ratio of emitter height and pairwise distance between neighboring emitters. The formula, verified against the exact solution of the linear LCM, was found to be adequate provided the mean separation between emitters is larger than half the emitter height. In this paper, we subject the analytical predictions to a more stringent test by simulating (i) an infinite regular array and (ii) an isolated cluster of 10 random emitters, using the finite element software COMSOL. In case of the array, the error in apex field enhancement factor (AFEF) is found to be less than $0.25\%$ for an infinite array when the lattice constant $c \geq 1.5h$, increasing to $2.9\%$ for $c = h$ and $8.1\%$ for $c = 0.75h$. For an isolated random cluster of 10 emitters, the error in large AFEF values is found to be small. Thus, the error in net emitted current is small for a random cluster compared to a regular infinite array with the same (mean) spacing. The line charge model thus provides a reasonable analytical tool for optimizing a large area field emitter.	physics.app-ph	physics	Verification of shielding effect predictions for large area field emitters Rashbihari Rudra1, 2 and Debabrata Biswas1, 2 1)Bhabha Atomic Research Centre, Mumbai 400 085, INDIA 2)Homi Bhabha National Institute, Mumbai 400 094, INDIA A recent analytical model for large area field emitters, based on the line charge model (LCM), provides a simple approximate formula for the field enhancement on hemiellipsoidal emitter tips in terms of the ratio of emitter height and pairwise distance between neighbouring emitters. The formula, verified against the exact solution of the linear LCM, was found to be adequate provided the mean separation between emitters is larger than half the emitter height. In this paper, we subject the analytical predictions to a more stringent test by simulating (i) an infinite regular array and (ii) an isolated cluster of 10 random emitters, using the finite element software COMSOL. In case of the array, the error in apex field enhancement factor (AFEF) is found to be less than 0.25% for an infinite array when the lattice constant c ≥ 1.5h, increasing to 2.9% for c = h and 8.1% for c = 0.75h. For an isolated random cluster of 10 emitters, the error in large AFEF values is found to be small. Thus, the error in net emitted current is small for a random cluster compared to a regular infinite array with the same (mean) spacing. The line charge model thus provides a reasonable analytical tool for optimizing a large area field emitter. I. INTRODUCTION Large area field emitters (LAFE) are promising as a high brightness source of cold electrons. They have been investigated for around four decades, as patterned arrays of pointed emitters or clusters of nanotubes or nanorods. They find applications in various vacuum devices such as X-ray tubes, terahertz generators and even in space navigation1 -- 6. The electron emission mechanism, at least in metals, is fairly well understood7 -- 10 and it is recognized that the enhanced local electric10 -- 12 field at a curved emitter tip leads to an increase in the quantum transmission coef- ficient. Thus, significant currents can be observed even at moderate macroscopic fields. The problem in bringing together a bunch of curved emitter tips is also well rec- ognized. The proximity of emitters causes electrostatic shielding and hence the local field enhancement on emit- ter tips is not as much compared to isolated emitters13 -- 16. Thus, a LAFE must have an optimal packing density such that the overall current density is maximum. Knowing the degree of field enhancement for an iso- lated emitter of given shape, height (h) and apex radius of curvature (Ra) is in itself a formidable task. For a LAFE, shielding makes this all the more complicated. At present, a theory of LAFE exists for only the sim- plest of emitters where the shape is hemiellipsoidal and for which the apex field enhancement factor (AFEF) for an isolated emitter is known analytically17,18. The the- ory predicts the AFEF, γa of an (ith) emitter in a given LAFE environment in terms of distances from all other emitters. Thus16, j(cid:54)=i αSij and 1 + 4δ2 ij + 2δij a (cid:39) γ(i) 2h/Ra ln(cid:0)4h/Ra (cid:1) − 2 + αSi j(cid:54)=i(λj/λi)αSij (cid:39)(cid:80) (cid:12)(cid:12)(cid:12)(cid:113) 1 −(cid:113) (cid:105) 1 + 4δ2 ij + ln where αSi =(cid:80) (cid:104) αSij = 1 δ12 (1) (cid:12)(cid:12)(cid:12) with δij = h/ρij, ρij = [(xi − xj)2 + (yi − yj)2]1/2 being the distance between on the cathode plane (XY ) between the ith and jth emitter. In the above, λ is the slope of the line charge density Λ(z) (i.e. Λ(z) = λz), obtained by projecting the surface charge density along the emitter axis15. Eq. (1) is approximate since αSi assumes the charge distribution on the ith and jth emitter to be identical. This is largely true when they are not too close so that Eq. (1) serves as a useful first approximation. It can however cause errors for separations smaller than the emitter height. A second source of errors concerns the nature of the charge distribution on the surface of the hemi-ellipsoid. An isolated hemiellipsoid with its axis aligned along an external macroscopic field E0 z, has a projected line charge distribution (along the emitter (Z) axis) that is linear15. However, when such emitters are close together, the line charge distribution can develop non-linear components. Since Eq. (1) is based on a linear model, this may contribute to the error in AFEF when emitters are close together. Note that Eq. (1) compares well with the exact lin- ear LCM model for an N -emitter random LAFE. For mean spacings larger than h/2, the observed error was < 6% while for mean spacings larger than h, the error is less than 1.5%. This comparison however neglects the non-linearity factor in the charge distribution. It is thus necessary to subject Eq. (1), obtained using linear LCM, to more stringent tests such as by comparing its predic- tions with numerical (finite element) simulations where the projected charge density does not have constraints of linearity. Such a comparison is also required since LCM predictions are reported to be at variance with other models/numerical predictions19. A reasonable outcome from the present study can pave the way for a greater reliance on LCMs for analytical investigations of large area field emitters where direct numerical methods are difficult to implement due to computational constraints. The paper is organized as follows. We shall first take a look at the computational domain required to model a large area field emitter using COMSOL with the 'anode- at-infinity' and also study the number of neighbouring emitters required for the convergence of the AFEF using Eq. (1). Next, we shall compare the predictions of linear LCM (i.e. Eq. (1)) with those of COMSOL for an infi- nite square array. Finally, we shall also study a bunch of random isolated emitters using COMSOL. Rather than studying just the error in apex field enhancement factor of individual emitters in the cluster, we shall also com- pute the error in net emitted current so that emitters in close proximity do not get a disproportionately large weight in deciding the error in LCM prediction. Finally, we shall discuss the implications of our results in design- ing large area field emitters. II. DOMAIN SIZE FOR COMSOL AND LCM The array-at-infinity is an idealization that simplifies the line charge model but is not essential to it. Compu- tationally, an infinite square array with lattice constant c can be simulated by imposing 'zero surface charge den- sity' at x, y = ±c/2. Thus, ∂V /∂(x, y) = 0 at x = ±c/2 and y = ±c/2. The boundary condition at the anode can be Dirichlet (V = VA, where VA is the anode poten- tial) or Neumann20 (∂V /∂z = 0E0, E0 being the magni- tude of the macroscopic field −E0 z) at z sufficiently far from the emitter tip. A generally accepted guideline is to place the anode at about 5 times the emitter height in order to impose the Dirichlet boundary condition while for the Neumann condition, the anode can be somewhat closer. It is important however to test for convergence to the anode-at-infinity result by pushing the anode further away in both cases. 2 lipsoidal array with h = 1500 µm, Ra = 1.5µm and c = h using COMSOL. The anode-cathode plane distance is in- creased from D = 5h for both the Dirichlet and Neumann boundary conditions and the apex field enhancement fac- tor γa is plotted against D. Clearly, the Neumann bound- ary condition achieves the anode-at-infinity condition at a much smaller D value while the Dirichlet condition in this case requires the anode to be at D = 100h. We shall henceforth use the Neumann boundary condition for simulating the anode-at-infinity. FIG. 2. The apex enhancement factor evaluated using the LCM prediction of Eq. (1). All shielding emitters within a cir- cle of radius R are included. The emitter height h = 1500µm, Ra = 1.5µm while c = h. We shall next fix the question of the number of emit- ters required for convergence of the LCM result, Eq. (1). Figure 2 shows the apex field enhancement factor calcu- lated by including all (jth) emitters in a circle of radius R. The R = 0 limit corresponds to an isolated emitter while at R/h = 5 and c = h, the number of emitters is 80. At R/h = 60, where approximate convergence is achieved, the number of shielding emitters is 11288. In all AFEF calculations henceforth, we shall consider R/h = 100 to ensure that convergence in AFEF has been achieved. III. THE ERROR IN LCM PREDICTION With the question of domain size settled, we are now in a position to investigate the error in LCM prediction. As mentioned earlier, we shall consider (a) an infinite square array and (b) an isolated cluster of randomly placed emit- ters. FIG. 1. The apex enhancement factor evaluated using COM- SOL with the anode plate having Dirichlet (V = VA) and Neumann (∂V /∂z = 0E0). The anode-at-infinity is eas- ily achievable with the Neumann boundary condition. The emitter height h = 1500µm, lattice constant c = h while Ra = 1.5µm . Fig. 1 shows a convergence study for an infinite hemiel- A. Infinite square array Consider an infinite square array with lattice constant c and an emitter of height h = 1500µm and apex ra- dius of curvature Ra = 1.5µm. The lattice constant c is now varied and the apex field enhancement factor γa is 210 215 220 225 230 235 240 10 20 30 40 50 60 70 80 90 100γa(D)D/h Neumann (c=h) Dirichlet (c=h) 205 225 245 265 285 305 0 10 20 30 40 50 60γa(R)R/h c=h, LCMcalculated using (i) the LCM prediction of Eq. (1) with R = 100h and (ii) COMSOL with Neumann boundary condition at the anode with the anode-cathode separa- tion fixed at D = 5h. The relative error Error(%) = γLCM a − γCOM SOL a γCOM SOL a × 100 (2) is calculated at different values of lattice constant c. The result is shown in Fig. 3. 3 can serve as a useful tool in optimizing the emitter den- sity of a LAFE. In the present context, an isolated cluster of randomly placed emitters is sought to be modelled. The compu- tational boundary must therefore be chosen to be suffi- ciently far away if a standard Neumann boundary is to be used in the X, Y directions. Typically, the domain con- sidered is [−10h, 10h] in the X, Y directions and [0, 5h] in the Z direction. The emitters are limited to a patch at the centre in the X, Y plane such that the mean spacing equals h. We have considered 2 such realizations, one having 5 emitters and the other having 10. We present here the results for 10 emitters. FIG. 3. The relative error in apex enhancement factor evalu- ated using the LCM prediction of Eq. (1). The 'exact' result is calculated using COMSOL. The emitter height h = 1500µm while Ra = 1.5µm. The error is about 2.9% for c = h and about 8.1% for c = 0.75h. It falls to about 1% for c = 1.25h and remains less than 0.5% for c ≥ 1.5h. For larger values of c, the error becomes smaller than 0.1%. Note that in the region where the optimal current density is expected to lie, the error remains small. Thus, the analytical result (Eq. (1)) based on the line charge model can be used to calculate the net emitted current accurately for a large array, provided the density of emitters is such that c ≥ h. B. Cluster of random emitters Randomly placed emitters pose a greater challenge in- sofar as verification of Eq. (1) is concerned. It is difficult to model these using a finite element software such as COMSOL since, even for reasonable computational re- sources, the number of emitters may be limited to about 25-30 depending on the h/Ra ratio. Note that the AFEF calculation for an array or cluster requires 3-dimensional modeling and the demands on resources increases as the curvature at the apex increases. Thus, an analytical model, if validated and found to be reasonably accurate, FIG. 4. A cluster of 10 randomly placed emitters in the XY plane The distribution of emitters in the XY plane is shown in Fig. 4. The apex field enhancement factor calculated using COMSOL and the line charge model is shown in Table I along with the error. Clearly, the error is small when the emitter has larger AFEF value (less shielding). This suggests that the error in net emission current may be smaller for a random cluster than a regular infinite array with the same mean spacing. The net field emission current can be calculated using the apex field enhancement factor and the generalized cosine law of local field variation21,22 around the emitter apex as shown in [23] for emitters having Ra > 100nm. For the random cluster under consideration, the error in net emission current density is a nominal 15.5% at a macroscopic field of 17.5 MV/m while for the infinite array having c = h, the roughly 3% error in the LCM- AFEF shifts the current density by a factor of 2. Both results are acceptable (given the inherent uncertainties in field emission theory predictions) though the random distribution has an edge insofar as the current estimation is concerned. These errors are likely to decrease as the mean separation or lattice constant increases and also with an increase in macroscopic field strengths. 0 2 4 6 8 10 2 4 6 8 10Error (%)c/hNeumann Anode at D = 5h-0.002-0.001 0 0.001 0.002-0.002-0.001 0 0.001 0.002Y (m)X (m)TABLE I. A comparison of apex field enhancement factors using COMSOL and LCM at different emitter locations in a cluster. X (m) Y (m) −1.418 × 10−3 1.867 × 10−3 1.186 × 10−4 −8.817 × 10−4 −9.511 × 10−4 −9.063 × 10−4 −1.913 × 10−3 9.696 × 10−4 2.114 × 10−3 1.839 × 10−3 −1.384 × 10−3 −3.521 × 10−4 4.507 × 10−4 −3.264 × 10−4 −1.652 × 10−4 2.220 × 10−3 1.852 × 10−3 −3.759 × 10−4 5.022 × 10−4 1.065 × 10−3 γCOM SOL a γLCM a Error (%) 274.97 258.35 262.71 275.49 294.34 264.13 254.63 278.38 285.85 268.47 271.81 250.64 255.78 272.08 293.54 257.51 246.74 275.79 283.91 263.78 1.15 2.98 2.63 1.23 0.275 2.50 3.10 0.932 0.677 1.75 4 V. REFERENCES 1C. A. Spindt, J. Appl. Phys., 39, 3504 (1968). 2C. A. Spindt, I. Brodie, L. Humphrey, and E. R. Westerberg, J. Appl. Phys. 47, 5248 (1976). 3R. J. Parmee, C. M. Collins, W. I. Milne, and M. T. Cole, Nano Convergence 2, 1 (2015). 4A. Basu, M. E. Swanwick, A. A. Fomani, and L. F. Velsquez- Garca, J. Phys. D Appl. Phys. 48, 225501 (2015). 5M. T. Cole, R. J. Parmee, and W. I. Milne, Nanotechnology 27, 082501 (2016). 6D. R. Whaley, C. M. Armstrong, C. E. Holland, C. A. Spindt, P. R. Schwoebel, 31st International Vacuum Nanoelectronics Conference (IVNC) (2018), 10.1109/IVNC.2018.8520271. 7R. H. Fowler and L. Nordheim, Proc. R. Soc. A 119, 173 (1928). 8E. L. Murphy and R. H. Good, Phys. Rev. 102, 1464 (1956). 9R. G. Forbes, App. Phys. Lett. 89, 113122 (2006). 10K. L. Jensen, Field emission - fundamental theory to usage, Wi- ley Encycl. Electr. Electron. Eng. (2014). 11R. G. Forbes, C.J. Edgcombe and U. Valdr`e, Ultramicroscopy 95, 57 (2003). 12D. Biswas, Phys. Plasmas 25, 043113 (2018). 13F. H. Read and N. J. Bowring, Nucl. Instrum. Meth. Phys. Res. A 519, 305 (2004). 14J. R. Harris, K. L. Jensen, D. A. Shiffler and J. J. Petillo, Appl. Phys. Lettrs. 106, 201603 (2015). 15D. Biswas, G. Singh and R. Kumar, J. App. Phys. 120, 124307 IV. SUMMARY AND CONCLUSIONS (2016). 16D. Biswas and R. Rudra, Phys. Plasmas, 25, 083105 (2018). 17H. G. Kosmahl, IEEE Trans. Electron Devices 38, 1534 (1991). 18E. G. Pogorelov, A. I. Zhbanov, and Y.-C. Chang, Ultrami- croscopy 109, 373 (2009). 19R. G. Forbes, J. Appl. Phys. 120, 054302 (2016). 20T. A. de Assis and F. F. Dall'Agnol, J. Vac. Sci. Tech. B, 37, 022902 (2019). 21D. Biswas, G. Singh, S. G. Sarkar and R. Kumar, Ultrami- croscopy 185, 1 (2018). 22D. Biswas, G. Singh, R. Ramachandran, Physica E 109, 179 (2019). 23D. Biswas, Phys. Plasmas 25, 043105 (2018). The analytical predictions of the line charge model for large area field emitters has been the subject of investi- gation in the paper. The model for LAFE uses hemi- ellipsoid emitters as the basic building blocks. It al- lows a computation of the apex field enhancement factor (AFEF) for any emitter in terms of normalized pair-wise distance to the emitters in its LAFE neighbourhood. As a purely geometric model which ignores charge distri- bution details, the approximate values of AFEF that it provides has been the subject of scrutiny in this study. Our results show that if emitters are separated by av- erage distances approximately equal to or greater than the height of the emitters, the errors in the prediction of LCM model are small and the values of net emitted cur- rent are acceptable. In the process of establishing this, we also demonstrated the domain size necessary for sim- ulating a large area field emitter, both from the point of view of finite element methods and the line charge model. The results are particularly encouraging for field emis- sion since the line charge model performs well for emitters that contribute significantly to the net current density. It can thus be used to study large clusters of emitters which are otherwise inaccessible to computations due to the re- sources involved. Practical emitter shapes used in field emission may vary from cylindrical structures to cones with the added possibility of differently shaped endcaps. The hemi- ellipsoid can be used to approximate these keeping the apex radius of curvature invariant. This will undoubtedly lead to errors in the AFEF calculation of single emitters since the emitter base plays an important role. For a random LAFE however, it is worth investigating if such finer points take a back seat.
1903.07538	1	1903	2019-03-18T16:25:08	InGaN/GaN {\mu}LED SPICE modelling with size dependent ABC model integration	[ "physics.app-ph" ]	The need of high brightness micro-displays in portable applications dedicated to mixed and/or virtual reality has drawn an important research wave on InGaN/GaN based micro-sized light emitting diodes ({\mu}LEDs). We propose to use a SPICE modelling technique to describe and simulate the electro-optical behavior of the {\mu}LED. A sub-circuit portrayal of the whole device will be used to describe current-voltage behavior and the optical power performance of the device based on the ABC model. We suggest an innovative method to derive instantaneously the carrier concentration from the simulated electrical current in order to determine the {\mu}LED quantum efficiency. In a second step, a statistical approach is also added into the SPICE model in order to apprehend the spread on experimental data. This {\mu}LED SPICE modelling approach is very important to allow the design of robust pixel driving circuits.	physics.app-ph	physics	InGaN/GaN µLED SPICE modelling with size dependent ABC model integration Anis Daamia, François Oliviera aUniversité Grenoble Alpes, CEA-LETI, Minatec Campus, III-V Lab., Grenoble, France ABSTRACT The need of high brightness micro-displays in portable applications dedicated to mixed and/or virtual reality has drawn an important research wave on InGaN/GaN based micro-sized light emitting diodes (µLEDs). We propose to use a SPICE modelling technique to describe and simulate the electro-optical behavior of the µLED. A sub-circuit portrayal of the whole device will be used to describe current-voltage behavior and the optical power performance of the device based on the ABC model. We suggest an innovative method to derive instantaneously the carrier concentration from the simulated electrical current in order to determine the µLED quantum efficiency. In a second step, a statistical approach is also added into the SPICE model in order to apprehend the spread on experimental data. This µLED SPICE modelling approach is very important to allow the design of robust pixel driving circuits. Keywords: InGaN/GaN, µLED, SPICE simulations, ABC model, quantum efficiency 1. INTRODUCTION Many research efforts have been carried out and a huge amount of interesting results can be found in literature dealing with the comprehension of electro-optical behavior of InGaN/GaN micro-light emitting diodes1, 2, 3 (µLEDs). Indeed, the growing interest to use these devices, especially in display and/or portable modules, is still at its very first years as no proven application can be found in the market today. An important reason is the high amount of µLEDs needed, that have to be efficiently similar to achieve high quality µ-displays4, 5, 6. The design of these latter, is based on the combination of a pixelated array and generally a CMOS active matrix to drive the µLEDs at their best quantum efficiency. Therefore, robust circuit design is a mandatory keystone to ensure a good functioning of these applications. Indeed process fabrication spread, even brought to a minimum, is inherent to any technology. As known and well employed in robust technologies, such as in microelectronics platforms, CMOS designs, are dependent on SPICE7 device models. Some work has been done on devices like organic photodiodes (OPDs) or organic light emitting diodes (OLEDs) to capture through a SPICE model their electro-optical behaviors8, 9, 10. Some efforts towards modelling LEDs behavior can be found in literature. Nevertheless, to our knowledge, these papers generally present an analytical modelling study11, 12, 13 of the device or a TCAD based model14. Therefore it has become urgent and almost mandatory to develop such models for devices like LEDs to simulate, understand and enhance the opto-electrical properties of µ-displays. Based on our latest experimental published data on blue emitting µLEDs15, 16, we report in this paper a complete SPICE model construction. This model aims to a complete description of the electro-optical behavior of µLEDs through a large size range. We demonstrate the use of a simple mathematical resolution technique in the simulation core to calculate the injected carrier density from the ABC model equations. Consequently, the LED external quantum efficiency (EQE) calculus becomes straight forward and is emulated inside the model itself, allowing the accurate determination of the emitted optical power. Furthermore, we show the importance of the statistical approach in this type of modelling to allow the design of robust pixel driving circuits dedicated to µ-display applications. [email protected] 2. LED OPTO-ELECTRICAL BEHAVIOR MODELLING 2.1 The sub-circuit approach We show on figure 1 the equivalent electrical circuit used to model the electrical behavior of the LED. It consists of a SPICE LEVEL 1 ideal diode (D), a series resistor (RS) and a parallel resistor (RP). This sub-circuit has two input pins representing the LED anode and cathode. The light output behavior is described by a virtual current-controlled current source (F). This latter is biased between two nodes VN and VP that are transparent to the final user. The current value (IF) flowing through this virtual source F emulates the optical output power (Popt) of the LED and is defined as: 𝐼(cid:3007) = 𝑃(cid:3042)(cid:3043)(cid:3047) = (cid:3010)(cid:3261)(cid:3254)(cid:3253) (cid:3044) ∙ 𝐸𝑄𝐸 ∙ (cid:3035)(cid:3030) (cid:3090) (1) where ILED,  and EQE are respectively, the current flowing through input pins (anode to cathode), the emitted wavelength (460 nm in our case) and the external quantum efficiency of the LED. The reader will easily recognize the universal constants h, c and q. We will focus on the determination of EQE in section 2.3. Figure 1. Equivalent electrical circuit implemented to model the LED electro-optical behavior. The components RS, RP and D are used to describe the current-voltage characteristic of the LED. The F current source is a virtual user-transparent component to mimic the optical output power of the LED. 2.2 Electrical parameters adjustments and size effect Figure 2 shows current-voltage measurements carried out on vertical square-shaped LEDs of different size, ranging from 10µm to 500µm. In order to adjust our SPICE model to these electrical characteristics, we have chosen to put to nil the internal series resistance of the SPICE LEVEL1 ideal diode D. The value of the implemented external resistor RS has been extracted for each LED size. The whole set permitted to derive an empirical area-dependent relationship equation as follows: 𝑅𝑆 = (cid:3020)(cid:2868) (cid:3020) ∙ (cid:3435)𝑅𝑆(cid:2868) + 𝑅𝑆(cid:3023) ∙ 𝑒(cid:2879)(cid:3080)((cid:3023)(cid:2879)(cid:3023)(cid:2868))(cid:3439) (2) where S0 and S define the reference LED area (500µm width), and the modelled one, respectively. Knowing that S ≤ S0, it is evident that RS will present a greater value for small devices. Furthermore, equation (2) can be interpreted as a sum of two series resistances contributions where the first term (RS0) is fixed and represents the common part. The second term is voltage dependent and is seen as a non-linearity deviation of RS. Indeed, the series resistance is a representation of all the physical obstacles that injected carriers have to overcome to flow through the device. Generally, anode and cathode contacts show a deviation from ideal ohmic behavior when bias is pushed to high values. Process fabrication can also generate defects and recombination centers preventing a smooth carrier flow. For all these reasons the RS parameter value cannot be considered as a fixed value. Finally, parameter V0 is inferred as a threshold voltage from which RS non-linearity begins to act, whereas  is a smoothing adjustment parameter. On the other hand, the value of the parallel resistance RP could not be adjusted for small geometries due to the limitation in our measurement instrumentation. This also demonstrates that our devices have extremely weak leakage currents at low voltage, proving a high process quality. To simplify the model and avoid any dysfunction during simulations, we have decided to give RP a fixed value, satisfying the lowest LED leakage current we have measured. Evidently, this choice underestimates the leakage of large geometries. Nevertheless, a light emitting device is often biased far from the leakage region above its threshold voltage, therefore this modelling hypothesis is acceptable. Beside the RS and RP adjustments, we have tuned the saturation current (IS) and ideality coefficient () parameters in the SPICE LEVEL1 diode model (D) to fit the low leakage and intermediate bias regions, far from the high voltage zone. Both parameter values seemed to be geometry independent. Figure 2. Current-voltage characteristics of different square LEDs. Width ranges from 10µm to 500µm. Shaded areas show where the SPICE model parameters, RS, RP, IS and  were adjusted. Note that the leakage current saturates due to a limitation in our measurement instrumentation. 2.3 Optical quantum efficiency and ABC model The quantum efficiency of LEDs has thoroughly been described by the well documented ABC model in literature. We have recently suggested a slight modification to take into account size in the ABC model9. This modification principally affects parameter A related to the Shockley Read Hall (SRH) non-radiative recombination. Indeed, we have demonstrated that A is proportional to the ratio perimeter/surface (P/S) of the LED. Parameters B and C, respectively linked to radiative and Auger non-radiative recombinations, are shown to be geometry independent. They represent the epitaxy material quality. Taking into account the electron charge density q and the quantum well thickness t, the current density JLED and the internal quantum efficiency (IQE) of the LED are expressed as follows: and 𝐽(cid:3013)(cid:3006)(cid:3005) = 𝑞 ∙ 𝑡 ∙ (cid:4672)𝐴 ∙ (cid:3017) (cid:3020) ∙ 𝑛 + 𝐵 ∙ 𝑛(cid:2870) + 𝐶 ∙ 𝑛(cid:2871)(cid:4673) 𝐼𝑄𝐸 = (cid:3003)∙(cid:3041)(cid:3118) (cid:3002)∙ (cid:3265) (cid:3268) ∙(cid:3041)(cid:2878)(cid:3003)∙(cid:3041)(cid:3118)(cid:2878)(cid:3004)∙(cid:3041)(cid:3119) (3) (4) Here, n is the injected carrier density flowing through the device at a given voltage. Knowing the light extraction efficiency (estimated LEE ≅ 0.14 by ray tracing simulations), the external quantum efficiency EQE is then determined from IQE in this way: 𝐸𝑄𝐸 = 𝐼𝑄𝐸 ∙ 𝐿𝐸𝐸 (5) We propose in this SPICE modelling methodology to integrate parameters A, B and C as input entities of the device, at the same level as its geometrical inputs, namely its perimeter P and surface S. It is then obvious that if the injected carrier density n is known, one will not have any difficulty to estimate the quantum efficiency of the LED. Nevertheless, any SPICE simulation only allows an electrical current determination. Our original submitted approach is then to derive instantaneously the value of n from the simulated electrical current. Indeed, equation (3) shows that for a given simulated current density JLED, the calculus of n is brought to the roots determination of a 3rd degree polynomial (𝑎 ∙ 𝑥(cid:2871) + 𝑏 ∙ 𝑥(cid:2870) + 𝑐 ∙ 𝑥 + 𝑑 = 0). This resolution is not straightforward as one might think. Indeed, depending on the polynomial coefficients a, b, c and d different mathematical resolution methods have been proposed. We have decided to implement in the core of the LED model library, an easy and common technique called the Cardano's method17, 18. This method allows a rapid calculus of all real solutions if the 3rd degree polynomial coefficients are real, which is the case in the ABC model. Therefore, for each applied voltage value, the current density is simulated using the adjusted electrical LED SPICE model, and the resolution of equation (3) to determine the carrier density n is simultaneously carried out. Subsequently, the quantum efficiency is determined using equations (4) and (5). 2.4 A statistical approach To provide a better robustness of the LED model, we have chosen to introduce a statistical modelling approach to allow the use of Monte Carlo SPICE simulations. At first approximation, the choice has been made to principally affect the injected carrier density into the LED. Therefore, we implemented Gaussian distributions on each recombination parameter A, B and C. Average parameter values have been fixed to those used in the typical model (from our latest experimental results9) and a given standard deviation around 15% has been fixed for all 3 parameters. This random choice has no particular reason, except the aim to apprehend the quantum efficiency spread due to the parameters cited above. Thus, for each random Monte Carlo draw of A, B and C, the same calculus of n, IQE and EQE will be carried out as stated in section 2.3. In this special statistical configuration, it is clear that the simulated LED current will not change from a draw to another, as no variation has been introduced in the LED model parameters. Evidently, the further step would be, to evaluate the mutual effects of current variation, and recombination parameters distributions together on the LED quantum efficiency spread. 3.1 Electrical characteristics 3. SIMULATION RESULTS We show on figure 3, the simulated current-voltage characteristics superimposed to measurement ones for the different LED sizes. It is clear that the implemented SPICE model describes well the electrical behavior, on a large voltage range. As discussed above, the current at low voltage (0 ≤ V ≤1) is a little bit underestimated for large geometries due to the choice of a constant value for RP accounting for LED leakage (not shown on figure 3). Despite this little discrepancy, the developed sub-circuit electrical SPICE model approach shows a high quality adjustment over the large variety of LED sizes. Figure 3. SPICE model simulated current-voltage characteristics (lines) versus experimental data (symbols) of different LED sizes. A good adjustment is observed on a large voltage range for all geometries. 3.2 Quantum efficiency Figure 4 shows external quantum efficiency EQE versus current density characteristics for all studied LED geometries. For the sake of clarity, experimental curves (figure 4 (a)) and simulated ones (figure 4 (b)) are not presented on the same graph. The first observation to come out is that the experimental EQE aspect is well represented by simulation. Indeed, first we observe the optical threshold shifting towards higher current density levels for small geometries. Moreover, the maximum efficiency EQEmax drops when size diminishes as perceived on experimental data. Additionally, the common sinking of EQE known as the ''droop'' effect is also present on simulated graphs. Yet, looking closely to figure 4, some differences exist between measurements and simulations. Indeed the EQE optical threshold appears at lower current densities for experimental data. Besides, the EQEmax drop versus LED size is somewhat exaggerated in the simulated EQE graphs. These differences are mainly due to the ABC model hypothesis, assuming a current-density JLED relationship to carrier density n expressed as stated in equation (3), which in reality is an empirical model based approximation. Nevertheless, for SPICE simulations it is sufficient to have the essence of the electro-optical behavior of InGaN/GaN based LEDs. Another point worth noting is the break appearing on simulated curves whatever the LED size. This is inherent to the resolution method we have used to determine the carrier density n. Indeed, Cardano's method for the roots determination of a 3rd degree polynomial, is highly dependent on the polynomial coefficients. This dependence implies 2 kinds of calculus subject to coefficient values. Knowing that one coefficient is directly related to current density JLED, this break shows the transit between both styles of the polynomial root resolution. Figure 4. External quantum efficiency versus current density for different LED sizes: experimental data behavior shown in symbols (a) are well described by the simulated SPICE model (b). Some differences are yet observable between measurements and simulation, due to the ABC model approximation on the LED current modelling. 3.3 Spread and recombination parameters effect To apprehend the effect of recombination mechanisms on the LED optical behavior, we have run Monte Carlo simulations for a 50µm width LED using the statistical approach we implemented in our SPICE model. We have focused our attention on the spread of EQEmax. Simulations results for 1000 runs are shown on figure 5. The first observation is that the maximum external quantum efficiency shows a Gaussian like distribution (figure 5 (a)) centered on an average value 〈𝐸𝑄𝐸(cid:3040)(cid:3028)(cid:3051)〉 = 0.11. with a spread deviation at 3𝜎 = 0.02. To seek the effect of the random A, B and C value draws on quantum efficiency, all extracted EQEmax values are presented on three different cloud graphs versus all drawn values of recombination parameters, respectively. Even though, no special correlation between recombination parameters has been used (random draws of A, B and C), we observe awaited evolutions of EQEmax versus each related recombination parameter. Indeed, we observe a tendency of having a better quantum efficiency when a higher parameter B value is drawn (figure 5 (c)), whatever are A and C values. This trend is also observable when A (figure 5 (b)) or C (figure 5 (d)) show lower values independently of the two other parameters. At last, we think that this kind of cloud representations of Monte Carlo simulations can also help understand and quantify the spread in quantum efficiency of µLEDs dedicated for display applications. Indeed high resolution µLED arrays contain a huge number of single µLEDs that will have to be controlled efficiently by an active matrix which also has its own spread. Therefore an InGaN/GaN LED SPICE model is mandatory to evaluate and design robust µLED displays. Figure 5. 1000 run Monte Carlo simulation results of a 50µm width LED. (a) Histogram Gaussian like distribution of EQEmax. Cloud graphs in (b), (c) and (d) represent EQEmax versus SRH, radiative and Auger recombination related parameters A, B and C, respectively. Awaited trends of EQEmax are observed versus each parameter showing the robustness of our implemented statistical approach in the SPICE model. 4. CONCLUSION An InGaN/GaN LED SPICE model taking into account known size effects is demonstrated. A sub-circuit description of the LED permitted an accurate modelling of the electrical behavior. The use of the ABC model recombination related parameters with a simultaneous mathematical resolution during current simulations allowed a precise determination of the external quantum efficiency versus current density. Moreover, a first simple statistical approach based on Gaussian distributions applied to recombination parameters show that this kind of SPICE modelling can help understand and ameliorate pixel designs dedicated to high resolution displays that are aimed to applications for virtual, augmented or mixed reality. ACKNOWLEGEMENTS The authors acknowledge partial fundings from the European Union's Horizon 2020 VOSTARS research and innovation programme under grant agreement No 731974 and H2020 HILICO European project (H2020- JTI-CS2-2016-CFP04-SYS- 01-03, Grant No. 755497). REFERENCES [1] Olivier, F., Tirano, S., Dupré, L., Aventurier, B., Largeron, C., Templier, F., "Influence of size reduction on the performances of GaN-based micro-LEDs for display application", J. Lumin., 191, pp.112-116, (2017). [2] Tian, P., McKendry, J. J. D., Gong, Z., Guilhabert, B., Watson, I. M., Gu, E., Chen, Z., Zhang, G., Dawson, M. D., "Size dependent efficiency and efficiency droop of blue InGaN micro-light emitting diodes", Appl. Phys. Lett., 101, 231110, (2012). [3] Badock, T. J., Ali, M., Zhu, T., Pristovsek, M., Oliver, R. A., Shields, A. J., "Radiative recombination mechanisms in polar and non-polar InGaN/GaN quantum well LED structures", Appl. Phys. Lett., 109, 151110, (2016). [4] Day, J., Li, J., Lie, D. Y. C., Bradford, C., Lin, J. Y., Jiang, H. X., "III-Nitride full-scale high-resolution microdisplays", Appl. Phys. Lett., 99, 031116 (2011). [5] El-Ghoroury, H. S., Alpaslan, Z. Y., "Quantum Photonic Imager (QPI): A New Display Technology and its Applications", Proc. of the 21st International Display Workshop (IDW'14), Niigata, Japan (2014). [6] Templier, F., Dupré, L., Tirano, S., Marra, M., Verney, V., Olivier, F., Aventurier, B., Sarrasin, D., Marion, F., Catelain, T., Berger, F., Mathieu, L., Dupont, B., Gamarra, P., "GaN-based Emissive Microdisplays: A very Promising Technology for Compact, Ultra-high Brightness Display Systems", SID Int. Symp. Digest Tech. Papers, 47(1), pp. 1013-1016, (2016). [7] Nagel, L. W, Pederson, D. O., "SPICE (Simulation Program with Integrated Circuit Emphasis)", Memorandum No. ERL-M382, University of California, Berkeley, (1973). [8] Daami, A., Vaillant, J., Gwoziecki, R., Serbutoviez, C., "A complete SPICE subcircuit-based model library for organic photodiodes", Solid-State Electronics, 75, 81-85, (2012). [9] Cummins, G., Underwood, I., Walton, A., "Electrical characterization and modelling of top-emitting PIN- OLEDs", Journal of the Society for Information Display, 19, 360-367, (2011). [10] Lee, S.-G., Hattori, R., "Physics-based OLED analog behavior modeling", Journal of Information Display, 10, 101-106, (2009). [11] Prajoon, P., Nirmal, D., Anuja Menokey, M., Charles Pravin, J., "A modified ABC model in InGaN MQW LED using compositionally step graded Alternating Barrier for efficiency improvement", Superlattices and Microstructures, 96, 155-163, (2016). [12] Lin, G.-B., Meyaard, D., Cho, J., Schubert, E. F., Shim, H., Sone, C., "Analytical model for the efficiency droop in semiconductors with asymmetric carrier-transport properties based on drift-induced reduction of injection efficiency", Appl. Phys. Lett., 100, 161106, (2012). [13] S. Karpov, "ABC-model for interpretation of internal quantum efficiency and its droop in III-nitride LEDs: a review" Opt. Quantum Electron., Vol. 47, pp.1293-1303, (2015) [14] Bulashevich, K. A., Konoplev, S. S., Karpov, S. Y., Photonics, 5(4), 41, (2018). [15] Olivier, F., Daami, A., Licitra, C., Templier, F., "Shockley-Read-Hall and Auger non-radiative recombination in GaN based LEDs: A size effect study", Appl. Phys. Lett., 111, 022104, (2017). [16] Daami, A., Olivier, F., Dupré, L., Henry, F., Templier, F., "Electro-optical size-dependence investigation in GaN micro-LED devices", SID Int. Symp. Digest Tech. Papers, 49(1), 790-793 (2018). [17] Nickalls, R. W. D., (1993), "A new approach to solving the cubic: Cardan's solution revealed" The mathematical Gazette, 77(480), 354-359, (1993). [18] Zucker, I. J., "The cubic equation: a new look at the irreducible case", The Mathematical Gazette, 92(52), 264- 268, (2008).
1708.07261	2	1708	2017-09-03T12:06:23	Machine Learning to Analyze Images of Shocked Materials for Precise and Accurate Measurements	[ "physics.app-ph" ]	A supervised machine learning algorithm, called locally adaptive discriminant analysis (LADA), has been developed to locate boundaries between identifiable image features that have varying intensities. LADA is an adaptation of image segmentation, which includes techniques that find the positions of image features (classes) using statistical intensity distributions for each class in the image. In order to place a pixel in the proper class, LADA considers the intensity at that pixel and the distribution of intensities in local (nearby) pixels. This paper presents the use of LADA to provide, with statistical uncertainties, the positions and shapes of features within ultrafast images of shock waves. We demonstrate the ability to locate image features including crystals, density changes associated with shock waves, and material jetting caused by shock waves. This algorithm can analyze images that exhibit a wide range of physical phenomena because it does not rely on comparison to a model. LADA enables analysis of images from shock physics with statistical rigor independent of underlying models or simulations	physics.app-ph	physics	Machine Learning to Analyze Images of Shocked Materials for Precise and Accurate Measurements Leora Dresselhaus-Cooper,a,b Marylesa Howard,c Margaret C. Hock,c B. T. Meehan,c Kyle Ramos,d Cindy Bolme,d Richard L. Sandberg,d Keith A. Nelsona,b a Department of Chemistry, Massachusetts Institute of Technology, Cambridge, MA, 02139, USA b Institute for Soldier Nanotechnology, Massachusetts Institute of Technology, Cambridge, MA, 02139, USA c National Security Technologies, North Las Vegas, NV, 89030, USA d Los Alamos National Laboratories, Los Alamos, NM, 87545, USA [Accepted to the Journal of Applied Physics] A supervised machine learning algorithm, called locally adaptive discriminant analysis (LADA), has been developed to locate boundaries between identifiable image features that have varying intensities. LADA is an adaptation of image segmentation, which includes techniques that find the positions of image features (classes) using statistical intensity distributions for each class in the image. In order to place a pixel in the proper class, LADA considers the intensity at that pixel and the distribution of intensities in local (nearby) pixels. This paper presents the use of LADA to provide, with statistical uncertainties, the positions and shapes of features within ultrafast images of shock waves. We demonstrate the ability to locate image features including crystals, density changes associated with shock waves, and material jetting caused by shock waves. This algorithm can analyze images that exhibit a wide range of physical phenomena because it does not rely on comparison to a model. LADA enables analysis of images from shock physics with statistical rigor independent of underlying models or simulations. I. INTRODUCTION Images that are acquired to characterize spatial variation in materials may show a wealth of information that cannot be obtained in other ways. Careful analysis of large datasets is often required for precise determination of material variability with meaningful uncertainties. In many settings, image acquisition is made difficult by factors including scattering of light, inhomogeneity of the light field that is used, small changes in material refractive index (i.e. low contrast), and other challenges. Interpretation of the resulting images can be strongly enhanced with advanced image processing approaches. ~ 1 ~ Image analysis algorithms have been widely developed for many different fields including satellite photography1 and biological imaging.2,3 These algorithms have used combinations of machine learning and artificial intelligence to extract precise locations for different types of objects. Image segmentation (also called image classification) has been used extensively to this end. Image segmentation is conducted using machine learning algorithms that separate (segment or classify) an image into different regions (classes). Supervised segmentation methods require the user to define a set of classes that correspond to features of interest (typically based on user insight about the image content) in the image.4 Several fields of science have already adopted variations on segmentation to appropriately analyze images. Satellite tomography uses a partial differential equation (PDE) approach to locate sharp changes in the intensity gradients caused by the edges of trees, houses, rivers, etc. in order to identify image features without user- provided information (unsupervised).1 Disease diagnostics use segmentation on microscopy images of samples from patients to separate benign cells from invasive bacteria in order to quickly determine if a patient has a bacterial illness.2,3 Images of shock waves and other high strain rate deformations are often difficult to interpret due to irreproducibility of successive measurements, unpredictable content in the images including spatially variable material responses and debris, spatial variations of the imaging light, and other causes. Because shock waves are high amplitude compression waves that move materials far from equilibrium, they usually cause irreversible damage such that each measurement must be made on a different sample or sample region, increasing variability from one measurement to the next. Reduction of shot-to-shot noise has presented challenges in engineering detectors with sufficiently high frame rates to obtain interpretable images with sufficient time resolution to measure shock-induced dynamics.5–8 A single image recorded at a specified time may capture a substantial region that has been traversed by a rapidly propagating shock wave (on the order of kilometers per second), revealing very wide-ranging features. 9–12 Shock-induced transformations can include defect formation, fracture, phase transitions and chemical reactions whose spatial locations and extents in an image can reveal material properties and dynamics.5,10,13,14 The combination of ultrafast imaging methods that capture spatially varying material responses with advanced image processing algorithms that reliably define the different regions under observation can further our understanding of complex shock-induced responses.5,10,13,15–21 In the field of shock physics, many images are manually segmented to extract material features from images. Most image analysis uses Fourier filtering, which suppresses high-frequency noise, but also diminishes the high-frequency features inherent to shock waves.22,23 Other image analysis algorithms rely on comparison of image features to theoretical models from fluid dynamics that describe the shock.24,25 These algorithms can only be used on images collected for shocks that have well-characterized spatial ~ 2 ~ dynamics. Images of shock waves or shock-induced responses that are not sufficiently well characterized to compare to a computational model require a general image analysis tool. Pribosek recently published a segmentation algorithm to perform unbiased image analysis using a PDE approach, which targets low- contrast images. This method has not been widely used within the field, as it is not optimized for high- noise images.26 Established image segmentation algorithms cannot give accurate results due to high noise content in the images and large overlap in pixel intensities between the different classes, i.e. the different shocked states. In this paper, we present the application of a recently developed image segmentation algorithm27 to the analysis of images of shocked materials. Conventional segmentation methods were attempted unsuccessfully for some of the images, as described in the Supplemental Information. The algorithm is unique within machine learning in its ability to segment images with high noise and low variation between classes. For the first image from a series collected from converging shocks in water28, we explain in detail the user-provided information and the algorithm processing routine through which a shock position is determined. We further illustrate the process for each of the subsequent images of the series. Shock positions with uncertainties are extracted from the set of images, and are used to calculate the presumed shock velocities with their corresponding uncertainties for the entire series. To show the wide applicability of this algorithm for images that are difficult to analyze, we present two additional images with many classes and several types of aberrations.18 The analysis demonstrates the algorithm's ability to analyze different types of images in order to make precise and accurate measurements. II. ANALYSIS OF CONVERGING SHOCK WAVES IN WATER WITH SEGMENTATION We describe statistical image segmentation that locates boundaries by sorting pixels from images into specific features (classes) based on the pixel intensity distributions. Supervised variants of statistical segmentation, like LADA, begin with the user providing a map to define the regions that are most characteristic of each class. LADA accounts for variation of intensities within a class by using both the intensity of each pixel and the distribution of intensities among surrounding pixels to sort the pixels into appropriate classes. The size of the surrounding region defines "local" for the algorithm, and can be set by the user with parameters d and n, based on the distance scale over which significant intensity changes are observed. In contrast, as described in the Supplementary Information, some global methods create a histogram with the intensities of all of the pixels defined by sparse training data for each class.4 The intensity distributions created from the well-defined regions of the image are then used to assign each ~ 3 ~ pixel in the image to the class which best defines its intensity. Intensity variation within classes, or overlapping distributions for the intensities of different classes can cause the algorithm to misidentify large regions of the image. To describe the mathematical process performed in LADA, we present its analysis of the first image from a sequence showing focusing shock wave propagation in water (Figure 1a). Images displaying real-time convergence of shock waves in this work were collected with a recently developed shock technique described fully elsewhere.10,29,30 These laser-driven shocks were generated by the interaction of a ring- shaped beam with a thin soft sample. Placement of the sample between two thick glass wafers was predicted to confine the shock wave within the sample layer, allowing for imaging to capture the wave propagation.10 Multi-frame imaging results similar to the ones shown in this work (using LADA) have indicated that further wave interactions may be present, suggesting possible multi-wave dynamics, as are being further investigated at this time.28 In our analysis of the strongest features evident in images of converging shocks, we demonstrate the use of analysis of variance (ANOVA) and maximum likelihood estimator (MLE) p-value maps to precisely calculate the boundary location uncertainties from the segmentation. The LADA measurements of shock position are then compiled from the entire sequence and are used to calculate the shock velocities and the uncertainties are propagated through the calculations. The results demonstrate the utility of this segmentation method, as quantitative determination of the shock properties from the images is difficult without the precision afforded by LADA. ~ 4 ~ Figure 1. (a) The first of six ultrafast images measuring converging shocks travelling in a multi-layered target (water between glass). (b) The training data form a user-specified map of a priori, known regions for the different classes in the image. Each class is represented by a distinct color. The dark regions correspond to user-unspecified regions to be segmented by the program. (c) The segmented image (d = 25, n = 5) showing the locations for each class. The white regions are an added class that corresponds to pixels that did not have any local training data when the image was restricted by parameter d. (d) The ANOVA p-value map of the segmented pixels. (e) The maximum likelihood estimator (MLE) p-value map for the segmented image, thresholded at 5%. For this image, the 1280 × 960 pixels are either black or white based on the thresholding, as mentioned in the text. Use of LADA first requires the user to provide training data, which gives a priori knowledge of the image features of interest to the algorithm. Training data form a map that can be overlaid on top of the image to show the approximate class locations in the image (Figure 1b). As a local algorithm, LADA requires the training data to provide regions of the image that are representative of both characteristic pixel intensities and positions for each class. LADA best analyzes images consisting mainly of classes that may be reliably definable by the user, and whose unknown regions are localized around the boundaries between classes. LADA analysis provides the locations of the boundaries. Figure 2. An illustration of the method by which LADA selects local training data for the image. Training data for an entire image (left) has four classes, shown in orange, green, blue and brown. The undefined region for the segmentation is shown in grey. For the red pixel p, a circle of radius d defines the set of pixels considered for the local training data (right). Within the local training data, the n nearest pixels used by the method are highlighted with stripes. ~ 5 ~ As shown with a cartoon in Figure 2, LADA also requires the user to provide two parameters, d and n, which define "local" for the algorithm. The value of d sets the distance within which the distribution of pixel intensities may be considered in assigning the class for a given pixel. Large values of d allow all of the defined classes to be considered for every pixel, while low values of d limit the number of classes available for comparison to those within the radius d around the pixel. Intensity histograms are assembled using the n nearest pixels in every class within the radius d. If there is a tie between pixels that are the same distance from p, the n points are selected using lexicographical ordering. High input values for both of these parameters bias the segmented output toward a global result. The optimal value of d will balance selecting enough local classes with avoiding the inclusion of training data that do not represent the local behavior around that pixel. That helps ensure that pixels that belong to one class are not misassigned to another class with a similar distribution. The optimal value of n typically corresponds to the number of pixels over which the local intensity variation within the class is homogenous. This helps avoid unwanted fluctuations within the histograms that are constructed. Pixels for which the radius d includes only one region are auto-assigned to that region. For a class to be considered local, we require at least three training pixels within radius d to allow us to gather its intensity information statistically. If a pixel does not have at least three training pixels of any local user-defined class, it is assigned to a bonus class, indicating uncertainty due to sparse training data. To summarize, the value of d can be set to affect the number of classes being considered, while the value of n determines the number of pixels compiled into each histogram. Once the user inputs have been collected, the algorithm locally segments the pixels into classes. The algorithm progresses through all of the pixels within the image, to assign each one to the class that is most probable, based on both its intensity and position. Uncertainty in class assignment may still be present due to incorrect training data, insufficient training data, or random noise caused by the camera or experiment that is manifested in the image. For each pixel, the algorithm crops the image within a radius d and for each class specified in the training data creates a histogram using up to n pixels. If the value of n exceeds the number of pixels in a given class within the cropped image, the histogram is formed by only the pixels that are within the cropped image. Each histogram is fitted to a Gaussian curve. The intensity of the pixel is then compared against the Gaussian distributions for each local class, and is assigned to the class with the highest probability for that intensity. The final segmentation of Figure 1a is given in Figure 1c, using d = 20 and n = 10. Segmentation uncertainty is measured for each pixel in two different ways, which cover the two most common causes of imprecise segmentation. Uncertainty can arise when multiple local classes have broadly overlapping intensity histograms, in which case some pixels with similar intensities may fall into ~ 6 ~ more than one class and there is a risk of misidentification. The uncertainty caused by overlapping distributions from local classes is characterized by analysis of variance (ANOVA),4 as shown in Figure 1d. Alternatively, uncertainty can arise when pixel intensities yield a low probability of belonging to any of the local classes specified in the training data. This uncertainty is measured by compiling an MLE uncertainty map1,4 (α = 0.05) to describe the p-values that in our work are thresholded at 5% (Figure 1e). ANOVA measures the extent of overlap among all of the local histograms there were constructed for the assignment of a given pixel. The p-values for ANOVA range from 0 to 1, with ascending values indicating a higher degree of overlap among the classes defined by d and n to be local within the training data. Pixels with high ANOVA values have poor differentiability among the available local classes. ANOVA maps that show high overlap among classes can be indicative of either non-optimized parameters for the user-inputs or regions of the image that cannot be precisely segmented. In determining the optimal parameters for LADA segmentation, ANOVA values can often be reduced by modifying the training data and the values of d and n. High ANOVA values often indicate that the image requires smaller values of d and n to correctly define the local classes according to the length scale over which intensity varies within a class. Inaccurate training data that include pixels from adjacent image features result in a local histogram for the class that is not representative of the a priori knowledge from the user. In those cases, two overlapping Gaussian fits of the histograms can show unfitted distributions that are both bimodal, representing one mode from each class. Low local contrast between image features covering a part of the image that are not well measured by LADA can be seen as regions for which the ANOVA values cannot be reduced by changing the user-inputs. The white regions in Figure 1d are areas that were assigned by the user and for which no ANOVA value was computed. Regions of the image that are not well-defined by any of the local classes are shown in Figure 1e with an MLE p-value map thresholded at α = 0.05.31 To compute this error map, p-values are calculated for each pixel based on the probability of that pixel being within the selected class. Calculated p-values are between 0 and 1, with higher values corresponding to high confidence for the segmentation. The uncertainty map in Figure 1e is thresholded to only have a black pixel corresponding to a p-value that is ≤0.05, or for any pixel that was placed into a bonus class. Figure 1e shows a random distribution of dark pixels throughout the image that corresponds to noise in the original image. A high density of dark pixels – located at a boundary between classes – indicates the uncertainty from the segmentation between the local classes for that region. The map of regions with low-confidence segmented pixels best describes the uncertainty in measurements of the boundaries between the adjacent classes and is used for propagation of error in subsequent calculations. ~ 7 ~ Appropriate results for the images in this paper were acquired by iteratively running the algorithm to optimize the user-input parameters consistent with a priori knowledge of the experimental system and the underlying science. All images shown in this paper were initially segmented using sparser training data than are shown here and high values for d and n to allow the algorithm to discern boundaries that are difficult to see by eye. After each iteration, we refined the training data and the local parameters based on the boundaries and sharp features identified by the method. The procedure ended when there were no significant changes between iterations and our a priori knowledge does not suggest any further refinements to the training data or the local parameters. Our primary interest with the LADA results was in locating the position of the leading edge of the dark feature we interpreted as the shock which, in Figure 1, is the innermost boundary. We defined the two regions on either side of this boundary, "shock" and "unshocked," as uniform arrays of 1 and 0 values, respectively. The positions of each pixel of the shock front was measured by taking the gradient of the segmented image to locate the edges between the classes.32 The resulting points were input into a circle fitting algorithm33 to define the center of the shock front. To measure the shock radius, we transformed the segmentation result into polar coordinates, and divided the shock ring into 60 azimuthal sections to make the measurement statistically. The position corresponding to a maximum value for the second derivative of the segmentation result in each azimuthal section was used to measure a distribution of shock radii for each image. The mean values are shown in Figure 4, with two times the standard deviation (95% certainty) shown by the error bars. The uncertainty that the algorithm had in locating the boundary was obtained from the MLE plot. That uncertainty was collected using 60 azimuthal slices to measure the position with the highest weighted uncertainty around that boundary. The resulting distributions then gave the mean and standard deviations for the position of highest uncertainty for each azimuthal slice. The two standard deviations about the mean position of highest uncertainty was extracted from the distribution for each image, corresponding to the positions of highest LADA uncertainty about the shock front. ~ 8 ~ Figure 3. LADA segmentation of the entire image sequence for a converging shock wave in water, using parameters d = 20 and n = 10. For each of the images in the sequence, a bonus class was generated due to uncertainty in assigning sufficient training data for that region of the image. Initially, a shock wave converges into a central region, then it diverges outwards. At the point of convergence, geometric instabilities create a Mach reflection,34 observed at 35 ns. After the shock diverges, additional features become evident beginning at 50 ns, which continue after the end of this image sequence.10 The LADA analysis of all of the images shown in Figure 3 was used to determine the shock position for all six images in the sequence. The shock positions, calculated as described above, were located for all time delays, and the average shock velocities were calculated with 𝑈𝑠(𝑡𝑛) = 𝑟𝑛 − 𝑟𝑛−1 𝑡𝑛 − 𝑡𝑛−1 ~ 9 ~ (1) for each pair of successive images, based on the measured time delay 𝑡𝑛 and shock position 𝑟𝑛 measured from segmentation and subsequent analysis. The average shock velocity US between each pair of images was calculated for all 60 azimuthal sections around the shock ring, to create a distribution of shock velocities. The plots in Figure 4 show the shock positions and shock velocities from the image sequence shown in Figure 3. For all of the data, the points and their error bars show the statistical values for position and velocity for each image, collected from the segmentation result. The positions and velocities that had the highest uncertainty in the segmentation are shown in the shaded gray and blue regions, respectively, behind the curves. For all of these values, two times the standard deviations (2σ) are shown, corresponding to the intervals of 95% confidence, consistent with the MLE confidence interval. The difference between the error bars and the shaded uncertainty bands highlight the unique separation of uncertainty that LADA provides. The azimuthal statistics for the shock positions and velocities from the segmentation result allow us to see variation in the shock front position. Variation in the position is caused by ellipticity in the shock ring (systematic error), and mis-assigned pixels (random and systematic error). Statistical variation in the velocity values include systematic error from variation in the shock velocity around the ring, which is often seen in converging shock waves with geometric distortions.35 The standard deviations show the extent of variation, but cannot determine the asymmetry of uncertainty, which can cause the mean value to be skewed. Deviation between the bands of MLE uncertainty and the error bars demonstrate results for which the mean value calculated from the segmentation result is skewed by uncertainty from the method's ability to segment that region of the image. ~ 10 ~ Figure 4. Measurements of shock positions and calculated velocities for the entire series of images shown in Figure 3. The measurements were done using the LADA analysis with d = 20 and n = 10 for each image in the sequence (analysis shown in Figure 3). The points and error bars show the mean and 2σ values for the shock position and velocities, calculated statistically from azimuthal slices around the shock ring. Shaded gray and blue regions behind the curves show the positions and corresponding velocities for which the algorithm had ≤5% confidence, as derived from the MLE plots. The temporal separation between the images was measured using an oscilloscope with 250 ps resolution, giving a separation of 5 ± 0.25 ns between frames. Temporal uncertainty was not included for this calculation, as the uncertainty in velocity is dominated by the much larger uncertainty in position. For experiments that do not have well-defined temporal resolution, an RMS average may be taken to account for the temporal and spatial uncertainty. New understanding of this experiment suggests that multi-wave interactions preclude the use of the principal Hugoniot to infer shock pressures from the velocity of the shock-induced image features. For other systems, the velocity measurements from LADA can be input to the equation of state in order to infer the shock pressure and uncertainty at each time. ~ 11 ~ III. EXTENDING LADA TO FURTHER COMPLICATED SHOCK IMAGES In the water image sequence above, we have explained the LADA methodology for the segmentation of converging waves in targets containing water for our experimental geometry. This analysis demonstrates the utility of LADA in finding the boundaries between the shock wave and unshocked material in order to measure shock position and velocity. We now demonstrate LADA's ability to separate locally varying images for which the boundaries between classes are difficult to identify by eye. Figure 5a shows an image captured with a 400-nm laser pulse using shadowgraphy (second derivative of density) of a converging shock wave transducing into an RDX crystal embedded in a polymer. Figure 6a shows an X- ray phase-contrast image of shock-induced void collapse causing jetting in acetaminophen. Subsequent figures in the respective sequences present the segmentation for the two different experiments. These images show the utility of the algorithm in extracting scientifically meaningful boundary measurements from the image segmentation afforded by LADA even with high intensity variation within individual classes. ~ 12 ~ Figure 5. The inputs and outputs showing the segmentation of (a) an image with a shock feature that appears mostly in the polymer, but partly in a crystal of RDX. The training data (b) show that there are many classes, describing a complicated experimental geometry. The segmented result using d = 20, n = ~ 13 ~ 15 (c) shows that the segmentation algorithm locates boundaries than the simple inspection. Error maps of (d) ANOVA and (e) MLE plots demonstrate that the image yields considerable uncertainty but can still be segmented reliably. Figure 5 shows the segmentation of an image which we understood to show a shock travelling through our converging shock geometry that transduces from polymer into an RDX crystal upon convergence. This image has many more classes than were seen in Figure 1. The a priori classes based on the experimental conditions are: the laser ring from the drive beam, the shocked polystyrene, the shock wave in the polystyrene, the unshocked polystyrene, the shocked crystal, the shock wave in the crystal, the unshocked crystal, and the region outside of the drive laser ring. The high number of classes for this image and the density of the classes make the image difficult to separate manually. As seen from the segmentation in Figure 5, LADA was able to locate boundaries between classes that were difficult to identify visually. While the classes are difficult to distinguish visually, the ANOVA error is quite low, indicating that the classes are locally quite well defined. The MLE plot demonstrates that the noise within the images generates substantial uncertainties for many but not all of the boundaries. Despite high uncertainty in some regions, the statistical analysis of LADA measures boundaries in Figure 4 that are difficult to discern by eye. This analysis allows for measurement of the features and structural variation of the shape change of the feature we interpret as the shock wave after it transduces into the polymer. Figure 6. Analysis of an image (a) of void collapse and the resultant jetting, as caused by a planar shock in acetaminophen collected with X-ray phase contrast imaging. The training data (b) includes arrows to indicate the original positions of the voids. The training data and parameters d = 24, n = 10 were used to ~ 14 ~ segment the image (c) with high precision, despite the aberrations from ghosting and saturation of the detector (photo bleaching of the scintillator) in the experiment. The error maps of (d) ANOVA and (e) MLE demonstrate the high accuracy of the measurement, despite features that are hard to discern in the original image. Reproduced with permission from J. Phys. Conf. Ser. 500, (2014) p. 142028. Copyright IOP Publishing. Figure 6 illustrates the use of LADA to analyze a phase contrast image from an 80-ps duration synchrotron X-ray pulse that shows a planar shock wave causing acetaminophen to jet as shocked voids collapse.6 This image shows an aperture within which appears an image of the jetting material extending into free space. The three different intensity classes in Figure 6b-c arise from ghosting of images from a series (153 ns inter-frame) caused by the scintillator detector.7 This ghosting causes Figure 5a to show three overlapping images with different intensities, each of which corresponds to an image produced by X-rays from adjacent synchrotron pulses during the jetting event. In this case, segmentation of the image can provide the temporal evolution of the jetting in the material. Analysis of this image (and similar ones) has been difficult, as the unsegmented image (Figure 6a) has boundaries that are challenging to discern with statistically accurate uncertainty. The segmentation with LADA allows for precise location of boundaries from Figure 6a between the different times within the jetting with relative ease. The uncertainty shown in the ANOVA plot demonstrates that the classes are locally distinguishable for most of the image. Figure 6e shows an MLE plot for the segmentation, which shows a case that is not seen in the previous examples. With the exception of the white class, boundaries between each class in the image show high confidence from the thresholded plot. There is a region of low p-values and high uncertainty which is located at the center of the image, in the white class. In this case, high uncertainty corresponds to a region which is defined by the user in the training data. The high uncertainty found from the white class is not an artifact, but is due to lack of local intensity variation in that class. The intensity distribution in the region of high uncertainty is extremely narrow, as the camera was saturated for those pixels. Saturated signal causes the distribution to narrow toward a delta function, which makes any small amount of variation in pixel intensity have extremely low overlap with the local distribution. Knowing which regions of an image gave a false signal due to saturated detectors was important to avoid misinterpretation of the science corresponding to the feature intensities. The same trend was observed in the aperture region, due to lack of variation within the class. Segmentation was completed successfully by artificially introducing noise to the lowest intensities to generate a statistical distribution in order to correctly identify the aperture class. ~ 15 ~ The application of LADA for quantitative analysis of ghosted images similar to Figure 6 has significant potential for use of recently developed pulse sequences in real-time imaging. These pulse sequences include the use of mode 324 at the Advanced Photon Source, which has 11.37 ns between pulses of 80ps duration.36 Use of X-rays for single-shot multi-frame imaging has been limited because there current scintillators do not have sufficient conversion efficiency from X-ray to optical light or decay times from that conversion process. Attempts at multi-frame imaging with these slow detectors causes ghosting and saturation (photo bleaching) by subsequent images.37 Improved methods for extracting data from sub- optically recorded images like the ones shown in this paper provide a possible route for data analysis using detectors near their temporal limits. IV. CONCLUSIONS The results presented here explain the use of LADA to extract quantitative information from images with rich information content about materials as they are shocked. With LADA, the computer uses a priori knowledge from the user to learn the approximate positions and local intensity distributions for each of the features of interest in the image. The algorithm then assigns all of the pixels in the image to the classes that best describes them. LADA outputs the segmented image with boundaries and their uncertainties determined. ANOVA uncertainty is used to optimize the user-input parameters and to identify regions that segmentation is not able to identify. An MLE p-value map is used to determine uncertainty in the local segmentation for each pixel, and is used to extract error bars from measurements made with LADA. We used LADA to determine the shock positions and velocities, including their uncertainties, for an image series showing a shock wave converging and then diverging in water. The generality of the algorithm is shown through the segmentations of complicated images with significant aberrations and variation of intensity across features. Software for LADA is available through individual license. Please contact Marylesa Howard at the Nevada National Security Site: [email protected] Supplementary Material See supplementary material for the results of global segmentation using quadratic discriminant analysis, using the image and training data from Figure 1. ~ 16 ~ Acknowledgements The authors would like to thank Aaron Luttman and Jing Kong for their suggestions and contributions to this work. Figure 6 has already been published by Ramos, et al. as cited in this work.6 The authors would like to acknowledge the Office of Naval Research for funding that supported two of the authors on from this work on grant numbers N00014-16-1-2090 and N00014-15-1-2694. This manuscript has been authored in part by National Security Technologies, LLC, under Contract No. DE-AC52-06NA25946 with the U.S. Department of Energy, National Nuclear Security Administration, NA-10 Office of Defense Programs, and supported by the Site-Directed Research and Development Program. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published content of this manuscript, or allow others to do so, for United States Government purposes. The U.S. Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan). DOE/NV/25946--3183 The authors would like to acknowledge support from the U.S. Department of Energy through LANL's Laboratory Directed Research and Development program. LANL is operated by Los Alamos National Security, LLC for the U.S. Department of Energy under contract DE-AC52-06NA25396. References 1 D. Manolakis and G. Shaw, IEEE Signal Process. Mag. 29 (2002). 2 S. Shen, K. Syal, N. Tao, and S. Wang, Rev. Sci. Instrum. 86, 7 (2015). 3 I. Valmianski, C. Monton, and I.K. Schuller, Rev. Sci. Instrum. 85, 33701 (2014). 4 T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning; Data Mining, Inference, and Prediction (2009). 5 K. Nakagawa, A. Iwasaki, Y. Oishi, R. Horisaki, A. Tsukamoto, A. Nakamura, and K. Hirosawa, Nat. Photonics 8, 695 (2014). 6 K.J. Ramos, B.J. Jensen, A.J. Iverson, J.D. Yeager, C.A. Carlson, D.S. Montgomery, D.G. Thompson, K. Fezzaa, and D.E. Hooks, in J. Phys. Conf. Ser. 500 (2014), p. 142028. ~ 17 ~ 7 S.N. Luo, B.J. Jensen, D.E. Hooks, K. Fezzaa, K.J. Ramos, J.D. Yeager, K. Kwiatkowski, and T. Shimada, Rev. Sci. Instrum. 83, 73903 (2012). 8 R.L. Sandberg, C. Bolme, K. Ramos, Q. Mcculloch, R. Martinez, V. Hamilton, T. Pierce, S. Mcgrane, J.L. Barber, B. Abbey, A. Schropp, F. Seiboth, P. Heiman, B. Nagler, and E. Granados, Microsc. Microanal. 21, 1851 (2015). 9 R.F. Smith, C.A. Bolme, D.J. Erskine, P.M. Celliers, S. Ali, J.H. Eggert, S.L. Brygoo, B.D. Hammel, J. Wang, and G.W. Collins, J. Appl. Phys. 114, 133504 (2013). 10 T. Pezeril, G. Saini, D. Veysset, S. Kooi, P. Fidkowski, R. Radovitzky, and K.A. Nelson, Phys. Rev. Lett. 106, 214503 (2011). 11 S. You, M. Chen, D.D. Dlott, and K.S. Suslick, Nat. Commun. 6, 6581 (2015). 12 A.E. Gleason, C.A. Bolme, H.J. Lee, B. Nagler, E. Galtier, D. Milathianaki, J. Hawreliak, R.G. Kraus, J.H. Eggert, D.E. Fratanduono, G.W. Collins, R. Sandberg, W. Yang, and W.L. Mao, Nat. Commun. 6, 1 (2015). 13 D.H. Dolan and Y.M. Gupta, J. Chem. Phys. 121, 9050 (2004). 14 D. McGonegle, D. Milathianaki, B.A. Remington, J.S. Wark, and A. Higginbotham, J. Appl. Phys. 118, 65902 (2015). 15 A. Schropp, R. Hoppe, V. Meier, J. Patommel, F. Seiboth, Y. Ping, D.G. Hicks, M.A. Beckwith, G.W. Collins, A. Higginbotham, J.S. Wark, H.J. Lee, B. Nagler, E.C. Galtier, B. Arnold, U. Zastrau, J.B. Hastings, and C.G. Schroer, Sci. Rep. 5, 1 (2015). 16 B. Sitalakshmi and A. Vudayagiri, J. Opt. Soc. Am. B 30, 2206 (2013). 17 E. Abraham, K. Minoshima, and H. Matsumoto, Opt. Commun. 176, 441 (2000). 18 R.L. Sandberg, C. Bolme, K. Ramos, Q. McCulloch, J.L. Barber, R. Martinez, M. Greenfield, S.D. McGrane, B. Abbey, A. Schropp, F. Sieboth, P. Heiman, B. Nagler, E.C. Galtier, and E. Granados, in CLEO 2014 Postdeadline Pap. Dig. (Optical Society of America, 2014), p. STh5C.8. 19 J.L. Gottfried, Phys. Chem. Chem. Phys. 16, 21452 (2014). 20 R.J. Hemley, Basic Research Needs for Materials under Extreme Environments (Washington D. C., 2007). 21 J. Noack and A. Vogel, Appl. Opt. 37, 4092 (1998). ~ 18 ~ 22 D. Estruch, N.J. Lawson, D.G. MacManus, K.P. Garry, and J.L. Stollery, Rev. Sci. Instrum. 79, 10 (2008). 23 S. Osher and L.I. Rudint, Siam J. Numer. Anal 27, 919 (1990). 24 R.P. Fedkiw, G. Sapiro, and C.W. Shu, J. Comput. Phys. 185, 309 (2003). 25 J.A. Sethian, J. Comput. Phys. 169, 503 (2001). 26 J. Pribosek, P. Gregorcic, and J. Diaci, Mach. Vis. Appl. 26, 485 (2015). 27 M. Howard, M.C. Hock, B.T. Meehan, and L. Dresselhaus-Cooper, ArXiv: 1707.09030 (2017). 28 L. Dresselhaus-Cooper, J.E. Gorfain, C.T. Key, B.K. Ofori-Okai, S.J. Ali, D.J. Martynowych, A. Gleason, S. Kooi, and K.A. Nelson, ArXiv: 1707.08940 (2017). 29 D. Veysset, A.A. Maznev, T. Pezeril, S.E. Kooi, and K.A. Nelson, Sci. Rep. 6, 1 (2016). 30 D. Veysset, T. Pezeril, S. Kooi, A. Bulou, and K.A. Nelson, Appl. Phys. Lett. 106, 161902 (2015). 31 G. Casella and R.L. Berger, Statistical Inference, Second Edi (Duxbury Advanced Series, Pacific Grove, CA, 2002). 32 S. Tabbone and D. Ziou, Int. J. Pattern Recognit. Image Anal. 8, 537 (1998). 33 W. Gander, G.H. Golub, and R. Strebel, Bit 34, 558 (1994). 34 R.W. Perry and A. Kantrowitz, J. Appl. Phys. 22, 878 (1951). 35 M. Watanabe and K. Takayama, Shock Waves 1, 149 (1991). 36 Argonne National Laboratory, http://www.aps.anl.gov/Accelerator_Systems_Division/Accelerator_Operations_Physics/SRparameters/n ode5.html (n.d.). 37 M. Nikl, Meas. Sci. Technol. 17, R37 (2006). ~ 19 ~ Supplemental Information GLOBAL SEGMENTATION OF IMAGE FROM FIGURE 1 The main text from the paper describes segmentation by a local algorithm that accounts for variation in image intensity across classes. Segmentation of the images shown in Figure 1-2 were attempted using established global segmentations, but the pixels were incorrectly sorted due to overlapping intensity distributions between classes. Figure S–1 demonstrates the unsuccessful segmentation of the image shown in Figure 1 using Quadratic Discriminant Analysis (QDA).1 Figure S–1. The QDA segmentation of the same image analyzed in Figure 1 from the main paper using LADA. The original figure (a) is shown again, with the exact same training data (b) as was shown in Figure 1, and used for the calculations that followed. The global histograms corresponding to the fits from each class are shown (c) and the resulting global segmentation is presented in (d). The QDA analysis presented in Figure S–1 used the same image and training data from the LADA analysis in Figure 1. As a global method, QDA compiled a histogram from all of the pixel intensities in the training data to create the Gaussian fit for the subsequent segmentation. The wide intensity variation within each class in Figure S–1 caused all of the histograms to overlap substantially, as shown in Figure S–1c. Overlapping histograms caused QDA to misidentify many pixels, as they often had high probability of belonging to more than one class. The main paper measures uncertainty from overlapping distributions using ANOVA p-values. Despite the global ANOVA uncertainty, Figure S–1a has visually discernable features that are not fully captured by QDA. In Figure S–1a, globally overlapping distributions arise from classes with similar intensities being spread out across the image. The image shown in Figure S–1a failed to be appropriately segmented by QDA because the presence of non-adjacent classes with similar intensities required the position of each pixel to be considered by the method. A thorough comparison of local and global histograms can be found in the seminal paper on LADA.2 REFERENCES 1 T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning; Data Mining, Inference, and Prediction (2009). 2 M. Howard, M.C. Hock, B.T. Meehan, and L. Dresselhaus-Cooper, ArXiv: 1707.09030 (2017).
1810.03320	2	1810	2018-10-09T11:12:12	Efficient nitrogen-vacancy centers' fluorescence excitation and collection from micrometer-sized diamond by a tapered optical fiber	[ "physics.app-ph" ]	Efficiently excite nitrogen-vacancy (NV) centers in diamond and collect their fluorescence significantly benefit the fiber-optic-based NV sensors. Here, using a tapered optical fiber (TOF) tip, we significantly improve the efficiency of the laser excitation and fluorescence collection of the NV, thus enhance the sensitivity of the fiber-optic based micron-sized diamond magnetic sensor. Numerical calculation shows that the TOF tip delivers a high numerical aperture (NA) and has a high fluorescence excitation and collection efficiency. Experiments demonstrate that using such TOF tip can obtain up to over 7-fold the fluorescence excitation efficiency and over15-fold the fluorescence collection efficiency of a flat-ended (non-TOF) fiber. Such fluorescence collection enhances the sensitivity of the optical fiber-based diamond NV magnetometer, thus extending its potential application region.	physics.app-ph	physics	Efficient nitrogen-vacancy centers' fluorescence excitation and collection from micrometer-sized diamond by a tapered optical fiber Dewen Duan,1, ∗ Guanxiang Du,1 Vinaya Kumar Kavatamane,1 Sri Ranjini Arumugam,1 Yan-Kai Tzeng,2, 3 Huan-Cheng Chang,2 and Gopalakrishnan Balasubramanian1, 4, † 1Max-Planck Research Group Nanoscale Spin Imaging, Max Planck Institute for Biophysical Chemistry,Am Fassberg 11,Gottingen, 37077, Germany 2Institute of Atomic and Molecular Sciences, Academia Sinica, Taipei 106, Taiwan 3Current adress: Department of Physics, Stanford University, Stanford, California 94305, USA 4Center Nanoscale Microscopy and Molecular Physiology of the Brain (CNMPB), Gottingen, 37077, Germany Efficiently excite nitrogen-vacancy (NV) centers in diamond and collect their fluorescence significantly benefit the fiber-optic-based NV sensors. Here, using a tapered optical fiber (TOF) tip, we significantly improve the efficiency of the laser excitation and fluorescence collection of the NV, thus enhance the sensitivity of the fiber-optic based micron-sized diamond magnetic sensor. Numerical calculation shows that the TOF tip delivers a high numerical aperture (NA) and has a high fluorescence excitation and collection efficiency. Experiments demonstrate that using such TOF tip can obtain up to over 7-fold the fluorescence excitation efficiency and over15-fold the fluorescence collection efficiency of a flat-ended (non-TOF) fiber. Such fluorescence collection enhances the sensitivity of the optical fiber-based diamond NV magnetometer, thus extending its potential application region. 1. INTRODUCTION Negatively charged Nitrogen-vacancy (NV) color cen- ters in diamond, as a remarkable solid-state quantum optical systems, has a high optical stability and a long spin lifetime at room temperature1 -- 4, is an attrac- tive candidate for biomarking/imaging5,6, magnetic field sensing7 -- 10, electric field monitoring11 and temperature measuring12 -- 14. In most of these applications, the spin state of the NV center is initialized and detected by light guided through optical components such as objectives, lenses, and dichroic mirrors. The bulky size and poor flexibility of these optical components limit its portabil- ity and using-environment compatibility, thus the appli- cations of the diamond NV-based sensor. Using an op- tical fiber to replace these bulk components will greatly extend the potential application areas of the NV-based sensor. In recent years, some research groups have integrated NV centers in an optical fiber to explore its potential application region. For example, Henderson et al. and Ruan et al. have embedded nanodiamonds into tellu- rite glass to fabricate optical fibers with nanodiamonds inside fibers15,16. Rabeau et al. have deposited a few diamonds onto a fiber end face17. Schroder et al. have placed a nanodiamond containing a single NV center on the end face center of a microstructure fiber to form an in-fiber single-photon source18. Mayer et al. have cou- pled a diamond containing NV centers to a single-mode photonic crystal fiber19. Liebermeister et al. have used a tapered optical nanofiber to collect the fluorescence from a single NV center in a nanodiamond on the surface of the nanofiber20. Patel et al. have fabricated an optical fiber nanotaper and collected over 16% of fluorescence from a single NV center in a nanodiamond wire21. Sim- ilarly, Burek et al. also used a single mode fiber tip and collected over 90% fluorescence from a single color cen- ter (Silicon-Vacancy) in a photonic crystal cavity fab- ricated in a diamond waveguide22. Fedotovet al. have glued a 200 µm and 30 µm sized high NV center den- sity diamond onto the flat end of a multimode fiber and a high numerical aperture photonic crystal optical fiber respectively; they excited the NV centers and collected its fluorescence response through the same fiber in their sensor systems23 -- 26. In NV spins' metrological applica- tions, the collected NV's fluorescence of each measure- ment is directly linked to its minimum detectable vari- ation; therefore, enhanced collection efficiency directly provides a better sensitivity27. Among the aforemen- tioned methods, a tapered nano-fiber is an efficient way to collect the diamond NV-center fluorescence. However, the enhancement does not scale well for micron-sized di- amond. On the other hand, attaching the diamond to the flat end of an optical fiber is easy to achieve, but it offers only a limited excitation and collection efficiency significantly lower than using a high numerical aperture (NA) objective. Here, we demonstrate a method to efficiently excite NV centers in micrometer-sized diamond crystal and col- lect its emission fluorescence through the same fiber. Our method is based on using a quasi-adiabatic tapered opti- cal fiber (TOF) tip made from a multimode optical fiber (shown in Fig.1(a)) to concentrate the excitation laser on to the diamond, at the same time, to collect the flu- orescence from the diamond with a high efficiency. The TOF tip is equivalent to a high NA objective on the op- tical fiber. Numerical calculations show that such TOF tip delivers a ultra-high NA larger than 1 and has a high fluorescence excitation and collection efficiency. Experi- ments demonstrate that when exciting and collecting flu- orescence from a ∼5-µm diamond, using such TOF-fiber tip can achieve over 7-fold the fluorescence excitation ef- 8 1 0 2 t c O 9 ] h p - p p a . s c i s y h p [ 2 v 0 2 3 3 0 . 0 1 8 1 : v i X r a ficiency and up to 15-fold the fluorescence collection ef- ficiency of a flat-ended (non-TOF) fiber (table.I). Com- paring Optically Detected Magnetic Resonance (ODMR) scanning experiments demonstrated that this TOF tip can boost the magnetic field sensitivity of a micro-sized NV sensor. In a direct current magnetic field measur- ing, a TOF based ∼7.9-µm diamond NV-magnetometer achieve 1/28 the sensitivity value of a non-TOF based ∼11.4-µm diamond NV-magnetometer, reach a sensitiv- √ ity of 180nT/ Hz. 2. PRINCIPLES AND METHODS 2 orescence entered the thin TOF tip within its maximum acceptance angle into the core of the multimode fiber (Fig.1(a) and Fig.2.(a)). This nearly adiabatic process enhances the NV centers' fluorescence excitation and col- lection. Calculated in an ideal geometry ray optics, the TOF tip's fluorescence collection efficiency can reach ap- proximately ∼16 times of a bare flat fiber end ( Fig.2 (b) and Fig.2 (c)) (when assuming the diamond's refractive index is ndi ≈ 2.4). In experiments, using a long TOF (∼ 7.4 µm in tip diameter and ∼ 15 mm in length), we achieved 15.03 times fluorescence collection efficiency of a bare flat fiber end in collecting fluorescence from the same 5 µm diamond (table.I ). A simulation example of multiple reflections comparison of both fluorescence col- lection and excitation between TOF tip and bare fiber end is shown in Fig.3. Figure 1: The tapered optical fiber (TOF) tip. (a) Schematic diagram of the light ray trace that resulting in the high NA for the multimode optical fiber based TOF tip. (b) Optical microscope image of part of a fabricated long TOF tip's tapered region (total TOF length is ∼15 mm, the photo is taken with 50 times magnification). (c) the side view and (d) the 45-degree view of another fabricated short TOF tip with a length of ∼1 mm in a scanning electron microscope. The TOF concept stems from the single-mode fiber TOF called adiabatic taper. Decreasing the diameter shrinkage ratio of an optical fiber taper can reduce the light transmission loss of the taper region; when the shrinkage ratio reached a certain value, the transmission loss can be almost ignored28,29. Through multiple total internal reflections, the diameter slowly shrinking TOF condenses the 532-nm laser from the multimode fiber core into the thin TOF tip, at the same time, converts the flu- Figure 2: The collection efficiency calculation schematic diagram of the TOF tip and flat fiber end. (a) is the fiber light collection solid angle. (b) and (c) are fluorescence acceptance angles of the TOF tip and bare flat fiber end when collecting light from a contacted diamond respectively. Figure 3: Simulation of multiple reflections in (a) NV fluorescence collection and in (b) NV excitation for both TOF tip (top) and bare flat fiber end (bottom). In (a) the original fluorescence rays are directed in 2◦, 171◦, 203◦ and 186◦ from same points/positions of the diamond in both the diamonds respectively. At the bottom of (b), the rays (blue) did not enter the diamond are not exist rays just for comparison. (The simulation is carried out by free software Optgeo: http://jeanmarie.biansan.free.fr/optgeo.html.) Not taperedtaper regionSmall tiplight beamncorencladnairncore>nclad>nairCoreCladding(a)(a)θmaxnfibernairnfiber>nairInternal reflection10oncore =1.496ndi ≈2.4(c)79.3o6.64oarcsin(1.47/1.496)≈79.3o6.64onpol =1.56nclad =1.470o0o10o26.67onfiber=1.47ndi ≈2.4arcsin(1/1.47)≈42.8o(b)26.67o30o42.8o2.1. TOF tip and diamonds preparing The optical fiber we used is a graded index multi- mode fiber (GIF625, Thorlabs. Core diameter: 62.5 µm, cladding diameter:125 µm, NA : 0.275±0.015 @ 850 nm). The TOF tip was fabricated by first melting and stretch- ing the fiber and then cutting the taper at its waist. By adjusting the heating and stretching force, the shape of the taper can be modified. More details about available tapering techniques to control the taper shapes and ta- per optical properties have been previously published in the literature30 -- 34. We fabricated two kinds of TOF tips: long TOF tip with a small taper's diameter shrinkage ra- tio tapered under flame heating (Fig.1(b), over ∼ 10 mm long), and short TOF tip with a large taper's diameter shrinkage ratio tapered under arc charging heating of an optical fusion splicer (Fig.1 (c) and (d), less than ∼1.5 mm long). The diamonds were prepared by irradiating the high- temperature high-pressure (HPHT) synthetic monocrys- talline diamond powders with 3-MeV H+ to form the vacancies in the diamond. Subsequently annealed the di- amond samples in vacuum at 800 oC for approximately two hours to concentrate the nitrogen and vacancies5. 2.2. TOF tip collection efficiency calculation An acceptance angle θmax related solid angle (the spherical surface area of the angle related ball crown di- vided by the square of the spherical radius) divided by 4π is its collection efficiency (Fig.2(a)). Taking the di- amond attached on the fiber tip surface as a thin sheet point sources ensemble, the fiber tip's collection efficiency is 2πr2[1 − cos(θmax)]/r2 4π 1 − cos(θmax) 2 = (1) η = core − n2 NA is defined as N A = (n2 clad)/2 = 0.275 for the used optical fiber; (ncore is the refractive index of the fiber core and nclad is the refractive index of the fiber cladding). The group refractive index of GIF625 fiber is 1.496 (@ 850 nm, the GIF625 fiber data sheet); then the cladding refractive index of the fiber nclad ≈ 1.47. Ignor- ing the refractive index distribution in the cross-section of the TOF tip and treating it as the core with air sur- rounding it as the cladding. The maximum acceptance angle of the TOF tip is calculated according to Fig.2(b). θmax = arcsin[ ndi sin(π/2 − arcsin(nclad/ncore) × ncore] (2) Taking ndi ≈ 2.4 as the refractive index of diamond, nf iber = nclad ≈ 1.47 as the TOF tip's refractive index and nclad = nair ≈ 1, a maximum acceptance angle of θmax = 26.676◦ is obtained for the TOF tip (Fig.2(b). If 3 we take nf iber = ncore ≈ 1.496 as the TOF tip refractive index, θmax = 27.620◦). Using equation (1), the TOF's collection efficiency can reach ηd ≈ 5.322% (If we take θmax = 27.620◦, ηd ≈ 5.698%) in collecting fluorescence from an attached diamond. For the flat bare fiber end, using equation (2), taking ncore = 1.496 and nclad = 1.47, the calculated θmax ≈ 6.644◦ and the collection efficiency η0.275d ≈ 0.336% for accepting fluorescence from a con- tacted diamond (Fig.2 (d)). The TOF tip's fluorescence collection efficiency is ∼ 16 times of a bare flat fiber end (ηd/η0.275d ≈ 5.322/0.336 ≈ 16; if we take ηd ≈ 5.698%, ηd/η0.275d ≈ 5.698/0.336 ≈ 17 ). Taking θmax = 26.676◦ the calculated NA = ndi × sin(θmax) of the TOF tip is N A ≈ 1.077 (if take θmax = 27.620◦, N A ≈ 1.113). As the strongest fluorescence of a diamond NV cen- ter is 630-780 nm instead of 850 nm at which the fiber parameters are given, and there are internal multiple re- flections in the diamond, the exact collection efficiency values may deviate from the calculated values. However, these simple calculations still illustrate that the TOF tip will enhance the fluorescence collection efficiency signifi- cantly. Since the not tapered graded-index multimode fiber can partly concentrate the laser light in the central core of the fiber, the bare flat fiber end has a relatively high excitation efficiency in illumination a micrometer-size di- amond attached at its end center. The TOF can further concentrate laser light on to its tip and enhance its flu- orescence excitation efficiency. But its enhancing value will be relatively smaller than its collection efficiency en- hancing. 3. RESULTS AND DISCUSSION The TOF fluorescence collection and excitation setup efficiency enhancement assessing experiment schematic and microscope images are shown in Fig.4 and Fig.5. The results are shown in table.I. We obtained the fluorescence collection efficiency enhancing value "col- lect" by comparing the intensities of fluorescence col- lected from the tested TOF tip and bare fiber end re- spectively when the diamond is attached on a bare fiber tip and irradiated with constant laser intensity through the bare fiber tip (Fig.4). In this way, the diamond emits constant fluorescence out, the collected fluorescence in- tensity will only depend on the tested collection tip's col- lection ability. Similarly, we obtained the fluorescence excitation efficiency enhancing value "excite" by com- paring the collected fluorescence intensities through the bare fiber tip attached to the diamond when the diamond is irradiated by the tested TOF tip and flat fiber end respectively under constant laser intensity coupled into them (fiber tip allying need to be reversed in the enlarged dashed box of Fig.4). In this way, the collection bare fiber tip's fluorescence collection ability is fixed, and the cou- pled laser intensity into the TOF tip and flat fiber end is constant, the diamond emits fluorescence intensity, so as 4 with a tip size of ∼10 µm can get the maximum excita- tion efficiency. the collected fluorescence intensity, will only depend on the tested TOF tip and flat fiber end's excitation abil- ity (concentrate the laser light into the diamond). In all experiments, both the tested TOF tip and the bare fiber end are dipped with a tiny amount of immersion oil (Fluka 51786, refractive index: 1.512 - 1.522) to en- hance its contact to the diamond. The lengths of long TOF (long taper in table.I) and short TOF (short taper in table.I) are ∼15mm and ∼1.1 mm respectively. Figure 4: TOF tip's fluorescence collection and excitation efficiency enhancement assessing setup. O1, O2, and O3 are objectives; CF is a 512-nm clear-up filter and LP is a 615-nm long-pass filter. The diamond is attached (by UV-curing glue) to the center of a bare fiber tip end (enlarged in the dashed box). In the fluorescence collection efficiency measuring, the diamond is illuminated under constant laser intensity through the bare fiber tip, the fluorescence is collected by the tested TOF tip or flat fiber end. In the fluorescence excitation efficiency measuring (fiber tip allying need to be reversed in the enlarged dashed box), the fluorescences are collected by the bare fiber tip and the diamond is illuminated through the bare fiber end or the tested TOF tip with constant laser intensity coupled into them. Table.I indicate that the long TOF tip can obtain a maximum value of 15.03 times the fluorescence collec- tion efficiency of a bare flat fiber end, it agrees to our calculated 16 times based on geometry ray optics. This is because our TOF and diamond have a dimension scale larger than ten times the wavelengths of the laser or fluo- rescence, fit the geometry ray optics using condition. Ta- ble.I also shows that the fluorescence collection efficiency is sensitive to the relative size of the TOF tip and dia- mond; generally, a smaller tip size TOF will have a better collection efficiency for a smaller diamond, when the di- amond size is set, using a TOF tip with a tip size close to the diamond can collect more fluorescence. While for fluorescence excitation, if the TOF tip size is too small, it will reduce the excitation efficiency. Using a TOF tip Figure 5: Microscope images of fluorescence excitation and collection efficiency measuring sets of (a) a TOF tip, (b) a flat fiber end and (c) a short TOF tip. We positioned another ∼12.5 µm NV enriched dia- mond crystal on a long TOF (tip diameter: ∼12.8 µm, length: ∼15 mm. Inset (a) of Fig.7). We used the setup shown in Fig.6 and coupled ∼151 µW 532-nm laser light into the fiber to illuminate the diamond attached on the TOF tip and collected ∼262 nW fluorescence through the same TOF tip (the power meter wavelength was set at 635 nm). Subsequently, we cut off the TOF tip and transferred the same diamond onto the center of the flat fiber end (inset (b) of Fig.7). We find that to obtain ∼262 nW fluorescence, approximately 4.57 mW of laser power (∼30 times more than the TOF tip needed) is re- quired to be coupled into the same fiber. Fig.7 shows the collected fluorescence spectra of the ∼12.5 µm NV enriched diamond attached on a long TOF tip, flat fiber end, and a ∼156 µm NV enriched diamond attached to a flat fiber end (inset (c) of Fig.7) with UV-curing glue when coupling ∼151 µW of the 532-nm laser into the op- tical fiber. We also positioned some other size diamond on to long TOF tip, short TOF tip, and bare flat fiber end respectively. Table.II shows the required laser power coupling into the optical fiber to obtain 262 nW fluores- cence through the same fiber for different fiber tips. It is clear, a TOF with short taper length exhibits a larger loss in transforming the light between the tapered tip and the not tapered fiber region due to the large fiber diam- eter shrink ratio cannot meet the adiabatic criteria; as a result, a short-TOF has a lower fluorescence excitation and collection efficiency than that of the long-TOF. How- ever, a shorter TOF tip can provide a better mechanical stability than the longer TOF, which may benefit certain TOF tipFlat fiber endlaserCFLPO3O2SpectrometerO1ComputerFiberTABLE I: Excitation and collection enhancement diamond size 5 µm 11.7 µm 50 µm 5 tip size collecta exciteb collecta exciteb collecta exciteb 0.97 7.4 µm 15.03 long taper1 1.11 long taper2 9.6 µm 13.13 long taper3 17.6 µm 5.35 1.12 short taper 2.04 2.13 1.93 2.42 10.79 8.51 5.64 1.35 1.85 2.11 6.1 12.6 µm bare fiber end 125 µm -- 1 7.68 7.84 5.73 -- 1 1 1 -- 1 -- 1 a ratio of the collected fluorescence intensity of the TOF tip and the bare fiber end; b ratio of the excited fluorescence intensity of the TOF tip and the bare fiber end. TABLE II: Obtain 262 nW fluoresence needed excitation laser power taper length require laser power a 12.8 µm 15mm diamond size fiber end type tip size 12.5 µm 12.5 µm 156 µm 7.9 µm 11.3 µm 11.7 µm 11.7 µm long taper bare fiber end 125 µm -- bare fiber end 125 µm -- long taper long taper short taper bare fiber end 125 µm -- 12.2 µm 12mm 13.3 µm 10mm 16.8 µm 1 mm 151 µW 4.57mW 1.52mW 333 µW 202 µW 846 µW 8.07 mW a The laser power coupled into the optical fiber attached to the diamond applications that require high TOF tip robustness. For comparison, we used the Fig.6 setup and inves- tigated the possibility of using an ultra-high NA pho- tonic microstructure fiber MM-HNA-35 (NKT Photon- ics, Denmark) in fluorescence excitation and collection from diamonds smaller than ten micrometers. Surpris- ingly, we find that when coupled in with equal laser power, the MM-HNA-35 fiber even collected less fluores- cence than not tapered GIF625 fiber. We believe this is because, unlike the graded-index core GIF625 fiber, the 35 µm pure silica core of the MM-HNA-35 fiber cannot focus the laser light into its center (Fig.8.(a)); thus, it is poor at excitation diamond smaller than its core. When tapering the MM-HNA-35 fiber, the air isolation band between the core and cladding will partly keep ( Fig.8 (b) shows its schematic illustration), most of the fluo- rescence entered the cladding region (92% of the tapered tip: 1−(35/125)2 ≈ 0.92) cannot be coupled into the core of the fiber; as a result, the tapering enhanced laser ex- citation cannot compensate for it decreased fluorescence collection. Thus it is not surprising to find out that the TOF form of the MM-HNA-35 tip even has collected less fluorescence than that of its un-tapered form when cou- pled in an equal laser power. In principle, tapering the MM-HNA-35 fiber into TOF tip with tip's core equal to the used diamond or directly design a fiber with a core equal to the used diamond will solve this problem. In addition, this TOF tip can combine the micro- concave technique35 (locate the diamond attached on the TOF tip at the focal point of a matched micro- concave mirror facing against the tip) to further enhance the fluorescence excitation and collection efficiency. As the micro-concave mirror can focus back the fluorescence emitted in the opposite direction of the TOF in a smaller incidence angle and the TOF has a relatively large flu- orescence acceptance angle, this combining can greatly improve the system's fluorescence collection. Ideally, if the matched micro-concave mirror is in parabolic reflec- tor shape, the total fluorescence collection efficiency will be around 50% + TOF tip (without using of the micro- concave mirror) fluorescence collect efficiency. If the TOF tip's fluorescence collection efficiency is 5%, com- bine it with a matched micro parabolic mirror will reach a fluorescence collection efficiency of 55%, and enhance10 times the fluorescence collection efficiency. In a simple experiment evaluation, for an unmatched relatively big micro-concave mirror, we obtained only ∼ 2.2 times the collected fluorescence efficiency of the TOF tip without using the micro-concave mirror. The excitation efficiency of NV-center on a TOF tip is the number of generated fluorescence photons divide the number of 532-nm photons coupled into the optical fiber. Taking the ∼12.5-µm diamond on the ∼12.8-µm TOF tip for example. The excitation laser is ∼151 µW and the collected fluorescence is ∼262 nW. Using a wavelength calibrated calculation, we obtained the collected fluores- cence photon number and incidence laser photon number ratio σ ≈ 1.864/1000. Divided by the calculated fluores- cence collection efficiency 5.3%, an NV-center excitation efficiency of 3.52% is obtained. As the NVs' fluorescence excitation and collection enhancing will greatly benefit NV-spin magnetometry applications9,27, the utilize of a TOF tip will certainly enhance its sensitivity. To give a hint on this enhanc- 6 Figure 6: Fiber-based diamond NV center system setup. The 532-nm laser (solid arrow), after passing the 532-nm clean-up filter CF and the dichroic mirror DM, is coupled by the objective O1 into the optical fiber OF1 tip where the diamond is attached (dashed triangular box). The fluorescence collected by the TOF/Flat fiber end cleared by 615-nm long-pass filter LP (dashed arrow) and coupled by objective O2 into optical fiber OF2 which is connected to the optical power meter/optical spectrometer/photodiode OM/OS/PD. The DC power supply is used to apply DC magnetic field to the diamond ing effect, we resorted to CW-ODMR by detecting the fluorescence emission and varying the frequency of mi- crowaves that induce spin transitions. We performed a series of experiments and compared the ODMR mag- netic field detection sensitivity of two diamonds crystals bonded on TOF tip and flat fiber end (non-TOF). The results (Fig.9) shown that under the same excitation laser power, the small ∼7.9-µm diamond on the 12.5-µm TOF tip (Fig.9 (a) ) have about 28.3 times the sensitivity of the ∼11.4-µm diamond on the flat fiber end. As a demon- stration, we bonded a ∼11.3-µm diamond on TOF tip by UV-curing glue and used it for measuring a DC magnetic field. We used only 0.238 mW of laser power to initial- ize the NV-centre ensemble, and -15 dBm of microwave power to induce the electronic spin resonance between the ms=0 spin state and the ms=±1 states of the NV centers. DC current was applied to a coil to vary the external magnetic field. The single shot ODMR spectra at various magnetic fields are shown in Fig.10; the sig- nals are obtained without utilizing any special techniques such as lock-in amplification or data averaging. 4. CONCLUSION To summarize, we have fabricated a quasi-adiabatic TOF tip that performs similarly to an objective in a clas- sical optical path for both the excitation and collection Figure 7: Comparison of measured fluorescence into the fiber is ∼151 µW. (a) is the spectrum of a spectra. In all cases, the excitation laser power launched ∼12.5-µm diamond on a TOF tip, (b) is the spectrum of a ∼12.5-µm diamond on a cleaved flat fiber end, and (c) is the spectrum of a ∼156-µm diamond on a cleaved flat fiber end. The inset shows microscope images of these samples. The length of the TOF is ∼15 mm, and its tip diameter is ∼12.8 µm. The fiber diameter is ∼125 µm. Figure 8: (a) Microscope images of a ∼5.2 µm diamond on the center of the cleaved flat fiber end of the Ultra-high NA photonic microsctructure fiber. (b) Schematic diagram of the tapered ultra-high NA photonic microstructure fiber. of the fluorescence from NV centers in micrometer-sized diamond. Calculation shows that the TOF tip possesses an ultra-high NA; experiments demonstrate that this technique can greatly enhance the excitation and fluo- rescence collection efficiency. We used this enhancing √ to boost the magnetic sensitivity of micro-sized NV magnetometer and achieve a sensitivity of 180nT Hz for a ∼7.9-µm diamond crystal for DC magnetic field StimulationFluorescenceMicrowave synthesizercoil TOF/flat fiberendOF1DCOM/OS/PDlaserOF2DMO1CFLPO27 Figure 10: ODMR spectra of the NV-center ensembles in a ∼11.3-µm diamond bonded on the tip of a ∼10-mm long TOF recorded for various magnetic field values (Fluorescence intensity have been normalized). The TOF tip' diameter is ∼13.3 µm. The magnetic field was applied by the DC coil. The test was executed using the Fig.6 setup, the laser power used to excite the diamond was 0.239 mW and the collected fluorescence was 0.309 µW (the power meter was set at 635 nm). Figure 9: (a) ODMR spectra of a ∼11.4-µm diamond UV-curing glue bonded on a flat fiber end center in a 9.6 mT magnetic field, and (b) the ODMR spectra of the UV-curing glue bonded∼7.9-µm diamond on TOF tip in a 9.0 mT magnetic field, and the laser launched into the optical fiber for stimulation both diamonds are the same ∼36.5 µW. The signals are read from a photodiode detector. (c) The Magnetic field sensitivity comparison of the two sensor heads (a) and (b). sensing. Furthermore, this improved excitation and collection enhancements reduce the size of the diamond sample that is required for precision sensing, potentially increases the spatial resolution (less than 5 µm) and retains the flexibility of optical fiber-based NV sensors. The authors thank Dr. Andrii Lazariev for critical reading of the manuscript. We gratefully acknowledge funding from the Max-Planck Society, Niedersachsisches Ministerium fur Wissenschaft und Kultur and DFG Research Center Nanoscale Microscopy and Molecular Physiology of the Brain. Magnetic field intensity (mT)Microwave frequency (GHz)Fluorescence intensity (normalized)2.832.60.960.98101015501.362.434.185.7110.87.549.2613.412.18 ∗ † [email protected] [email protected] 1 F. Jelezko, J. Wrachtrup, Single defect centres in diamond: A review, physica status solidi (a) 203 (13) (2006) 3207 -- 3225. 2 M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, L. C. Hollenberg, The nitrogen-vacancy colour centre in diamond, Physics Reports 528 (1) (2013) 1 -- 45. 3 L. Childress, M. V. Gurudev Dutt, J. M. Taylor, A. S. Zibrov, F. Jelezko, J. Wrachtrup, P. R. Hemmer, M. D. Lukin, Coherent dynamics of coupled electron and nuclear spin qubits in diamond, Science 314 (5797) (2006) 281 -- 285. 4 G. Balasubramanian, P. Neumann, D. Twitchen, M. Markham, R. Kolesov, N. Mizuochi, J. Isoya, J. Achard, J. Beck, J. Tissler, V. Jacques, P. R. Hemmer, F. Jelezko, J. Wrachtrup, Ultralong spin coherence time in isotopically engineered diamond, Nat Mater 8 (5) (2009) 644 -- 647. 5 T.-J. Wu, Y.-K. Tzeng, W.-W. Chang, C.-A. Cheng, Y. Kuo, C.-H. Chien, H.-C. Chang, J. Yu, Tracking the engraftment and regenerative capabilities of trans- planted lung stem cells using fluorescent nanodiamonds, Nat. Nano. 8 (2013) 682 -- 689. 6 K. Y. Han, S. K. Kim, C. Eggeling, S. W. Hell, Metastable dark states enable ground state depletion microscopy of nitrogen vacancy centers in diamond with diffraction- unlimited resolution, Nano Letters 10 (8) (2010) 3199 -- 3203. 7 G. Balasubramanian, I. Y. Chan, R. Kolesov, M. Al- Hmoud, J. Tisler, C. Shin, C. Kim, A. Wojcik, P. R. Hem- mer, A. Krueger, T. Hanke, A. Leitenstorfer, R. Brats- chitsch, F. Jelezko, J. Wrachtrup, Nanoscale imaging mag- netometry with diamond spins under ambient conditions, Nature 455 (2008) 648. 8 J. R. Maze, P. L. Stanwix, J. S. Hodges, S. Hong, J. M. Taylor, P. Cappellaro, L. Jiang, M. V. G. Dutt, E. Togan, A. S. Zibrov, A. Yacoby, R. L. Walsworth, M. D. Lukin, Nanoscale magnetic sensing with an individual electronic spin in diamond, Nature 455 (2008) 644. 9 J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, D. Bud- ker, P. R. Hemmer, A. Yacoby, R. Walsworth, M. D. Lukin, High-sensitivity diamond magnetometer with nanoscale resolution, Nat Phys 4 (10) (2008) 810 -- 816. 10 L. Rondin, J.-P. Tetienne, T. Hingant, J.-F. Roch, P. Maletinsky, V. Jacques, Magnetometry with nitrogen- vacancy defects in diamond, Reports on Progress in Physics 77 (5) (2014) 056503. 11 F. Dolde, M. W. Doherty, J. Michl, I. Jakobi, B. Naydenov, S. Pezzagna, J. Meijer, P. Neumann, F. Jelezko, N. B. Man- son, J. Wrachtrup, Nanoscale detection of a single funda- mental charge in ambient conditions using the N V − center in diamond, Phys. Rev. Lett. 112 (2014) 097603. 12 P. Neumann, I. Jakobi, F. Dolde, C. Burk, R. Reuter, G. Waldherr, J. Honert, T. Wolf, A. Brunner, J. H. Shim, D. Suter, H. Sumiya, J. Isoya, J. Wrachtrup, High- precision nanoscale temperature sensing using single de- fects in diamond, Nano Letters 13 (6) (2013) 2738 -- 2742. 13 G. Kucsko, P. C. Maurer, N. Y. Yao, M. Kubo, H. J. Noh, P. K. Lo, H. Park, M. D. Lukin, Nanometre-scale ther- mometry in a living cell, Nature 500 (7460) (2013) 54 -- 58. 14 D. M. Toyli, C. F. de las Casas, D. J. Christle, V. V. Dobrovitski, D. D. Awschalom, Fluorescence thermome- try enhanced by the quantum coherence of single spins in diamond, PNAS 110 (21) (2013) 8417 -- 8421. 15 M. R. Henderson, B. C. Gibson, H. Ebendorff-Heidepriem, K. Kuan, S. Afshar V., J. O. Orwa, I. Aharonovich, S. Tomljenovic-Hanic, A. D. Greentree, S. Prawer, T. M. Monro, Hybrid materials: Diamond in tellurite glass: a new medium for quantum information (adv. mater. 25/2011), Advanced Materials 23 (25) (2011) 2772 -- 2772. 16 Y. Ruan, H. Ji, B. C. Johnson, T. Ohshima, A. D. Greentree, B. C. Gibson, T. M. Monro, H. Ebendorff- Heidepriem, Nanodiamond in tellurite glass part ii: practi- cal nanodiamond-doped fibers, Opt. Mater. Express 5 (1) (2015) 73 -- 87. 17 J. R. Rabeau, S. T. Huntington, A. D. Greentree, S. Prawer, Diamond chemical-vapor deposition on optical fibers for fluorescence waveguiding, Applied Physics Let- ters 86 (13) (2005) 134104. 18 T. Schroder, A. W. Schell, G. Kewes, T. Aichele, O. Ben- son, Fiber-integrated diamond-based single photon source, Nano Letters 11 (1) (2011) 198 -- 202, pMID: 21138271. 19 L. Mayer, T. Debuisschert, Direct optical interfacing of cvd diamond for deported sensing experiments involving nitrogen-vacancy centres, physica status solidi (a) 213 (10) (2016) 2608 -- 2613. 20 L. Liebermeister, F. Petersen, A. v. Mnchow, D. Bur- chardt, J. Hermelbracht, T. Tashima, A. W. Schell, O. Benson, T. Meinhardt, A. Krueger, A. Stiebeiner, A. Rauschenbeutel, H. Weinfurter, M. Weber, Tapered fiber coupling of single photons emitted by a determinis- tically positioned single nitrogen vacancy center, Applied Physics Letters 104 (3) (2014) 031101. 21 R. N. Patel, T. Schroder, N. Wan, L. Li, S. L. Mouradian, E. H. Chen, D. R. Englund, Efficient photon coupling from a diamond nitrogen vacancy center by integration with sil- ica fiber, Light: Sci. Appl. 5 (2016) e16032. 22 M. J. Burek, C. Meuwly, R. E. Evans, M. K. Bhaskar, A. Sipahigil, S. Meesala, B. Machielse, D. D. Sukachev, C. T. Nguyen, J. L. Pacheco, E. Bielejec, M. D. Lukin, M. Loncar, Fiber-coupled diamond quantum nanophotonic interface, Phys. Rev. Applied 8 (2017) 024026. 23 I. V. Fedotov, L. V. Doronina-Amitonova, A. A. Voronin, A. O. Levchenko, S. A. Zibrov, D. A. Sidorov-Biryukov, A. B. Fedotov, V. L. Velichansky, A. M. Zheltikov, Elec- tron spin manipulation and readout through an optical fiber, Scientific Reports 4 (2014) 5362. 24 I. V. Fedotov, L. V. Doronina-Amitonova, D. A. Sidorov- Biryukov, N. A. Safronov, S. Blakley, A. O. Levchenko, S. A. Zibrov, A. B. Fedotov, S. Y. Kilin, M. O. Scully, V. L. Velichansky, A. M. Zheltikov, Fiber-optic magnetic- field imaging, Opt. Lett. 39 (24) (2014) 6954 -- 6957. 25 I. V. Fedotov, S. Blakley, E. E. Serebryannikov, N. A. Safronov, V. L. Velichansky, M. O. Scully, A. M. Zheltikov, Fiber-based thermometry using optically detected mag- netic resonance, Applied Physics Letters 105 (26) (2014) 261109. 26 I. V. Fedotov, S. M. Blakley, E. E. Serebryannikov, P. Hem- mer, M. O. Scully, A. M. Zheltikov, High-resolution mag- netic field imaging with a nitrogen-vacancy diamond sensor integrated with a photonic-crystal fiber, Opt. Lett. 41 (3) (2016) 472 -- 475. 9 27 L. M. Pham, D. L. Sage, P. L. Stanwix, T. K. Yeung, D. Glenn, A. Trifonov, P. Cappellaro, P. R. Hemmer, M. D. Lukin, H. Park, A. Yacoby, R. L. Walsworth, Magnetic field imaging with nitrogen-vacancy ensembles, New Jour- nal of Physics 13 (4) (2011) 045021. 28 J. D. Love, W. M. Henry, W. J. Stewart, R. J. Black, S. Lacroix, F. Gonthier, Tapered single-mode fibers and devices: Part 1, adiabaticity criteria, IEE Proc., Optoelec- tron. 138 (5) (1991) 343 -- 354. 29 R. J. Black, S. Lacroix, F. Gonthier, J. D. Love, Tapered single mode fibers and devices: Part 2, experimental and theoretical quantification., IEE Proc. 138 (5) (1991) 355 -- 364. 30 T. A. Birks, Y. W. Li, The shape of fiber tapers, Journal of Lightwave Technology 10 (4) (1992) 432 -- 438. 31 A. Stiebeiner, R. Garcia-Fernandez, A. Rauschenbeutel, Design and optimization of broadband tapered optical fibers with a nanofiber waist, Opt. Express 18 (22) (2010) 22677 -- 22685. 32 S. Pricking, H. Giessen, Tapering fibers with complex shape, Opt. Express 18 (4) (2010) 3426 -- 3437. 33 H. Latifi, M. I. Zibaii, S. M. Hosseini, P. Jorge, Nona- diabatic tapered optical fiber for biosensor applications, Photonic Sensors 2 (4) (2012) 340 -- 356. 34 S. Harun, K. Lim, C. Tio, K. Dimyati, H. Ahmad, Theoret- ical analysis and fabrication of tapered fiber, Optik 124 (6) (2013) 538 -- 543. 35 D. Duan, V. K. Kavatamane, S. R. Arumugam, G. Ra- hane, Y.-K. Tzeng, H.-C. Chang, H. Sumiya, S. Onoda, J. Isoya, G. Balasubramanian, Enhancing fluorescence ex- citation and collection from the nitrogen-vacancy center in diamond through a micro-concave mirror, Applied Physics Letters 113 (4) (2018) 041107. Additional Material 10 Figure Fig-add1: Microscope images of (a) the side view of the ∼ 11.7-µm diamond on the TOF tip (the TOF tip has a diameter of ∼16.8-µm), (b) the same diamond on the centre of the cleaved flat fiber end (the fiber diameter is ∼125 µm), (c) the side view of the ∼7.9-µm diamond bonded on the ∼12.2 µm TOF tip by UV glue, and (d) a ∼11.4 µm diamond bonded on the centre of the cleaved flat fiber end by UV glue. 5. FLUORESCENCE COLLECTION SIMULATION The simulation is carried out by free software Optgeo which can be downloded for free from http://jeanmarie.biansan.free.fr/optgeo.html.The results are shown in Fig.Fig-add2.a and b. According to the comparison of the simulation of the diamond NV center fluorescence collection of the tapered optical fiber tip and flat optical fiber end results. The incidence angle of fluorescence for the tapered tip are both 26, while for flat optical fiber tip, the incidence anges are 26 and 6.5 respectively. The refractive index of the diamond, fiber core, fiber cladding, polymer coating and air are 2.4, 1.496, 1.47, 1.56 and 1, respectively. The results shows that for the tapered optical fiber tip, the fluorescence have enter angle as large as 26 can be grasped by the fiber taper and guided into the optical fiber core through the tapered region, while that for the flat optical fiber end, only the fluorescence have incidence angle smaller than 6.5 can be guided in the fiber core, and the larger incidence angle light will be absorbed or trapped out of the fiber by the polymer coating. Fig.Fig-add3.b shows the simulation results of combining the micro concave technique (locate the diamond at the focal point of a micro-concave mirror facing the optical fiber tip) with the TOF tip, and Fig.Fig-add3.a shows the simulation results of using a sphered fiber end for comparison. It is clear that the micro-concave technique will enhance the collection efficiency for the TOF tip configuration. 11 Figure Fig-add2: Comparison of the simulation of the diamond NV center fluorescence collection of the(top)tapered optical fiber tip and (bottom)flat optical fiber end results.Simulation parameters: Fiber diameter is 125 µm,tapered optical fiber tip diameter is 12.5 µm,diamond diameter is 30 µm, length of the taper is 1.2 mm. Figure Fig-add3: Comparison of the simulation of the diamond NV center fluorescence collection of the (a) sphered end combine with micro-concave mirror and (b) TOF tip combine with micro-concave mirror. Simulation parameters: Fiber diameter is 125 µm, tapered optical fiber tip diameter is 20 µm, micro-concave mirror curvature radius is 120 µm, length of the taper is 500 µm.
1812.02673	1	1812	2018-12-06T17:15:18	Formation of periodic surface structures on dielectrics after irradiation with laser beams of spatially variant polarisation: a comparative study	[ "physics.app-ph", "cond-mat.mtrl-sci", "physics.optics" ]	A comparative study is performed to explore the periodic structure formation upon intense femtosecond pulsed irradiation of dielectrics with radially and azimuthally polarised beams. Laser conditions have been selected appropriately to produce excited carriers with densities below the optical breakdown threshold in order to highlight the role of phase transitions in surface modification mechanisms. The frequency of the laser induced structures is calculated based on a theoretical model that comprises estimation of electron density excitation, heat transfer, relaxation processes, and hydrodynamics-related mass transport. The influence of the laser wavelength in the periodicity of the structures is also unveiled. The decreased energy absorption for azimuthally polarised beams yields periodic structures with smaller frequencies which are more pronounced as the number of laser pulses applied to the irradiation spot increases. Similar results are obtained for laser pulses of larger photon energy and higher fluences. All induced periodic structures are oriented parallel to the laser beam polarisation.	physics.app-ph	physics	Formation of periodic surface structures on dielectrics after irradiation with laser beams of spatially variant polarisation: a comparative study Antonis Papadopoulos 1,2, Evangelos Skoulas 1,2 , George D.Tsibidis 1♣, and Emmanuel Stratakis 1,2♣ 1 Institute of Electronic Structure and Laser (IESL), Foundation for Research and Technology (FORTH), N. Plastira 100, 2 Materials Science and Technology Department, University of Crete, 71003 Heraklion, Greece Vassilika Vouton, 70013, Heraklion, Crete, Greece ABSTRACT A comparative study is performed to explore the periodic structure formation upon intense femtosecond pulsed irradiation of dielectrics with radially and azimuthally polarised beams. Laser conditions have been selected appropriately to produce excited carriers with densities below the optical breakdown threshold in order to highlight the role of phase transitions in surface modification mechanisms. The frequency of the laser induced structures is calculated based on a theoretical model that comprises estimation of electron density excitation, heat transfer, relaxation processes, and hydrodynamics-related mass transport. The influence of the laser wavelength in the periodicity of the structures is also unveiled. The decreased energy absorption for azimuthally polarised beams yields periodic structures with smaller frequencies which are more pronounced as the number of laser pulses applied to the irradiation spot increases. Similar results are obtained for laser pulses of larger photon energy and higher fluences. All induced periodic structures are oriented parallel to the laser beam polarisation. Keywords: Ultrashort pulses, dielectrics, radial and azimuthal polarization, phase transitions, modelling, periodic structures ♣ [email protected] (Theory) ♣ [email protected] (Experiment) 1 1. Introduction Laser processing of dielectrics with femtosecond pulses has received remarkable attention due to its important technological applications [1-9]. Regarding the underlying physical processes that lead to surface modification, the predominant emphasis was initially centred on the role of electron density in the induced surface profile through the estimation of the optical breakdown damage threshold (OBT) [10-12]. Nevertheless, a more precise approach requires a direct correlation of the laser beam characteristics, the excited electrons density and the induced thermal effects suggests that morphological changes on the surface of dielectrics should involve a thermal criterion [13-15]. One type of surface modification, the laser induced periodic surface structures (LIPSS) (see [16] and references therein) has been extensively explored assuming linearly polarised laser beams (LPB). Nevertheless, it is important to elaborate on whether other types of polarisation states lead to structures with different morphological profiles. More specifically, laser beams with cylindrical states, namely radial and azimuthal polarization, have gained considerable attention in the past two decades, as the symmetry of the polarization enables new processing strategies [17] with applications in various fields including microscopy, lithography [18], electron acceleration [19], material processing [17, 20, 21] and optical trapping [22]. The so-called cylindrical vector beams (CVB) have been the topic of numerous theoretical and experimental investigations [20, 23-29]. A recent work on nickel irradiation with femtosecond pulses revealed morphological changes for LPB and CVB for various fluences and number of pulses [30]. Given the wide range of applications of laser processed dielectrics, it is of paramount importance to explore the fundamental mechanisms using CVB. To the best of our understanding, no comparative study has been performed to underline the role of the laser beam polarisation state in the features (i.e. periodicity, height, spot size, dependence on laser pulses applied to the laser spot, etc.) of the induced surface morphology. In this work, we investigate the mechanism of periodic structure formation induced on fused silica upon irradiation with, radially and azimuthally polarised, (RPB and APB, respectively) femtosecond pulses. To evaluate the role of the polarisation state in the LIPSS periodicity, a comparative study is conducted to quantify morphological changes for using CVB laser beams with respect to fluence and number of pulses variation. The role of the laser beam wavelength on the periodicity of the produced structures is also highlighted. 2. Theory 2.1 Electron excitation and energy absorption. To simulate the excitation mechanism in fused silica, we consider the multiple rate equation model (MRE) [31, 32] 2 (1) where ne, nv, are the number densities of the excited electrons, and valence band electrons (nv=2.2×1022 cm-3), respectively. The last term in the first equation corresponds to free electron decay that is characterised by a time constant τr (τr~150fs in fused silica) that leads to a decrease of the electron density. For the sake of simplicity, formation of self- trapped excitons is ignored in Eq.1 [13, 32, 33]. The rest of the terms in Eq.1, PI, and correspond to the photoionization rate, impact ionization probability and the probability for single photon intraband absorption, respectively [32]. With respect to the intensity of the laser beam I, the attenuation of the local laser intensity is determined by multiphoton ionisation and inverse bremsstrahlung absorption assuming a laser beam of frequency ωL (i.e. wavelength λL) [13, 14, 34]. where In Eq.2, (2) (3) corresponds to the minimum number of photons necessary to be absorbed by an electron in the valence band to overcome the relevant energy gap EG and reach the conduction band. Also, R(t,x,y,z=0) stands for the reflectivity of the material, Ed is the fluence while F(x,y) is proportional to is the applied laser electric field). Hence, (as , , where a. For a radially polarized beam, denotes radial polarization are the unit vectors along the x- and y-axis, respectively. Hence, the field is and expressed as the superposition of orthogonally Hermite-Gauss HG01 and HG10 modes with x- and y-polarisation, respectively, which yields [23, 35] (4) where intensity is higher at distance from the spot-centre. b. On the other hand, for an azimuthal beam, , where denotes azimuthal polarization. Hence, the field is expressed as the superposition of 3 ve1t1Gk1pt1vrjtj1ptj11ptjrktk1ptk1krnnnnPI(E)2nWnnnnWnWn,1jk,nnWnnα1ptWvephLGevnnI(t,x)NPI(E)(n)I(t,x)zn2pdppt32log(2)I(t,x,y,z0)(1R(t,x,y,z0))Eexp4log(2)F(x,y,z)phN2EE1001rEHGxHGyrExy22222200xy2(xy)F(x,y)expRR0R/21001EHGyHGxEorthogonally Hermite-Gauss HG01 and HG10 modes with y- and x-polarisation, respectively, which yields F equal to Eq.(4) [23, 35] The transient optical properties of the irradiated material (i.e. refractive index, extinction and absorption coefficient and reflectivity) are computed through the evaluation of the dielectric constant for materials with a band gap [36] (5) where εun corresponds to the dielectric constant of the unexcited material, ε0 stands for the vacuum permittivity, e is the electron charge, mr (=0.5me, me stands for the electron's mass) is the electron's reduced mass, and τc=0.5 fs [37] is the Drude damping time. It is evident that in both types of CVB, F(x,y) is the same, however, the absorbed energy in the material is characterised by the difference of the reflectivity R between RP and AP beams. More specifically, a distinction should be identified on the calculated values of the reflectivity for s- or p-polarised beams. Hence, since azimuthal (radial) CVB are s- (p-) polarised, the reflectivity R is larger (smaller) which leads to smaller (larger) energy absorption at increasing angle of incidence (where morphological profile becomes steeper as a result of repetitive irradiation) [38]. To take into account the distribution of the absorbed energy, the Fresnel expressions for the transient reflectivity Rs and Rp for s- and p-polarised beams, respectively, are directly incorporated in the modelling approach through the following expressions (6) (7) is the extinction coefficient) and θi stands for the angle of incidence. to quantify the effect of the polarisation state. In the above expressions, N=n+i is the complex refractive index of the irradiated material (n is the real part of the refractive index and 2.2 Electron-Lattice relaxation processes. Due to the metallic character of the excited material, a TTM model is used to describe the spatio-temporal dependence of the temperatures Te and TL of the electron and lattice subsystems [10, 32, 39], respectively 4 GE2eeeun2v0rLLcnne1(n)1(1)11nm1i222iis22iicos()Nsin()Rcos()Nsin()2222iip222iiNcos()Nsin()RNcos()Nsin()rikrik (8) The complete expression for source term is given by [14, 32] (9) to account for photoionisation of electrons, avalanche ionisation, free electron absorption and energy loss due to trapping processes. We point out that a term that describes the divergence of the current of the carriers ( ) has not been taken into account (it appears when carrier dynamics in irradiated silicon is explored upon irradiation [40], however, it turns out that for small pulse durations its contribution is negligible [41]). The temperature, electron density and temporal dependence of the thermophysical properties, Ce, ke, CL and g are provided from well-established expressions derived from free electron gas based on the metallic character of the excited material [13, 32, 42]. 2.3 Carrier density related LIPSS periodicity. To correlate the excited electron density with surface structures, the inhomogeneous energy deposition into the irradiated material is computed through the calculation of the product η(kL,ki)×b(kL) as described in the Sipe-Drude model [43]. In the above expression, η describes the efficacy with which the surface roughness at the wave vector kL (i.e. normalized wavevector kL=λL/Λ, where Λ stands for the predicted periodicity due to electrodynamics) induces inhomogeneous radiation absorption, ki is the component of the wave vector of the incident beam on the material's surface plane and b represents a measure of the amplitude of the surface roughness at kL. To introduce surface roughness, the value b=0.4 is initially assumed [32], while the computation of the influence of efficacy influence due to a variable amplitude and shape factor of the produced profile is also considered that accounts for expected corrugation changes upon successive irradiation. To take into account the changes in the periodicity values for AP and RP beams, the computation of η is performed for s- and p- polarized beams, respectively [43, 44]. It is assumed that ki is along the unit radius vector. 2.4 Fluid dynamics To correlate the induced surface profile with thermal effects, two processes are considered resulting from the intensity profile of the laser beam and the laser energy. In principle, surface modification is usually associated with a phase transition (i.e. melting of the material). Nevertheless, near the centre of the laser spot, the deposition of energy is 5 eeeeeLLLeLTCkTgTTS(x,t)tTCgTTtS(x,t)veeekphGGGeBeBevVrnnnnn33S(x,t)NEPI(E)E(n)I(t,x)kTkTnn2t2Jrnot large enough to produce a phase transition and, thermoplastic effects are predominantly responsible for the permanent deformations. Due to the, significantly, smaller size of surface deformation in the irradiated spot due to the latter effects [30, 45], the role of elastoplastic effects in the surface modification is ignored in this work. fluid and therefore, the Navier-Stokes equation is used to describe the fluid dynamics It is also assumed that the material in the molten phase behaves as an uncompressed (10) and is the fluid velocity, μ(m) is the dynamic viscosity of the liquid, ) is the where density of the molten material, Tmelt is the melting temperature for SiO2, P stands for pressure, κ stand for lattice heat conductivity (see Table 1) and lattice heat capacity respectively, of the molten material. The second equation in Eq.10 results from energy conservation requirements. To describe the fluid dynamics, the movement of the isothermal Tmelt is followed to determine the resolidification process. 3. Simulation To determine the modification characteristics, a solidification criterion is used based on the lattice displacements when the associated lattice temperature drops to lower values than the melting point [30, 32, 46, 47]. To simulate the induced surface modification, on a spot of R0=15μm a three dimensional finite-difference method in a staggered grid is employed [46] to solve numerically the heat transfer equations, phase change and Table 1 Simulations parameters for SiO2 Parameter CL [106 Jm−3 K−1] [106 Jm−3 K−1] Tmelt [K] σ [N/m] [Kgr/m3] [mPa sec] Value 1.6 [42] Fitting [32, 48] 1988K [32] Fitting [32, 48] Fitting [32, 48] Fitting [32, 48] resolidification process [32]. At time t=0, both electron and lattice temperatures are set to 300K. For the part of the material which is in the liquid phase non-slipping conditions (i.e. velocity field is zero) are applied on the solid-liquid interface. The grid size taken for 6 ()()()()()0()())()()mmLLLLLmmmTLuTCuTTtuuuPuutu()mL()mL()mLC()mLC()mL()msimulations is 2nm (vertical dimension) and 2nm (horizontal dimension). The temporal step is adapted so that the stability Neumann condition is satisfied [49]. In the simulations, it is also assumed that a region where the lattice temperature exceeds the values ~1.8Tb (~4000K)[14] (where Tb is the boiling temperature) is ablated [45, 46, 50]. Regarding the adequate description of fluid dynamics, the hydrodynamic equations are solved in the subregion that contains either solid or molten material. To include the "hydrodynamic" effect of the solid domain, material in the solid phase is modelled as an extremely viscous liquid (μsolid = 105μliquid), which results in velocity fields that are infinitesimally small. An apparent viscosity is then defined with a smooth switch function to emulate the step of viscosity at the melting temperature. A similar step function is also introduced to allow a smooth transition for the rest of the temperature-dependent quantities (i.e., heat conductivity, heat capacity, density, etc.) between the solid and liquid phases. For time-dependent flows, a common technique to solve the Navier-Stokes equations is the projection method, and the velocity and pressure fields are calculated on a staggered grid using fully implicit formulations. More specifically, the horizontal and vertical velocities are defined in the centers of the horizontal and vertical cells faces, respectively, where the pressure and temperature fields are defined in the cell centers (see description in Tsibidis et al. [46]). Similarly, all temperature-dependent quantities (i.e., viscosity, heat capacity, density, etc.) are defined in the cell centers. While second-order finite difference schemes appear to be accurate for NP=1, where the surface profile has not been modified substantially, finer meshes and higher-order methodologies are performed for more complex profiles [51, 52]. Furthermore, techniques that assume moving boundaries (i.e. solid-liquid interface) are employed [53]. 4. Experimental protocol Commercial polished samples of SiO2 of 99.9% purity and average thickness of 1 mm were used. An Yb:KGW laser source was used to produce linearly polarized pulses with pulse duration equal to 170 fs, 1 kHz repetition rate, at 1026 nm central wavelength and 56μm Gaussian spot diameter. Beams with radial and azimuthal polarisation were generated using an s-waveplate. Following the s-waveplate, the beams were focused on the sample via an achromatic convex lens of 100mm focal length. The samples were fixed onto a 3-axis motorized stage and positioned perpendicular to the incident beam. The number of pulses was controlled by an electromagnetic beam shutter and all irradiations were performed in ambient environment. The morphology of the laser- induced structures has been characterized by scanning electron microscopy (JEOL JSM- 7500F). The morphological features of the structures were computed by a two dimensional fast Fourier transform analysis of the respective SEM images using the Gwyddion software (See Supplementary Material for a more detailed description). The experimental investigation for the period analysis of the irradiated laser spots was conducted for multiple experiments, for both azimuthal and radial polarization states. In particular, the presented measured period values are not a product of a single experiment, but of five different experiments, performed under identical conditions. Moreover, for each experiment , at least four measurements of the periodicity at different areas of each 7 spot have been obtained. The standard deviation and the mean of the error bars, presented in the discussion, are the outcome of those measurements. 5. Results and discussion As repetitive irradiation of a material leads to a gradually deepening of the affected zone (an initially flat profile is transformed in a curved morphology that suggests that an increase of the irradiation pulses produce a larger incidence angle), the need to distinguish the influence in the energy absorption of the differently polarised beams, APB and RPB, suggests that it is imperative to estimate the reflectivity of the material at (a) (b) , for s- FIG. 1. (Color online) (a) Carrier density and (b) Reflectivity at z=0, x=y= and p- polarisation as a function of θi. (NP=1, τp=170fs, Ed=7J/cm2). various angles of incidence (up to the Brewster's angle that is approximately 550 for fused silica) as a spatially variable energy absorption will influence the overall density of the excited carriers. Hence, a generic scheme is developed to determine the amount of energy that is absorbed from each segment of the curved geometry as a result of exposure of the material to a sequence of laser pulses [45, 46]. The solution of Eqs.1-7 yields the spatio-temporal distribution of the produced excited carriers. The induced maximum carrier density and the associated reflectivity values for RPB and APB as a function of the angle of incidence are presented in Fig.1 (see also Supplementary Material). The simulation parameters are Ed=7J/cm2, τp=170fs, λL=513nm and 1026nm. It turns out that the produced excited carrier densities for RPB increases at larger values of θi (Fig.1a) that leads to a decreased reflectivity and therefore enhanced energy absorption from the irradiated material (Fig.1b). It also appears that for the above simulation parameters (i.e. that leads to excitation of relevantly small carrier densities [37]), the influence of the laser wavelength does not yield significant changes in the reflectivity (Fig.1b) despite the expected increase of the excited carrier density at smaller λL resulting from the larger photon energy. Nevertheless, a focus on an enlarged area (see Supplementary Material) illustrates that for both s- and p-polarised beams, the induced reflectivity becomes higher at smaller laser beam wavelengths. 8 0R/20204000.51i [Degrees]Ne [1021 cm-3] S-pol (513nm)P-pol (513nm)S-pol (1026nm)P-pol (1026nm)0204000.050.10.150.2i [Degrees]Reflectivity S-pol (513nm)P-pol (513nm)S-pol (1026nm)P-pol (1026nm) On the other hand, the produced excited electron densities are always lower than the OBT nob (i.e. nob=1.06×1021cm-3 and 4.25×1021cm-3 at λL=1026 nm and λL=513nm, respectively) which in previous works have been related to the damage threshold [10-12]. Simulation results for θi =0 (similar results follow for other angles) illustrate the evolution of the maximum electron density inside the bulk (z-axis at x=y= where the absorbed energy is maximum) (Fig.2a,b). Nevertheless, herein, a thermal criterion is used to introduce surface damage which is demonstrated through the development of lattice temperatures larger than the melting temperature of fused silica (~1988K). As seen in Fig.2c,d, the spatio-temporal lattice temperature shows that a substantially large part (a) (c) (b) (d) FIG. 2. (Color online) Evolution of maximum free electron density at x=y= λL=513nm and (b) λL=1026nm (distance from the spot centre). Evolution of maximum lattice temperature at x=y= for (a) λL=513nm and (b) λL=1026nm (distance R0 from the spot centre) (NP=1, τp=170fs Ed=7J/cm2). In (c) and (d) dashed- dotted and dashed lines indicate ablation and melting depths, respectively. of the material has undergone a phase transition. More specifically, there are two regions that require special attention, one that is related to ablation (TL>4000K) and a second with Tmelt<TL<4000K that describes material in a molten phase. While at λL=513nm, the two phenomena occur in the regions z≤290nm and 290nm≤z≤1.268μm (Fig.2c), for (a) 9 0R/20R/20R/20R/2Time [ps]Z-axis [m] 0.511.5024Ne [1021 cm-3]0.20.40.6L=513nmTime [ps]Z-axis [m] 0.511.5012345TL [103 K]12345L=513nmTime [ps]Z-axis [m] 0.511.5012345Ne [1021 cm-3]00.050.10.150.20.25L=1026nmTime [ps]Z-axis [m] 0.511.5012345TL [103 K]1234L=1026nmrespectively, for λL=1026nm a less pronounced damaged region is produced at z≤150nm and 150nm≤z≤1.228μm (Fig.2d) as a result of a smaller photon energy. The differences in the thermal effects for the two polarisation states and the significance of the angle of incidence (up to the Brewster's angle) are also depicted in Fig.3. The theoretical predictions indicate that the induced maximum lattice temperature for RPB is higher than the value computed for APB which is also in agreement with the trend the excited carrier densities follow (Fig.1a). This observation also underlines the more dominant role of p-polarisation in the production of thermal effects. On the other hand, the decrease of TL difference between the values for irradiation with 513nm and 1026nm laser beam for APB (compared to RPB) with increasing angle of incidence can be attributed to the increase of the reflectivity if s-polarised waves are used. Simulations show, also, that due to the induced larger lattice temperatures a deeper damaged region is produced for irradiation with RPB compared to APB. In Fig.4, it is shown that irradiation with a laser beam of λL=1026 nm at an angle θi=300, produces a molten volume of depth equal to 963 nm (for RPB) while for APB it is smaller (759 nm). Similar conclusions can be deduced at other angles and λL. Theoretical models and experimental observations have indicated that upon repetitive irradiation periodic structures are formed [30, 32, 45-47, 54, 55]. To present a mechanism that explains the development of periodic structures on fused silica for RPB and APB, simulations are carried out firstly to correlate the spatio-temporal distribution of the excited electrons with the anticipated LIPSS periodicity based on Sipe's theory [43]. The computations have been performed to estimate, also, the role of the angle of incidence and how the previously calculated carrier density and dielectric constant influences the average periodicity of the induced structures. The orientation and size of rippled structures are predicted by computing the efficacy factor for APB (Fig.5) and RPB (Fig.6) for θi=00, 100, 300, respectively, (similar conclusions can be drawn for other values of θi). The efficacy factor cross-line along kr (for s-polarised APB) and kθ (for p- polarised RPB) provide an estimate of the ripple periodicity due to electrodynamics. Periodicity changes are due to the variation of the electron density resulting from differences in reflectivity, energy absorption and curvature of the irradiated profile. Results for RPB and APB indicate low spatial frequency periodic structures (LSFPS) FIG. 3. (Color online): Maximum Lattice temperature dependence on polarisation and angle of incidence. 10 02040468i [Degrees]Lattice Temperature [103 K] AP (513nm)RP (513nm)AP (1026nm)RP(1026nm) are formed with an orientation always parallel to the polarisation of the electric field on the plane of incidence (Fig.5,6). A similar orientation was also predicted for linearly polarised beams [32, 37]. Results shown in Figs.5,6 are similar for λL=513nm and 1026nm as kr, kθ (the components of the LIPSS wave vectors kL along the r- and θ- axes) are normalised with the wavelength (the computed different carrier density does not produce substantially different shapes of the efficacy factor distribution). Interestingly, the efficacy factor field in the k-space resulting from the s-polarised beam does not exhibit a similar symmetry as in the case of the RPB that emphasises on the significant influence of the incidence angle. A similar behaviour has been observed in (a) (b) FIG. 4. (Color online): Evolution of maximum lattice temperature at x=y= for (a) APB and (b) RPB (distance R0 from the spot centre) (NP=1, θi=300, τp=170fs, λL=1026nm, Ed=7J/cm2). Dashed lines indicate maximum melting depths. previous studies in semiconductors [44]. The dependence of the computed periodicities derived from the relation of the efficacy factor and the incidence angle is illustrated in Fig.7. The presence of sharp points in the efficacy factor curves (Fig.5b,6b) indicate a remarkably strong absorption which therefore leads to the formation of LIPSS. It is shown that for both λL=513nm and 1026 nm, there are two types of periodic 11 0R/2Time [ps]Z-axis [m] 0.511.5012345TL [103 K]1234APB (=300)Time [ps]Z-axis [m] 0.511.5012345TL [103 K]1234RPB (=300)(a) (b) (c) (d) (e) (f) FIG.5. (Color online): APB: (a) Efficacy factor computation as a function of the components of kL (for ne=0.2913×1021 cm-3 , θi=00). (b) Efficacy factor along the black dashed line in (a). (c) Efficacy factor computation as a function of the components of kL (for ne=0.29128×1021 cm-3 , θi=100). (d) Efficacy factor along the black dashed line in (c). (e) Efficacy factor computation as a function of the components of kL (for 12 kkr -202-2-1012Efficacy Factor0.10.20.30.40.5s-polarisation-2-101200.20.40.60.8krEfficacy Factorkkr -202-2-1012Efficacy Factor0.10.20.30.40.5s-polarisation-2-101200.20.40.60.8krEfficacy Factorkkr -202-2-1012Efficacy Factor0.10.20.30.4s-polarisation-2-101200.20.40.60.8krEfficacy Factorne=0.3132×1021 cm-3, θi=300). (f) Efficacy factor along the black dashed line in (e). (λL=1026 nm). (a) (b) (c) (d) (e) (f) FIG.6. (Color online): RPB: (a) Efficacy factor computation as a function of the components of kL (for ne=0.2913×1021 cm-3 , θi=00). (b) Efficacy factor along the black 13 kkr -202-2-1012Efficacy Factor0.10.20.30.40.50.6p-polarisation-2-101200.20.40.60.8kEfficacy Factorkkr -202-2-1012Efficacy Factor0.10.20.30.40.50.6p-polarisation-2-101200.20.40.60.8kEfficacy Factorkkr -202-2-1012Efficacy Factor0.20.40.60.811.2p-polarisation-2-101200.20.40.60.81kEfficacy Factordashed line in (b). (c) Efficacy factor computation as a function of the components of kL (for ne=0.2933×1021 cm-3 , θi=100). (d) Efficacy factor along the black dashed line in (c). (e) Efficacy factor computation as a function of the components of kL (for ne=0.2932×1021 cm-3, θi=300). (f) Efficacy factor along the black dashed line in (e). (λL=1026 nm). structures that are formed; more specifically, subwavelength (Fig.7a) or suprwavelength (Fig.7b) sized structures are formed at larger angles. The experimental results shown in the discussion (Fig.8 and periodicity measurements in Fig.9) suggest that the second type of periodic structures is preferential. To correlate the periodic structure formation with the development of thermal effects, the ablation (by removing the lattice points with TL>4000K), carrier density, relaxation processes, phase transformation, Marangoni effects and resolification process are taken into account for every laser pulse. This is due to the fact that hydrodynamics is expected to contribute to the change of periodicity size (correction to the value Λ derived from the efficacy factor computation) of the final profile due to fluid movement [45, 56, 57]. In previous studies, where LSFPS were assumed to be formed through a surface plasmon (SP) excitation mechanism, a periodic function in the source term S in Eq.8 was incorporated that corresponded to the grating period of the structure [30, 46]. As a result, a spatially periodic carrier density, Te and TL were produced that led to a similarly periodic thermal response of the lattice system. By contrast, the simulation parameters used in this work yield carrier densities substantially lower than the value that would allow SP excitation in dielectrics (~1022 cm-3) which also conforms with experimental studies as such modes have never been observed in dielectrics [37]. Nevertheless, to evaluate the frequency of the produced structures upon repetitive irradiation, a spatially (a) (b) FIG. 7. (Color online): Computed periodicities Λ as a function of the incidence angle for structures with (a) large and (b) small Λ, respectively. (Ed=7J/cm2). periodic function was introduced based on the value computed by the efficacy factor calculation. Hence, the carrier densities, the inhomogeneous energy deposition into the irradiated material, the produced new efficacy factor-based periodicity (Λ) and the 14 020406050010001500i [Degrees] [nm] S-pol (513nm)P-pol (513nm)S-pol (1026nm)P-pol (1026nm)L=1026nmL=513nm02040600200040006000i [Degrees] [nm] S-pol (513nm)P-pol (513nm)S-pol (1026nm)P-pol (1026nm)L=513nmL=1026nmcontribution of the thermal effects were computed to estimate the periodicity at the new NP after the material has resolidified. The aforementioned scheme towards solving Eqs.1-10, allows a parametric investigation of the morphological features of the produced structures as a function of the laser wavelength, polarisation state, fluence, and number of pulses. In Fig.8a,b, simulation results are illustrated that show the induced morphology after irradiation with APB and RPB, respectively, for NP=5. On the other hand, SEM images for the two polarisation states are shown (Fig.8c,d) to provide a qualitative comparison. Notably, similar SEM images were produced using tightly focused beams (in the present work, the beams were not tightly focused) in a previous work in which the simulated intensity distribution was produced assuming longitudinal and transverse electric field for the beams [29]. In that work, the estimation of the magnitude of the electric fields was based on their values for RPB (i.e. strong longitudinal component) and APB, respectively [23]. In the present work (where the role of phase transition was particularly highlighted towards determining the final morphological profile), simulations show that periodic structures are formed either perpendicularly to the r-axis (APB) or the θ-axis (RPB) and they are more pronounced inside the crater (around ) where the energy deposition is higher. It is noted that, firstly, mass displacement [46], and, secondly, stress-related effects (ignored in these simulations) that are generated around the spot centre are expected to produce an elevated peak the x=y=0 region that leads eventually to the formation of a 'Mexican-hat' [30] (i.e. the colorbar shows there is a structure formed above the initially flat surface). On the other hand, the simulated rippled morphology for RPB is presented in Fig. 8b, which corresponds to the white rectangular regions A and B in Fig.8d (where the energy deposition is higher according to the spatial distribution of fluence). It is noted that simulation results do not produce a similar (stable and periodic) profile as the respective SEM image for RPB at small radii from the centre of the spot (Fig.8d). This is attributed to strong hydrodynamical forces that do not allow the formation of a regular profile for radii smaller than . Furthermore, it is possible that the induced enhanced turbulence that has been observed in those regions require a particular simulation approach to predict a stable and symmetrical profile at smaller radii, however, this is beyond the scope of this work. Nevertheless inside the two (perpendicularly to each other) boxes A and B the ripples are locally parallel to each other. Notably, in Fig.8b the orientation (locally) of the laser beam polarization vector is denoted by the black doubled vectors (the notation should not be confused with the direction of a linearly polarised beam). As illustrated in Fig.8b, the height profile distribution in A and B are symmetrical (a more detailed analysis of the height profile is illustrated in the Supplementary Material). To estimate the periodicity, two subsequent peaks of the rippled structures (maxima across the Z-axis in the black boxes in Fig.8b which correspond to the hills) were considered and the distance between them represent the periodicity of the structure. It is noted that the periodicities were computed for two subsequent structures around the position of maximum energy deposition. The computed periodicities for the two distinct polarisation states are 1023nm (for APB), and 990nm (for RPB), on average, respectively. It is noted that the maximum depth around the region where the energy deposition is higher, is larger for RPB that is justified by the induced larger lattice temperature. 15 0R/20R/2 (a) (b) B A (c) (d) FIG. 8. (Color online): (a) Surface profile for APB (one quadrant). (b) Surface profile for RPB (in white rectangular boxes A and B in (d)). SEM image for NP=5 for: (c) APB and (d) RPB. (Ed=7J/cm2, λL=1026nm). In (b), Black doubled vectors indicate the direction of the polarisation vector (locally) while the white doubled arrows in the rippled region indicate the periodicity. Results for the periodicity change as a function of the number of pulses for λL=513nm and λL=1026nm for Ed=7J/cm2 show that the ripple periodicity increases with increasing NP for APB in contrast to the periodicity due to RPB (Fig.9a). Although repetitive irradiation leads to a deeper profile and a larger angle of incidence for RPB compared to APB the produced carrier densities influence the efficiency of the energy absorption of the energy and leads to a shift of the sharp points of the efficacy factor to smaller kL for APB and therefore, larger wavelengths (Fig.7b) [37]. To evaluate the influence of the energy deposited on the material, a similar approach was ensued by keeping the number of NP constant (NP=20) while the fluence remained constant. It is noted that the periodicity of the produced periodic structures as a function of the fluence follows a similar monotonicity with the one described in the previous paragraph. This behaviour is 16 Y [m]X [m] 0102005101520Height [m]-1-0.501023nmY [m]X [m] 2.533.54681012141618Height [m]-1.4-1.2-1990nmRegion ARippleRippleY [m]X [m] 6810121416182.533.54Height [m]-1.4-1.2-1990nmRippleRippleRegion Balso attributed to the carrier density increase at larger Ed that produced a more efficient energy absorption at smaller kL (and therefore larger periodicities). (a) (b) (c) (d) FIG. 9. (Color online): Periodicity of the laser-induced structures as a function of NP (a) for λL=1026nm, (b) for λL=513nm for Ed=7/cm2. Periodicity of the laser-induced structures as a function of fluence (c) for λL=1026nm, (d) for λL=513nm for NP=20. To validate quantitatively the theoretical results, data based on experimental observations are also illustrated in Fig.9 for irradiation with laser beam of λL=1026nm (SEM images for λL=513nm were difficult to analyse due to the small size periodicity and therefore experimental data for this wavelength have not been included). It is evident that although the theoretical model predicts a small variation of the periodicities for APB and RPB, the experimental estimation of periodicity includes a large error which does not allow a conclusive physical interpretation on which polarisation type produces larger periodicities and structures. By contrast, the proposed theoretical mechanism correlates a distinct hierarchy of the periodicity values with respect to the polarization state (Fig.9a,b). This tendency is very interesting as it can reveal special features of processing of materials by using laser beams of various polarization types. Although, the discrepancy of the periodicities for the two cases is difficult to be distinguished experimentally, the trend can be sufficiently explained by the differences in the efficiency of the energy absorption and therefore the density of the excited carriers; hence, periodicity size for APB irradiation should lean above that which is produced using RPB (Fig.9a,b). It is evident that at decreasing angle of incidence, the difference between the 17 020406060080010001200Number of PulsesPeriodicity [nm] AP (Theory)RP (Theory)AP (Experiment)RP (Experiment)L=1026nm0204060350400450500550600Number of PulsesPeriodicity [nm] AP (Theory)RP (Theory)L=513nm51015600800100012001400Periodicity [nm] Fluence [J/cm2]AP (Theory)RP (Theory)AP (Experiment)RP (Experiment)L=1026nm51015300400500600Fluence [J/cm2]Periodicity [nm] AP (Theory)RP (Theory)L=513nmproduced periodicities due to the two polarisation states disappears as there is no energy absorption difference. Finally, the periodicity values appear to saturate at larger values of fluence and NP (Fig.9c,d) which has also been observed theoretically and experimentally in previous studies [37, 46]. To summarise, the proposed multi-scale methodology provides new insights into the mechanism that characterises laser-matter interaction as it can allow a parametric investigation of influence the laser beam parameters on the features of the induced morphological changes on the irradiated material. The approach emphasises on the role of the laser beam polarisation in the modulation of the morphological features of the irradiated material and it can provide a recipe for a systematic analysis of the arising possibilities in ultrafast laser-based micro- and nano-fabrication. It should be noted that in addition to the differences between the periodicities values (at these laser energies) of APB and RPB, one aspect that is also significant to underline is that the two polarisation states produce periodic structures with different orientation. Hence, a combination of these (or even more complex) states could be used to produce interesting biomimetic structures [27, 28] that can be used in a wide range of applications. 6. Conclusions In conclusion, we have performed a comparative study to explore and interpret the surface profile and the periodicity of the self-assembled periodic structures formed upon irradiation of fused silica with CVB femtosecond laser pulses at two laser beam wavelengths. It was shown that compared to RPB, APB lead to periodic structures with a larger periodicity than that produced with APB and this difference is more enhanced as the irradiation pulses/fluence increase. On the other hand, the two polarisation states lead to the formation of periodic structures with different orientation. This significant conclusion emphasises particularly the ability to control the size/orientation of the morphological changes via modulating the beam polarization; it is evident that, laser processing through control of use of beams with various polarisation states may provide novel types of surface and bulk structures with significant advantages for potential applications. Acknowledgement This work has been supported by the project LiNaBioFluid, funded by EU's H2020 framework programme for research and innovation under Grant Agreement No. 665337 and from Nanoscience Foundries and Fine Analysis (NFFA) -- Europe H2020-INFRAIA- 2014-2015 (Grant agreement No 654360). Funding is also acknowledged from the General Secretariat for Research and Technology (GSRT) and Hellenic Foundation for Research and Innovation (HFRI), No. 130229/I2. References 18 I. M. Burakov, N. M. Bulgakova, R. Stoian, A. Rosenfeld, and I. V. Hertel, Applied Physics a- N. S. Shcheblanov and T. E. Itina, Applied Physics a-Materials Science & Processing 110, 579 A. Y. Vorobyev and C. Guo, Laser & Photonics Reviews 7, 385 (2012). O. J. Allegre, W. Perrie, S. P. Edwardson, G. Dearden, and K. G. Watkins, Journal of Optics 14, H. B. Sun, Y. Xu, S. Juodkazis, K. Sun, M. Watanabe, S. Matsuo, H. Misawa, and J. Nishii, R. Taylor, C. Hnatovsky, and E. Simova, Laser & Photonics Reviews 2, 26 (2008). E. L. Papadopoulou, M. Barberoglou, V. Zorba, A. Manousaki, A. Pagkozidis, E. Stratakis, and C. M. Livitziis and S. Pissadakis, Optics Letters 33, 1449 (2008). L. Jiang and H. L. Tsai, International Journal of Heat and Mass Transfer 48, 487 (2005). L. Jiang and H. L. Tsai, Journal of Applied Physics 100, 023116 (2006). B. C. Stuart, M. D. Feit, S. Herman, A. M. Rubenchik, B. W. Shore, and M. D. Perry, Physical B. Chimier, O. Utéza, N. Sanner, M. Sentis, T. Itina, P. Lassonde, F. Légaré, F. Vidal, and J. C. W. J. Cai, A. R. Libertun, and R. Piestun, Optics Express 14, 3785 (2006). E. Bricchi, B. G. Klappauf, and P. G. Kazansky, Optics Letters 29, 119 (2004). L. Sudrie, M. Franco, B. Prade, and A. Mysyrowicz, Optics Communications 191, 333 (2001). H. G. de Chatellus and E. Freysz, Optics Letters 27, 1165 (2002). E. N. Glezer, M. Milosavljevic, L. Huang, R. J. Finlay, T. H. Her, J. P. Callan, and E. Mazur, [1] [2] [3] [4] [5] Optics Letters 21, 2023 (1996). [6] Optics Letters 26, 325 (2001). [7] [8] Fotakis, Journal of Physical Chemistry C 113, 2891 (2009). [9] [10] [11] [12] Review B 53, 1749 (1996). [13] Kieffer, Physical Review B 84, 094104 (2011). [14] Materials Science & Processing 81, 1639 (2005). [15] (2013). [16] [17] 085601 (2012). [18] [19] [20] [21] Mottay, Journal of Laser Micro Nanoengineering 8, 188 (2013). [22] 78, 4713 (1997). [23] [24] [25] 72, 109 (2001). [26] M. Beresna, M. Gecevicius, P. G. Kazansky, and T. Gertus, Applied Physics Letters 98, 201101 (2011). [27] Amoruso, Sci Rep-Uk 5, 17929 (2015). [28] [29] (2011). [30] [31] [32] (2016). [33] Materials Science & Processing 79, 1695 (2004). [34] [35] W. L. Erikson and S. Singh, Physical Review E 49, 5778 (1994). [36] [37] [38] Applied Physics a-Materials Science & Processing 68, 563 (1999). L. E. Helseth, Optics Communications 191, 161 (2001). B. Hafizi, E. Esarey, and P. Sprangle, Physical Review E 55, 3539 (1997). V. G. Niziev and A. V. Nesterov, Journal of Physics D-Applied Physics 32, 1455 (1999). R. Torres, T. Kaempfe, M. Delaigue, O. Parriaux, C. Honninger, J. Lopez, R. Kling, and E. T. Kuga, Y. Torii, N. Shiokawa, T. Hirano, Y. Shimizu, and H. Sasada, Physical Review Letters Q. W. Zhan, Adv Opt Photonics 1, 1 (2009). K. S. Youngworth and T. G. Brown, Optics Express 7, 77 (2000). S. Quabis, R. Dorn, M. Eberler, O. Glockl, and G. Leuchs, Applied Physics B-Lasers and Optics E. Skoulas, A. Manousaki, C. Fotakis, and E. Stratakis, Sci Rep-Uk 7, 45114 (2017). C. Hnatovsky, V. Shvedov, W. Krolikowski, and A. Rode, Physical Review Letters 106, 123901 G. D. Tsibidis, E. Skoulas, and E. Stratakis, Optics Letters 40, 5172 (2015). B. Rethfeld, Physical Review Letters 92, 187401 (2004). G. D. Tsibidis, E. Skoulas, A. Papadopoulos, and E. Stratakis, Physical Review B 94, 081305(R) K. Sokolowski-Tinten and D. von der Linde, Physical Review B 61, 2643 (2000). S. Höhm, A. Rosenfeld, J. Krüger, and J. Bonse, Journal of Applied Physics 112, 014901 (2012). S. Nolte, C. Momma, G. Kamlage, A. Ostendorf, C. Fallnich, F. von Alvensleben, and H. Welling, J. JJ Nivas, S. He, A. Rubano, A. Vecchione, D. Paparo, L. Marrucci, R. Bruzzese, and S. S. S. Mao, F. Quere, S. Guizard, X. Mao, R. E. Russo, G. Petite, and P. Martin, Applied Physics a- P. Balling and J. Schou, Reports on Progress in Physics 76, 036502 (2013). 19 S. I. Anisimov, Kapeliov.Bl, and T. L. Perelman, Zhurnal Eksperimentalnoi Teor. Fiz. 66, 776 H. M. Vandriel, Physical Review B 35, 8166 (1987). J. K. Chen, D. Y. Tzou, and J. E. Beraun, International Journal of Heat and Mass Transfer 48, 501 E. G. Gamaly, A. V. Rode, B. Luther-Davies, and V. T. Tikhonchuk, Physics of Plasmas 9, 949 J. E. Sipe, J. F. Young, J. S. Preston, and H. M. van Driel, Physical Review B 27, 1141 (1983). J. Bonse, M. Munz, and H. Sturm, Journal of Applied Physics 97, 013538 (2005). G. Tsibidis and E. Stratakis, Journal of Applied Physics 121, 163106 (2017). G. D. Tsibidis, M. Barberoglou, P. A. Loukakos, E. Stratakis, and C. Fotakis, Physical Review B [39] (1974 [Sov. Phys. Tech. Phys. 11, 945 (1967)]). [40] [41] (2005). [42] (2002). [43] [44] [45] [46] 86, 115316 (2012). [47] [48] [49] (2008). [50] [51] [52] (1998). [53] M. Zerroukat and C. R. Chatwin, Journal of Computational Physics 112, 298 (1994). [54] [55] Rep-Uk 7, 12306 (2017). [56] M. Barberoglou, G. D. Tsibidis, D. Gray, E. Magoulakis, C. Fotakis, E. Stratakis, and P. A. Loukakos, Applied Physics A: Materials Science and Processing 113, 273 (2013). [57] J. Bonse, J. Krüger, S. Höhm, and A. Rosenfeld, Journal of Laser Applications 24, 042006 (2012). A. Rudenko, J. P. Colombier, S. Hohm, A. Rosenfeld, J. Kruger, J. Bonse, and T. E. Itina, Sci G. D. Tsibidis, C. Fotakis, and E. Stratakis, Physical Review B 92, 041405(R) (2015). H. Shibata, A. Suzuki, and H. Ohta, Mater. Trans. 46, 1877 (2005). H. J. Wang, W. Z. Dai, and L. G. Hewavitharana, International Journal of Thermal Sciences 47, 7 R. Kelly and A. Miotello, Applied Surface Science 96-98, 205 (1996). Y. Morinishi, O. V. Vasilyev, and T. Ogi, Journal of Computational Physics 197, 686 (2004). Y. Morinishi, T. S. Lund, O. V. Vasilyev, and P. Moin, Journal of Computational Physics 143, 90 G. D. Tsibidis, E. Stratakis, P. A. Loukakos, and C. Fotakis, Applied Physics A 114, 57 (2014). 20
1710.10758	2	1710	2017-11-14T08:56:16	Optimization of honeycomb battery package based on space mapping algorithm	[ "physics.app-ph" ]	A new honeycomb battery package structure is designed and optimized in this study. It is a honeycomb structure which uses grid to reinforce the strength. To obtain the highly accurate finite element (FE) model, the material parameters of 18650 cylindrical Li-ion battery are identified by using optimization techniques based on flat compression test data. Due to the expensive cost of finite element evaluation, the space mapping (SM) algorithm is suggested to optimize the structure of the package. Compared with other space mapping algorithms, the coarse model of space mapping in this work is based on a pseudo-plane-strain model. Moreover, to guarantee the reliability, the mean and variance values of battery stress are used to be the objective function. The final optimum solution is obtained in 3 days, and it shows the magnitude of stress and the distribution of stress are improved significantly compared with initial structure. Moreover, the computational cost of optimization for the problem is decreased greatly.	physics.app-ph	physics	Optimization of honeycomb battery package based on space mapping algorithm Wenquan Shuai1, Xin Luo1, Hu Wang1,2, 1State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha, China 2Joint Center for Intelligent New Energy Vehicle, Shanghai, China Jian Wang3 ABSTRACT A new honeycomb battery package structure is designed and optimized in this study. It's a honeycomb structure which uses grid to reinforce the strength. To obtain the highly accurate finite element (FE) model, the material parameters of 18650 cylindrical Li-ion battery are identified by using optimization techniques based on flat compression test data. Due to the expensive cost of finite element evaluation, the space mapping (SM) algorithm is suggested to optimize the structure of the package. Compared with other space mapping algorithms, the coarse model of space mapping in this work is based on a pseudo-plane-strain model. Moreover, to guarantee the reliability, the mean and variance values of battery stress are used to be the objective function. The final optimum solution is obtained in 3 days, and it shows the magnitude of stress and the distribution of stress are improved significantly compared with initial structure. Moreover, the computational cost of optimization for the problem is decreased greatly. Email: [email protected], Tel 86-731-88821417, Fax 86-731-88821807 Keyword: Li-ion battery package; Optimization; Simulation; Space mapping(SM) algorithm 1. Introduction The Li-ion battery is considered to be an ideal electric vehicle battery due to its higher energy density, higher power density and longer cycle life. The safety of the cell should be more valued in its application. When the cell is abused, it might explode or get fire due to internal short circuit [2]. These abuses include mechanical abuse, overcharge, over-discharge and overheating. In this work, only mechanical abuse is concerned. In order to understand the mechanism of internal short circuit, experiments are conducted. However, these experiments are ruinous and unrepeatable. Therefore, FE method is commonly used to simulate and predict the behavior of battery. However, it is difficult to develop the constitutive model of Li-ion battery when building the FE models. In order to solve the problem, a lot of researches had been done. For examples, Sahraei et al., (2012a) performed mechanical test on pouch battery under five loading conditions and used compressible foam to build FE models of pouch battery. Sahraei et al., (2012b) modeled 18650Li-ion battery and obtained the short circuit detection criterion under mechanical abuse condition. They also used the compressible foam to build FE models of cylinder battery but the material parameters were obtained from the strain-stress curve of pouch battery. Wierzbicki and Sahraei (2013) constructed FE models of 18650 Li-ion battery and studied the homogenized mechanical properties of the battery. They conducted the flat compression test and used the principle of virtual work to obtain the strain-stress curve of cylinder battery. Greve and Fehrenbach (2012) performed quasi-static mechanical test on cylindrical battery and built a macro- mechanical FE model to simulate the deformation and short circuit initiation of cylindrical battery. Ali et al., (2013) used computational models of Li-ion battery to simulate the behavior of battery under constrained compression tests. Sahraei et al., (2014) tested three types of pouch battery ranging from small cells to large cells under several different global and local compression conditions to obtain their mechanical property. They also constructed models for these three types of battery and investigated conditions leading to internal short circuit in the cells. Sahraei et al., (2016) developed a micro model of representative volume element (RVE) for Li-ion battery to study the failure mechanisms of internal component under complex loading scenarios. The available model for applying can be created based on these achievements. In this work, the crushable foam constitutive model is used to model the FE model of 18650 battery and the material parameters are obtained by inverse identification method. In this study, the new battery pack structure is suggested. Compared with the typical honeycomb structure, the distinctive characteristic of this structure is adopting a grid to reinforce the strength [1]. It is well known that typical honeycomb structure is an excellent function structure which has outstanding crashworthiness and energy-absorption property. Therefore it has been wildly used and investigated. Yamashita and Gotoh (2005) utilized numerical simulation to analyze the effect of honeycomb cell shape and foil thickness on crush strength of honeycomb which is compressed in the longitudinal direction. Ivañez et al., (2017) studied the influence of structure size and material properties for energy-absorption property of honeycomb core. Zhang et al,. (2017) investigated the indentation response and energy absorption of honeycomb sandwich panels under drop-weight impact behavior and obtained some semi-empirical formulations under different impact energy. However, the typical honeycomb structure is too soft to be used as battery package structure directly. Therefore, its strength should be reinforced by using grid, and the largest difference from typical honeycomb is that the honeycomb is not applied as buffer structure of battery package. The batteries are placed into its hexagon cells, so that the special hexagon cell is utilized to protect internal battery. In order to obtain better performance of battery package structure, optimization should be conducted. However, the computational cost of FE model is too high. It results in that typical direct optimization algorithm cannot be used in this problem. Space mapping (SM) algorithm is an efficient optimization method, especially for the problems of high calculating cost. The SM algorithm establishes two types of model which are fine model and coarse model, respectively. Fine model is a high accurate model or evaluation, which can achieves a more accurate result but requires highly computational cost. Coarse model is much simpler and achieves a result efficiently, but the accuracy of the result is lower. The SM would establish a mapping relation between response of fine model and coarse model. The method can get the optimized result of coarse model in a short time, and then obtain the optimized result of fine model through the mapping relation. Leary et al., (2001) used constraint mapping method to optimize an expensive model. The method mapped the coarse model constraints so that the coarse model constraints could approximate the fine model constraints. It is accurate to apply to an expensive model whose real objective varies little but constraints change signally. Redhe et al., (2002) used the SM algorithm which uses surrogate models and response surface method (RSM) to optimize a series of vehicle crashworthiness problems. Florentie et al., (2016) applied the SM algorithm in fluid-structure interaction problems. The output SM was used to optimize partitioned fluid-structure interaction problems, and resulted in the computational cost decreased to 50% in comparison with quasi-Newton method. Wang et al., (2017) integrated reanalysis method with SM method to propose an available optimization method. The Reanalysis-based space mapping method shows a significant improvement in the efficiency of expensive simulation-based problems. Due to the mentioned advantages above, the SM algorithm seems to be suitable for the problems with highly computational cost. Therefore, the SM is applied to structure optimization of honeycomb package in this study. The rest of paper is organized as follows. In Section 2, the material parameters of Li-ion battery are obtained by the inverse identification. What's more, the material parameters of Al alloy are obtained from uniaxial tensile test. In Section 3, the modeling method of battery package is validated by experiments and then the FE model of battery package is established. Subsequently, the trust region SM (TRSM) algorithm is introduced and used to optimize the battery package structure. In the final Section, the conclusions are made. 2. Identification of material parameters of Li-ion battery and Al alloy 6061-T6 2.1 Inverse identification of material parameters of battery 2.1.1 Inverse identification method For the battery package design, it is important to use highly accurate FE evaluation in the optimization procedure. The material parameters are the critical issue for simulation accuracy. Therefore, the hybrid numerical method is used to obtain the material parameters in this study. The flow chart of the inverse identification method is shown in Fig. 1. 2.1.2 Flat compression test of 18650 battery The Li-ion batteries studied in this work are commercial 18650 ternary polymer lithium batteries. Their capacity is smaller than pouch battery's, only 2600mAh. It means that an electric vehicle needs more batteries. However, they are safer than pouch battery. The specifications of these 18650 batteries are listed in Table 1. The contribution of shell casing to total force is less than 1% in the flat compression test [5]. Therefore, whether there is shell casing or not is not important for the test. In order to measure the voltage of battery, the shell casing is retained, as shown in Fig. 2(a). The compression test is performed in INSTRON 150kN Universal Test Machine, and digital image correlation (DIC) is used to measure the local strain (Fig. 2(c)). The UNI-T U322 Thermometer and UNI-T UT61C Multimeter are used to obtain temperature and voltage respectively. Fig. 1 The flow chart of the inverse identification method The deformation rate is set as 1mm/min. When the internal short circuit happens, the test should be stopped. The response of internal short circuit is that the load and the voltage drops rapidly and the temperature increases simultaneously [2]. The force, displacement, temperature and voltage are measured in the test. After accomplishment of the test, it is observed that the electrolyte of battery is leaked and the shell casing is flawed. The deformed battery is shown in Fig. 2(d). Figure. 3 shows the data which is obtained in the test. It reveals the load drops rapidly from 5.7×104N when the displacement is about 6.3mm. Meantime, the Inverse identification methodmin S=abs(f(x)-F(x)) x={x1,x2,x3,……x100}Subject to : xil≤xi≤xiu for i=1:100Flat compression test of 18650 batteryConstruction of a FE model of 18650 batteryOutput of the force-displacement curve f(x)Output of the force-displacement curve F(x)Is the S minimumChange of the material parameters of batteryNoDetermination of the material parameters of batteryYesEnd of the inverse identification of material parameters of batterytemperature start to increase significantly and the voltage drops from 3.6V to 0.096V in a short time. It means that internal short circuit has happened. Table 1 Specifications of the 18650 battery Parameters Nominal capacity Values 2600mAh Size Weight Nominal voltage material Operating temperature 18mm18mm65mm 48g 3.7V Li(NiCoMn)O2 -20℃ 60℃ Fig. 2 (a) The entity of battery (b) The FE model of battery (c) The experimental facilities of the flat compression test (d) The deformed battery Fig. 3 The result of the flat compression test 2.1.3 FE model of 18650 battery A FE model of 18650 battery is constructed in ABAQUS. The contribution of shell casing to total force is too small, therefore, the whole cell is regarded as a homogeneous and isotropous material and the shell case is ignored. The crushable foam is used as the constitutive model of the material in the plastic phase[2]. To simulate the real compression test, the model of the cell is compressed by two rigid shells. The cell is modeled by using 8-node linear brick and reduced integration solid elements (C3D8R). The simulation parameters are detailed in Table 2. The whole model is shown in Fig. 2(b). The contact force between rigid board and battery is exported as history output. Table 2 The simulation parameters of battery FE analysis Parameters Global seed size of the cell Global seed size of the rigid shell Values 1mm 5mm Simulation step Loading speed Friction coefficient Dynamic explicit; 0.12s 50mm/s 0.1 2.1.4 Inverse identification of material parameters of battery The material parameters of the cell are obtained by inverse identification method. The two force-displacement curves are obtained by experiments and simulation, respectively. The objective function is minimizing the error between two force-displacement curves, and there are 11 design variables in this problem. The mathematical expression can be formatted as follows (1) where is the nearest displacement from the in the experimental displacement data. represents the design variables where is elasticity modulus, is Poisson ratio, is yield stress in harden curve which shown in Table 3. represents the contact force when the displacement is in simulations. represents the force when the displacement is in experiments. Each FE evaluation outputs 200 sets of data of contact force and time, and the interval between two sources is about 0.03mm. Similarly, the experiment outputs 4312 sets of data of force and displacement. The interval between two sources is nearly 0.002mm. The 100 sets of simulation data are selected to calculate the error. To obtain reliable result, the global evolutionary optimization method, Genetic algorithm (GA) is employed. The final result is listed in Table 4 and corresponding curves are demonstrated in Fig. 4. It can be found that the curves derived from experimental data and simulation are well matched. is a criterion which is applied in fitting nonlinear curves, as shown in Eq. (2) where represents the true value and represents the predicted value. The criterion is 0.9 in this work. It means the inversed parameters are well matched with experimental data and can be used for subsequent simulation. (2) Table 3 The parameters of harden curve of material of battery Yield stress/MPa Plastic strain σ0=7.6958 σ1 σ2 σ3 σ4 σ5 σ6 σ7 σ8 σ9 0 0.010 0.025 0.065 0.100 0.125 0.165 0.215 0.255 0.325 Table 4 The material parameters of battery Parameters Values Unit 544.8866 MPa E ν σ1 σ2 σ3 σ4 σ5 σ6 σ7 σ8 σ9 0.0234 9.4787 MPa 7.7454 MPa 7.8069 MPa 7.7137 MPa 7.6989 MPa 7.7091 MPa 10.1857 MPa 106.6688 MPa 161.4426 MPa Fig. 4 The force-displacement curves of experiments and simulation 2.2 The material parameters of Al alloy 6061-T6 2.2.1 Uniaxial tensile test of Al alloy 6061-T6 The material of the wall of battery package is Al alloy 6061-T6. In order to obtain the material parameters of Al alloy 6061-T6, the uniaxial tensile test is conducted. The size of tensile test coupon is obtained according to GB/T 228- 2010, as shown in Fig. 5(a). The uniaxial tensile test is conducted in INSTRON 150kN Universal Test Machine. To obtain the strain of tensile test coupon during the test, the extensometer is used. The tensile rate is 2mm/min. Figure. 5(b) shows the fractured tensile test coupon. The engineering strain- stress curve is obtained from test. It should be transformed to true stress and strain. The transformation formula is (2) The true strain- stress curve is shown in Fig. 6(a). Fig. 5 (a) The size of tensile test coupon (b) The fractured tensile test coupon and corresponding FE model baFig. 6 (a) The true strain- stress curve of tensile test (b) The result curves of parameters validation 2.2.2 The validation of material parameters of Al alloy 6061-T6 According to the above illustration, the material parameters can be obtained from the true strain- stress curve. The material parameters of Al alloy 6061-T6 are summarized in Table 5. In order to validate the material parameters, the FE model of tensile sample is built, as shown in Fig. 5(b). The strain-stress curve of elements of FE model is output to validate the accuracy of parameters. The validation result is shown in Fig. 6(b). The result indicates the material parameters of Al alloy are accurate. baTable 5 The material parameters of Al alloy Parameters Elasticity modulus /MPa Poisson ratio Yield strength/ MPa Density/ (T/mm3) value 71.275 0.33 241.5 2.9e-9 Plastic curve Yield stress/MPa Plastic strain 241.5 263.0 278.8 318.8 346.7 374.5 388.8 423.8 464.3 473.6 0 0.0069 0.0217 0.0921 0.1408 0.1914 0.2181 0.2862 0.3728 0.4078 3. Design and optimization of battery package 3.1 The validation of modeling method of battery package 3.1.1 Quasi-static compression test of battery package The computational cost would be expensive if the whole package is modeled by solid elements. We hope to use a specific simplified FE model to complete the package design. In order to validate the modeling method of battery package, a quasi-static compression test is conducted to compare with FE model. The structure of the test battery package without battery is shown in Fig. 7(a). Figure. 7(b) shows the test package sample. The quasi-static compression test is conducted in INSTRON 2000kN Universal Test Machine. The compression rate is 1mm/min. The deformed sample is shown in Fig. 7(d). Fig. 7 (a) The structure of the battery package (b) The test package sample (c) The FE model of test sample (d) The stress contours of FE model and the deformed sample 3.1.2 The validation of FE model In order to improve the efficiency of simulation, the FE model is simplified and different from the entity. The FE model is divided into two parts. One is the honeycomb core which is modeled by 4-nodes shell element (S4RSW). Another is the reinforce grid which is modeled by solid element (C3D8R). These two parts are tied together by using tie constraint. The whole model includes 428805 solid and 96980 shell elements. The friction coefficient of model is 0.17. The test is quasi-static compression test. In order to improve its computational efficiency, the compression rate is set as 80mm/s. It can be observed that the kinetic energy is less than 5% of the total energy, therefore, it is still a quasi-static process. Figure. 7(c) shows the FE model of test sample. The compression displacement is 3mm. The stress contours of FE model is shown in Fig. 7(d). By comparing the two figures in Fig. 7(d), it can be observed that the deformed shape of test and simulation are similar. The bottom honeycomb cores of test and simulation are both buckling outward and the grid both bend. Therefore, the modeling method can be accepted in this work. 3.2 The FE model of battery package The structure of optimized battery package is shown in Fig. 6(a). The FE model of package also consisted of two parts which are honeycomb core and grid. The honeycomb core is modeled by using 4-node shell elements (S4RSW), its thickness is 0.9mm. The grid is modeled by using solid elements (C3D8R). The whole model includes 672165 solid and 135200 shell elements. The friction coefficient of model is 0.17. Figure. 8(b) shows the FE model of the whole battery package. Figure. 9(a) shows the stress contours of the package in 0.018s, 0.036s, 0.054s and 0.072s, respectively. These contours show the stress of model increases with the time increases. In order to show the integral stress contours of battery only, the grid and honeycomb are excluded from the simulation output and only the battery is output as shown in Fig. 9(b). The maximum stress of battery is 32.1733MPa which is still in safe loading range. The FE model of battery package can simulate the stress-strain behaviors of the battery package accurately, however, the computational cost is high. It needs 2.5 hours to finish a simulation when using 8 CPUs (Inter(R) Xeon(R) CPU,3.40GHz). Therefore, the typical optimization method cannot be available for optimizing the battery package. However, the SM algorithm can do well in this kind of expensive problems. In later section, the TRSM algorithm is introduced and applied in optimization of the battery package. Fig. 8 (a) The structure of optimized battery package (b) The FE model of the whole battery package (c) The stress contours of the batteries (d) The coarse FE model Fig. 9 The stress contours of fine FE model in different time (a) The stress contours of the whole package (b) The stress contours of the battery ab3.3 Trust region space mapping 3.3.1 Introduction of TRSM algorithm The TRSM is a mainstream SM algorithm. Its characteristic is setting a trust region for design parameter of fine model. The trust region of every cycle is used to judge whether the algorithm is convergent. Then, the TRSM will be introduced in detail. Assuming is the design parameter of fine model and is the objective function of fine model. Similarly, the symbol and represent the design parameter and the objective function of coarse model, respectively. The optimization problem of the fine model is expressed as (3) The constraint function of the fine model is which is expressed as (4) (5) where and are the lower and upper bounds of constraint, respectively, and are the lower and upper bounds of the design parameter . The optimal solution of the fine model is expressed as (6) The optimal solution of the coarse model is expressed as (7) where and represent the design space of the fine model and the coarse model, respectively. The algorithm wants to establish the mapping relation between the two design spaces, that is . The establishment of the mapping relation is based on minimizing the residual. The expression is (8) where is residual, and is norm. According to the expression, if the residual is small enough, it follows (9) If the parameters of two spaces satisfy (10) a perfect mapping can be established. Therefore, the Eq. (6) can be transformed into (11) Now, the complex and expensive optimization problem is transformed into easy and cheap problem. The SM algorithm will solve Eq. (6) iteratively and solve Eq.(7) mainly at each iteration. At the kth cycles, the function is replaced with linear approximation which is expressed as (12) where , . donates approximation Jacobin matrix of function . is the size of the optimization range in the kth cycles. The specific steps of the algorithm are presented in Table 6. In every iteration, is accepted if it satisfies The update rule of is (13) (14) where . The update rule of is (15) The algorithm will converge if it satisfied any one of these conditions: (16) where both and are constant coefficients, is the maximum number of cycles Table 6 Descriptions of TRSM algorithm Given ; set ; ; ; Evaluate ; Evaluate ; While 1: Obtain ; If algorithm converges: Break; End; ; Evaluate ; Update and ; Evaluate ; If algorithm converges: Break; End; ; End; 3.3.2 Numerical example In order to validate the superiority of TRSM algorithm, a simple function example is conducted. The fine model is expressed as (17) The coarse model is expressed as (18) These two models are shown in Fig. 10. It is obvious that these two models are similar, but the coarse model is simpler. The objective function is (19) There are two methods which are applied in the optimization problem. The one is TRSM algorithm whose optimization algorithm of coarse model is GA. The another one is GA which is used to optimize the fine model directly. When using TRSM algorithm to optimize the example, there are only one fine model and 230 coarse models are calculated. When the GA is used to optimize directly, the algorithm converges after 50 iterations. The result of two methods are shown in Table. 7. According to the result, the error between two optimal solutions which are obtained by TRSM and GA is small, and the computational cost of TRSM is much less than GA. Therefore, the TRSM algorithm is superior when the efficiency and accurate are taken into consideration together. Fig. 10 The fine model and the coarse model Table 7 The result of optimization of two algorithms Iterations Time/s TRSM algorithm 7.5915 -1.1273 1 0.039 GA 7.4248 -1.137 50 2.23 3.4 The coarse FE model of battery package In SM algorithm, an easier and cheaper FE model is needed which called coarse FE model. As shown in Fig. 8(c), the stress of battery is almost changeless in the axial direction of battery. Therefore, the FE model can be simplified as a pseudo- plane-strain model. The coarse FE model is shown in Fig. 8(d). All the translational degree of freedoms (DOFs) along the Z axis and the rotational DOFs along the X and Y axis are constrained. The simulation parameters of the coarse FE model are the same as the fine model. The whole model includes 10341 solid and 2080 shell elements. Figure. 11(a) shows the stress contours of coarse FE model in 0.018s, 0.036s, 0.054s and 0.072s, respectively. In order to show the integral stress contours of battery, the grid and honeycomb are excluded from the simulation output and only the battery is output as shown in Fig. 11(b). The maximum stress of battery is 15.1035MPa. The deformation of coarse FE model is similar to the deformation of fine FE model by comparing these two stress contours which in the same time. Because the coarse FE model needs only 5 minutes to finish a simulation when using the same CPUs and its stress states is similar to fine FE model, it can be an excellent coarse model of SM algorithms. Fig. 11 (a) The stress contours of coarse FE model (b) The stress contours of the battery of coarse model in different time ab3.5 Optimization of battery package In the optimization problem, the design variables are the lengthways thickness of grid , the lateral thickness of grid and the thickness of honeycomb core . The range of these design variables are listed in Table 8. Because the stress of battery increases with time, the stress state of battery in the final time is concerned. In order to improve the magnitude and distribution of stress of batteries, the optimized target is suggested to minimize the mean and the variance of stress of all the batteries which in the final time. It is a multi-objective problem. The objective function is (20) where is the stress of all batteries in 0.072s. GA is used to optimize the coarse model. Ultimately, GA obtains a series of feasible solutions, and the solution with minimum square sum of mean and variance is selected be the optimum solution. Then, the optimum solution obtains from coarse model is used to build the mapping relation with fine model and then, we will obtain the optimum solution of fine model. During the whole optimization, two fine models and 797 coarse models are calculated. The optimization costs 3 days to obtain a global optimal solution with 8 CPUS. With the same time, only 29 fine models could be calculated and it is impossible to obtain the global optimal solution. If the fine model was optimized by using GA, 230 fine models would be calculated at least and it would cost 24 days. Therefore, the TRSM algorithm is greatly helpful for the efficiency of this work. The optimum solution of fine model is listed in Table 9. The design variable and objective history for the optimization problem are listed in Table 10. The norm of of the first and the second cycle are shown in Fig. 12. According to the result listed in Table 10, the convergence condition is satisfied. What's more, due to the convergence judgment in the second cycle, the second cycle is interrupted. The algorithm obtains three ultimately. In the second cycle, the norm of is less than the minimum of norm of . According to the Eq. (9), it can indicate the objective value is worse if and the globally optimal solution is not in the optimal range of the second cycle. Figure. 12 reveals that the result of the second cycle is much worse than the result of the first cycle. Therefore, the algorithm converges, and the optimal solution of first cycle is the globally optimal solution. The initial mean and variance of stress of fine model in 0.072s are 5.3231MPa and 5.9944, respectively. The optimized mean and optimized variance of stress of fine model are 3.5836MPa and 4.3904, respectively. Correspondently, The initial mean and variance of stress of coarse model in 0.072s are 5.2257MPa and 5.2641, and the optimized mean and variance of stress of coarse model are 3.6988MPa and 4.4957. Obviously, they close to the optimized solution of fine model. It means the fine model and the coarse model are similar essentially, therefore, the TRSM algorithm is greatly suitable for this work. Figure. 13 (a) shows the stress contours of the optimized fine model in 0.018s, 0.036s, 0.054s and 0.072s, respectively. Figure. 13(b) only shows the stress contour of coarse model in 0.072s. These stress contours only output the stress of battery. According to the figures, it is clear to know that the stress state of fine model and coarse model in 0.072s are similar. The maximum stress of batteries of fine model is 12.9642MPa in 0.072s. It drops by 59.71% compared to the maximum stress of initial fine model which is 32.1733MPa. The optimization consequence reveals that the magnitude and distribution of stress of batteries have been improved significantly. Table 8 The range of design variables Parameters/mm The lengthways thickness of grid The crosswise thickness of grid The thickness of honeycomb core Range [0.5,5] [0.5,5] [0.02,1] Table 9 The optimum solution of fine model Parameters/mm Values 2.0184 The lengthways thickness of grid The crosswise thickness of grid 0.8023 The thickness of honeycomb core 0.03865 Table 10 The design variable and objective history for the optimization problem Cycles Parameters /mm [1.6676 0.9112 0.0422] [3.5765 4.6412] / [2.01840 0.80230 0.03865] [3.5643 4.5399] [3.5991,4.4420] 0 1 2 [3.5263 4.6484] [3.5836 4.3904] [1.4514 0.9041 0.0503] / [3.6112,4.5736] / Fig. 12 The norm of of the first and the second cycle Fig. 13 (a) The stress contours of the optimized fine model (b) The stress contours of the optimized coarse model 4. Conclusions A new honeycomb battery package structure is designed in this study. compared with poplar honeycomb structure, the proposed honeycomb structure uses a grid to reinforce its strength. In order to construct the FE model of the package, the material parameters of battery and Al alloy have to be obtained. Therefore, 0.072sabthe flat compression test of 18650 battery and the uniaxial tensile test of Al alloy 6061-T6 are conducted. And then according to the data that obtained from flat compression test of battery, the material parameters of battery are identified by inverse identification method. Subsequently, the quasi-static compression test of battery package is conducted to validate the modeling method of battery package. The verification results reveal that the deformation shape of the test package sample and the deformation shape of FE model are similar. Therefore, the FE model of battery package is accurate. In order to obtain the optimized structure of package, TRSM algorithm is used to optimize the structure. Compared with other SM algorithms, the coarse model of SM is based on a pseudo-plane-strain model. Moreover, to guarantee the reliability, the mean and variance values of battery stress are used to be the objective function. Optimized consequence shows the maximum stress of inner battery is decreased from 32.1733MPa to 12.9642MPa, the mean of battery is decreased from 5.3231MPa to 3.5836MPa and the variance is decreased from 5.9944 to 4.3904. The optimization effect is quite remarkable. What's more, the computational cost of the optimization decreases significantly. Generally, the highlights of this work can be summarized as follows:  A new structure of battery package is designed. It is a kind of honeycomb structure. In order to reinforce its strength, an Al alloy grid is used to be the framework.  The material parameters of cylindrical battery are obtained by inverse identification method instead of replacing by material parameters of pouch battery.  Due to the expensive cost of FE evaluation, the typical optimization method cannot be available for optimizing the battery package. Therefore, the TRSM algorithm can be used to optimize the structure and obtain an ideal result in an acceptable time. Compared with other application of the TRSM, the coarse model is specified modeled for the full FE package model. 5. Acknowledgments This work has been supported by National Key R&D Program of China 2017YFB0203701, Project of the Key Program of National Natural Science Foundation of China under the Grant Numbers 11572120 References [1] Sun Z, Shi S, Guo X, et al. On compressive properties of composite sandwich structures with grid reinforced honeycomb core. Composites Part B Engineering, 2016, 94:245-252. [2] Meier J D. Material characterization of high-voltage lithium-ion battery models for crashworthiness analysis. Cambridge, Massachussetts, Massachussetts Institute of Technology, 2013. [3] Sahraei E, Hill R, Wierzbicki T. Calibration and finite element simulation of pouch lithium-ion batteries for mechanical integrity. Journal of Power Sources, 2012, 201(3):307-321. [4] Greve L, Fehrenbach C. Mechanical testing and macro-mechanical finite element simulation of the deformation, fracture, and short circuit initiation of cylindrical Lithium ion battery cells. Journal of Power Sources, 2012, 214(4):377-385. [5] Sahraei E, Campbell J, Wierzbicki T. Modeling and short circuit detection of 18650 Li-ion cells under mechanical abuse conditions. Journal of Power Sources, 2012, 220(4):360-372. [6] Ali M Y, Lai W J, Pan J. Computational models for simulations of lithium-ion battery cells under constrained compression tests. Journal of Power Sources, 2013, 242(22):325-340. [7] Wierzbicki T, Sahraei E. Homogenized mechanical properties for the jellyroll of cylindrical Lithium-ion cells. Journal of Power Sources, 2013, 241(6):467-476. [8] Sahraei E, Meier J, Wierzbicki T. Characterizing and modeling mechanical properties and onset of short circuit forthree types of lithium-ion pouch cells. Journal of Power Sources, 2014, 247(2):503-516. [9] Sahraei E, Bosco E, Dixon B, et al. Microscale failure mechanisms leading to internal short circuit in Li-ion batteries under complex loading scenarios. Journal of Power Sources, 2016, 319:56-65. [10] Yamashita M, Gotoh M. Impact behavior of honeycomb structures with various cell specifications-numerical simulation and experiment. International Journal of Impact Engineering, 2005, 32(1–4):618-630. [11] Ivañez I, Fernandez-Cañadas L M, Sanchez-Saez S. Compressive deformation and energy-absorption capability of aluminium honeycomb core. Composite Structures, 2017, 174: 123-133. [12] Zhang D, Fei Q, Zhang P. Drop-weight impact behavior of honeycomb sandwich panels under a spherical impactor. Composite Structures, 2017, 168: 633-645. [13] Leary S J, Bhaskar A, Keane A J. A Constraint Mapping Approach to the Structural Optimization of an Expensive Model using Surrogates. Optimization and Engineering, 2001, 2(4):385-398. [14] Redhe M, Nilsson L. Using space mapping and surrogate models to optimize vehicle crashworthiness design//9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimization. 2002: 5536. [15] Florentie L, Blom D S, Scholcz T P, et al. Analysis of space mapping algorithms for application to partitioned fluid–structure interaction problems. International Journal for Numerical Methods in Engineering, 2016, 105(2):138-160. [16] Wang H, Fan T, Li G. Reanalysis-based space mapping method, an alternative optimization way for expensive simulation-based problems. Structural & Multidisciplinary Optimization, 2017, 55(6):2143-2157.
1907.06000	1	1907	2019-07-13T02:24:24	Detection of Influenza A Virus Nucleoprotein Through the Self-Assembly of Nanoparticles in Magnetic Particle Spectroscopy-Based Bioassays: A Method for Rapid, Sensitive, and Wash-free Magnetic Immunoassays	[ "physics.app-ph" ]	Magnetic nanoparticles (MNPs) with proper surface functionalization have been extensively applied as labels for magnetic immunoassays, carriers for controlled drug/gene delivery, tracers and contrasts for magnetic imaging, etc. Here, we introduce a new biosensing scheme based on magnetic particle spectroscopy (MPS) and the self-assembly of MNPs to quantitatively detect H1N1 nucleoprotein molecules. MPS monitors the harmonics of oscillating MNPs as a metric for the freedom of rotational motion, thus indicating the bound states of MNPs. These harmonics can be readily collected from nanogram quantities of iron oxide nanoparticles within 10 s. H1N1 nucleoprotein molecule hosts multiple different epitopes that forms binding sites for many IgG polyclonal antibodies. Anchoring IgG polyclonal antibodies onto MNPs triggers the cross-linking between MNPs and H1N1 nucleoprotein molecules, thereby forming MNP self-assemblies. Using MPS and the self-assembly of MNPs, we achieved the sensitivity of 44 nM (442 pmole) for detecting H1N1 nucleoprotein. In addition, the morphologies and the hydrodynamic sizes of the MNP self-assemblies are characterized to verify the MPS results. Different MNP self-assembly models such as classical cluster, open ring tetramer, chain model as well as multimers (from dimer to pentamer) are proposed in this paper. Herein, we claim the feasibility of using MPS and the self-assembly of MNPs as a new biosensing scheme for detecting ultralow concentrations of target biomolecules, which can be employed as rapid, sensitive, and wash-free magnetic immunoassays.	physics.app-ph	physics	Detection of Influenza A Virus Nucleoprotein Through the Self-Assembly of Nanoparticles in Magnetic Particle Spectroscopy-Based Bioassays: A Method for Rapid, Sensitive, and Wash-free Magnetic Immunoassays Kai Wu, Jinming Liu, Renata Saha, Diqing Su, Venkatramana D. Krishna, Maxim C-J Cheeran* and Jian-Ping Wang* Dr. K. Wu, J. Liu, R. Saha, Prof. J.-P. Wang Department of Electrical and Computer Engineering University of Minnesota, Minneapolis, MN 55455, United States Email: [email protected] D. Su Department of Chemical Engineering and Material Science University of Minnesota, Minneapolis, MN 55455, United States Dr. V.D. Krishna, Prof. M.C-J. Cheeran Department of Veterinary Population Medicine University of Minnesota, St. Paul, Minnesota 55108, USA E-mail: [email protected] Abstract Magnetic nanoparticles (MNPs) with proper surface functionalization have been extensively applied as labels for magnetic immunoassays, carriers for controlled drug/gene delivery, tracers and contrasts for magnetic imaging, etc. Here, we introduce a new biosensing scheme based on magnetic particle spectroscopy (MPS) and the self- assembly of MNPs to quantitatively detect H1N1 nucleoprotein molecules. MPS monitors the harmonics of oscillating MNPs as a metric for the freedom of rotational motion, thus indicating the bound states of MNPs. These harmonics can be readily collected from nanogram quantities of iron oxide nanoparticles within 10 s. H1N1 nucleoprotein molecule hosts multiple different epitopes that forms binding sites for many IgG polyclonal antibodies. Anchoring IgG polyclonal antibodies onto MNPs triggers the cross-linking between MNPs and H1N1 nucleoprotein molecules, thereby forming MNP self-assemblies. Using MPS and the self-assembly of MNPs, we achieved the sensitivity of 44 nM (442 pmole) for detecting H1N1 nucleoprotein. In addition, the morphologies and the hydrodynamic sizes of the MNP self-assemblies are characterized to verify the MPS results. Different MNP self-assembly models such as classical cluster, open ring tetramer, chain model as well as multimers (from dimer to pentamer) are proposed in this paper. Herein, we claim the feasibility of using MPS and the self-assembly 1 of MNPs as a new biosensing scheme for detecting ultralow concentrations of target biomolecules, which can be employed as rapid, sensitive, and wash-free magnetic immunoassays. 1. Introduction Influenza viruses belong to the family Orthomyxoviridae and contain single stranded, negative sense, segmented RNA genome. Based on their differences in nucleoprotein (NP) and matrix (M) protein, influenza viruses are classified into four types, A, B, C, and D.[1 -- 4] Influenza A viruses (IAV) are further sub-divided based on combinations of their surface glycoproteins hemagglutinin (HA) and neuraminidase (NA) into different subtypes. There are 18 different types of HA (H1 to H18) and 11 different types of NA (N1 to N11). Each IAV subtype is virus with one type of H and N combination (such as the H1N1 subtype investigated in this work). IAV infects many vertebrates including humans and is responsible for seasonal epidemics of acute respiratory illness known as influenza or flu. IAV poses a significant public health concern that causes substantial morbidity and mortality and has the ability to cause worldwide pandemics.[5,6] According to World Health Organization (WHO), influenza viruses are responsible for 250,000 to 500,000 deaths annually. Rapid, accurate and sensitive method for early diagnosis of IAV infections are critical for rapid initiation of antiviral therapy to control infection and to reduce the impact of possible influenza pandemics. IAV nucleoprotein (NP) is a basic protein that binds to viral RNA along with polymerase and is the most abundant component of the ribonucleoprotein (RNP) complex.[7] Each RNP consists of one genomic RNA segment associated with a single trimeric polymerase complex and multiple NP monomers.[8,9] In addition to binding single stranded RNA, IAV nucleoprotein has been shown to self-assemble to form large oligomeric complexes.[10] Moreover, nucleoprotein is well conserved among different IAV strains isolated from different host species with amino acid difference of less than 11%.[11] These properties make IAV nucleoprotein a good target for detection of multiple IAV subtypes. The conventional detection schemes of IAV can be divided into several categories based on the type of assays employed. Immunofluorescence assays use either direct or indirect fluorescent antibody staining techniques to detect IAV. The signal can be readout through fluorescent microscopes with sensitivities of 60 -- 80%. However, they are unable to distinguish different IAV subtypes.[12] Serological tests such as hemagglutination inhibition assays (HAI), microneutralization or virus neutralization assays (VN) and enzyme linked immunosorbent assays (ELISA) are the most commonly used techniques to detect influenza virus-specific antibody responses. But they usually need paired samples with strict requirements of the sample collection time.[13] Nucleic acid-based assays such as reverse transcription-polymerase chain reaction (RT-PCR) are based on the detection of DNA/RNA sequences of the viruses. Apart from the fact that long reaction time (1 -- 8 hr) is required, RT-PCR is one of the most expensive testing techniques.[14] To cut down the time of detection, rapid influenza diagnostic tests (RIDTs) have also been developed with RT-PCR and ELISA as the two gold standards.[15,16] Although the total testing 2 time is less than 30 min with high specificities (95 -- 99%), the sensitivities of the RIDTs are lower than other techniques (10 -- 70%) and false negative results should be considered.[17,18] Giant magnetoresistive (GMR) biosensors have also been reported for the detection of IAV both in lab-based and point-of-care platforms.[19 -- 21] However, this technique is limited by the complexity of nanofabricating the GMR sensors as well as the cost per chip/test. Magnetic particle spectroscopy (MPS), a novel measurement tool that closely related to magnetic particle imaging (MPI), has emerged recent years as an alternative to the aforementioned techniques and is gaining increasing attentions in the area of volumetric-based bioassays.[22 -- 26] In MPS, also called 0D MPI, oscillating magnetic fields periodically drive magnetic nanoparticles (MNPs) into saturated states where the magnetic responses contain not only the fundamental driving field frequencies but also a series of harmonic frequencies. These harmonics are very useful metrics for characterizing the MNP ferrofluids such as the viscosity and temperature of the solution as well as the conjugations of chemical compounds on the MNPs.[26 -- 31] Nowadays, with the ease of fabricating and surface-functionalizing MNPs, MNPs functionalized with ligands, DNA, RNA, and protein molecules can be readily applied as probes, labels, contrasts, and tracers for different bioassays.[26,31 -- 39] Unlike the surface-based biosensing platforms (ELISA, GMR, Hall effect sensors, etc.) that directly detect individual target objects near the sensing elements, the volumetric-based biosensing platforms measure the analytical signals that directly come from the entire detection volume, making the bioassays simple and fast. [40 -- 43] Representative examples of volumetric-based sensors are the conventional superconducting quantum interference devices (SQUIDs), the nuclear magnetic resonance (NMR) devices, and the magnetic susceptometers.[44 -- 50] MPS is one type of volumetric-based biosensing platform where the MNPs are acting as the mini-sensing probes and their dynamic magnetic responses are used as metrics for the characterization of target analytes from the fluidic samples.[51 -- 55] In this current work, we report the MPS-based bioassay platform as a rapid, sensitive, and wash-free method for the detection of H1N1 nucleoprotein. Due to the IgG antibodies anchored on the MNPs, cross-linking takes place between MNPs and H1N1 nucleoprotein which forms MNP clusters. Characterization of these MNP clusters reveal their controlled formation and hence in the upcoming sections we have referred to them as MNP 'self- assemblies'. The 3rd and the 5th harmonics along with the 3rd over the 5th harmonic ratios (R35) can be used as metrics for the characterization of target analyte concentrations from fluidic samples. These harmonics can be detected from nanogram quantities of iron oxide MNPs within 10 s. We show that H1N1 nucleoprotein can be detected with high sensitivity (44 nM or 442 pmole) using this detection scheme. By combining MPS technology and the self-assembly of MNPs, we are able to achieve rapid, sensitive, and wash-free magnetic bioassays. Furthermore, this detection scheme utilizing the self-assembly of MNPs is suitable for detecting and quantifying a wide range of biomarkers/analytes. 3 2. Results and Discussion 2.1 Conjugating IgG Polyclonal Antibodies on MNPs. To immobilize rabbit IgG polyclonal antibodies onto MNPs, the zero-length carbodiimide crosslinker EDC is used to couple the carboxyl groups from MNPs to the primary amines from IgG. As shown in Figure 1(a):i -- iii, the EDC reacts with carboxylic acid group from MNP to form an O-acylisourea active ester that can be easily displaced by nucleophilic attack from primary amino group on IgG. In this step, 1.5 mL microcentrifuge tube containing 200 L, 0.29 nmole/mL SHP-25 MNP suspension is mixed with 200 L, 25 mM MES and 10 L, 10 mg/mL EDC, react at room temperature for 15 min with continuous mixing. Then, 100 L, 5 g/mL polyclonal IgG is added to the suspension and react at room temperature for 2.5 hours with continuous mixing. As is shown in Figure 1(a):iii -- v, the primary amine from IgG forms an amide bond with the original carboxyl group and an EDC by-product is released as a soluble urea derivative (not shown in the figure). Hereafter, 100 L, 100 mM Trizma hydrochloride buffer is added to the suspension and react at room temperature for 30 minutes with continuous mixing to quench the EDC activation reaction. The MNP and IgG antibody complexes (denoted as 'MNP+Aby' in this paper) are formed and the unconjugated IgG polyclonal antibodies are removed by ultra- centrifuging the MNP ferrofluid at 11 000 RPM, acceleration 11 200 g, for 45 min (PowerSpin™ BX Centrifuge), then the supernatant is removed and the remaining MNP+Aby complexes are resuspended in PBS to a total volume of 200 L (in order to maintain the original concentration of 0.29 nmole/mL nanoparticles). This wash step is repeated for three times to thoroughly remove the unconjugated IgG polyclonal antibodies. Using the same method, eight samples each containing 100 L, 0.29 nmole/mL MNP+Aby complexes in PBS buffer are prepared (labeled as samples I -- VIII). The as-prepared MNPs anchored with rabbit IgG polyclonal antibodies are ready to be used for bioassays and can be stored at 2C -- 8C under sterile condition for one month without detectable loss of magnetic properties of MNPs and activity of IgG. This test kit containing microliter volume surface functionalized MNPs allows for future on field bioassays along with a portable MPS platform. 4 5 Figure 1. (a) EDC (carbodiimide) crosslinking reaction scheme. i) SHP-25 MNP with carboxylic acid group; ii) EDC crosslinker; iii) O-acylisourea active ester (unstable) on MNP; iv) The polyclonal IgG with primary amine group; v) The reaction of O-acylistourea with the amine from IgG forms stable amide bond on MNP. (b) Schematic view of the self-assembly of MNPs through proper choice of IgG polyclonal antibodies and target antigen hosting multiple different epitopes. vi) SHP-25 MNP surface conjugated with IgG polyclonal antibodies; vii) H1N1 nucleoprotein (AAM75159.1) (Met1-Gly490) possessing multiple different epitopes that allows a number of IgG polyclonal antibodies to bind; viii) The final product of the self-assembled MNPs. (c) Sample preparation flow charts of the experimental and negative control groups. Nine MNP samples are prepared: sample index I -- VII are MNP-antibody complexes in the presence of different concentrations of H1N1 nucleoprotein; sample index VIII is MNP-antibody complexes in the absence of H1N1 nucleoprotein (denoted as '0 nM (MNP+Aby)' in this paper); sample index IX is bare MNP suspension (denoted as 'Bare MNP' in this paper). The details of samples I -- IX are listed in Table 1. (d) Schematic view of MPS system setups. (e) and (f) are the 3rd and the 5th harmonics along varying driving field frequencies (only samples I, VIII, and IX are plotted) collected by the MPS system. (g) Boxplots of the harmonic ratios (R35) collected from samples I, VIII, and IX. Note that figures (a) -- (c) are not drawn to scale. 2.2 The Experimental and Control Groups. As shown in Figure 1(c), to the as-prepared MNP-IgG samples I -- VII were added 100 L H1N1 nucleoprotein of varying concentrations (from 4.42 M down to 44 nM as listed in Table 1). To sample VIII, 100 L PBS buffer was added (instead of antigen) as negative control group #1 (denoted as '0 nM (MNP+Aby)' in this paper). Sample IX containing 100 L, 0.29 nmole/mL SHP-25 MNP and 100 L PBS buffer (without antibody or H1N1 nucleoprotein) is prepared as negative control group #2 (denoted as 'Bare MNP' in this paper). Samples I -- IX are incubated at room temperature for 2 hours with continuous mixing to allow the cross-linking of MNPs and H1N1 nucleoprotein molecules, followed by storage at 2C -- 8C under sterile condition before the MPS measurements. Table 1. Experimental design of samples I - IX Sample Index I II III IV V VI VII VIII IX MNP 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL Polyclonal IgG Yes Yes Yes Yes Yes Yes Yes Yes No Concentration 4.42 µM 2.21 µM 884 nM 442 nM 221 nM 88 nM 44 nM 0 nM 0 nM 6 H1N1 Volume* 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL 100 µL Nucleoprotein Note: to samples I -- VII we added 100 µL of varying concentrations of H1N1 nucleoprotein and to samples VIII & IX are added with 100 µL PBS buffer. 2.3 Self-Assembly of MNPs. Different degrees of MNP self-assemblies (here different degrees indicate the number and average sizes of MNP self-assemblies) are formed in samples I -- VII due to the fact that the H1N1 nucleoprotein molecule has multiple epitopes recognized by polyclonal antibodies (see Figure 1(b)). As a result, each H1N1 nucleoprotein is bound to more than one MNPs. On the other hand, each MNP functionalized with multiple IgG polyclonal antibodies is able to bind with multiple H1N1 nucleoprotein molecules. Thus, the cross-linking of MNPs and H1N1 nucleoproteins leads to different degrees of MNP self-assemblies depending on the number/concentration of H1N1 nucleoproteins in the MNP ferrofluid. 2.4 Detection of H1N1 Nucleoprotein via MPS and Magnetic Relaxation Dynamics. When the MNPs are suspended in a solution under oscillating magnetic fields, there are two mechanisms governing the rotation of magnetic moments in response to the magnetic fields (see Scheme 1(a) & (b)). One is the intrinsic Neel motion (rotating magnetic moment inside the stationary MNP) and the other is the extrinsic Brownian motion (rotating the entire MNP along with its magnetic moment). In principle, the joint effects of Néel and Brownian relaxation mechanisms contribute to the macroscopic magnetic response of MNP ferrofluid sample subjected to external oscillating magnetic fields. Herein, we use the SHP-25 MNPs with averaged core diameter of 25 nm, where Brownian motion is the dominating relaxation mechanism and thus determines the macroscopic magnetic responses of MNPs (in this case the effect of Neel motion is negligible, see Scheme 1(c)). For these Brownian motion-dominated MNPs, their macroscopic magnetic responses are directly related to the overall hydrodynamic sizes. MPS-based bioassay uses the harmonics of oscillating MNPs as a metric for the freedom of rotational motion, thus indicating the degree of MNP self-assembly. In the presence of H1N1 nucleoprotein molecules, the polyclonal IgG functionalized MNPs can bind to one nucleoprotein molecule or cross-link to multiple nucleoprotein molecules, thus, forming MNP self-assemblies (see Scheme 1(d)). Such an interaction leads to an increased hydrodynamic size of the MNPs, resulting in an increased Brownian relaxation time, which is monitored by the MPS. The 3rd and the 5th harmonics are immediately recorded at 𝑓𝐻 ± 2𝑓𝐿 and 𝑓𝐻 ± 2𝑓𝐿. In addition, the ratio of the 3rd over the 5th harmonic (R35) is used as a MNP quantity-independent metric for the detection of H1N1 nucleoprotein. Herein, we report using the 3rd and the 5th harmonics along with the 3rd over the 5th harmonic 7 ratio (R35) as metrics for the characterization of target analyte concentrations from fluidic samples (see example plots in Figure 1(e) -- (g)). Scheme 1. (a) Neel motion is the rotation of magnetic moment inside the stationary MNP. (b) Brownian motion is the rotation of the entire MNP along with its magnetic moment. (c) The Néel, Brownian, and effective relaxation times are depicted as a function of iron oxide MNP core size, assuming MNPs are dispersed in a solution with viscosity η=1 cP and temperature T=300 K. The vertical orange line highlights that the Brownian motion is the dominating relaxation mechanism for iron oxide MNPs with core size of 25 nm. The magnetic relaxation time models can be found from Supporting Information Note S2. (d) As the degree of MNP self-assembly increases in the presence of H1N1 nucleoprotein, the average hydrodynamic sizes of particles (MNPs and MNP self-assemblies) increases, as a result, the measured harmonic amplitude decreases. The dashed black lines represent the hydrodynamic sizes of MNPs and MNP self- assemblies. Note that figures (a), (b) and (d) are schematics and are not drawn to scale. 8 2.5 The 3rd and the 5th Harmonics as Metrics for the Detection of H1N1 Nucleoprotein. The magnetic responses of MNPs from the as-prepared samples I -- IX are monitored and recorded by our lab- based MPS system. One negative control group 1 (sample VIII, see Figure 1(c)) is added to this experiment to validate the selectivity in detection of target biomarker, i.e., H1N1 nucleoprotein. The other negative control group 2 (sample IX, see Figure 1(c)) is added to verify the efficiency of anchoring rabbit IgG polyclonal antibodies on MNPs. The description of our lab-based MPS system setup can be found in Supporting Information Note S1 and a representative signal chain is shown in Figure 1(d). The frequency 𝑓𝐻 of the high-frequency driving field is swept from 400 Hz to 20 kHz, and the frequency of the low-frequency driving field is set at 10 Hz. The amplitudes of the high- and low-frequency driving fields are set at 17 Oe and 170 Oe, respectively. The sampling rate is set at 500 kHz. In each MPS measurement, the background noise floor is collected for 5 s followed by inserting the sample for another 5 s of data collection. The 3rd and the 5th harmonics from the magnetic responses of MNPs are reconstructed from the signal and background noise. As shown in Figure 2, according to the Faraday's Law of Induction and the Langevin model (see Notes S3 - S6 from Supporting Information), the amplitudes of the 3rd and the 5th harmonics are modulated by the driving field frequency 𝑓𝐻. With increasing concentration/quantity of H1N1 nucleoprotein in the samples, MNPs are likely form larger self-assemblies, as a result, the Brownian relaxation time (as well as the effective relaxation time) increases. This increased relaxation time causes an increased phase lag (see Notes S4 from Supporting Information) between the magnetic moments of MNPs and the external oscillating magnetic fields. 9 Figure 2. (a, b) MPS measurements of the 3rd and the 5th harmonics from samples I -- IX at varying driving field frequencies from 400 Hz to 20 kHz. The insets in i) -- iv) highlight the 3rd harmonic amplitudes measured at 1 kHz, 5 kHz, 10 kHz and 20 kHz, respectively. Insets v) -- viii) highlight the 5th harmonic amplitudes measured at 1 kHz, 5 kHz, 10 kHz and 20 kHz, respectively. The error bar represents standard deviation. The amplitudes of the 3rd and the 5th harmonics from samples I -- IX are summarized in Figure 2(a) & (b), which shows a clear trend that at all driving field frequencies, the samples with higher concentrations of H1N1 nucleoprotein, yield weaker magnetic responses (i.e., smaller harmonic amplitudes). This is due to the following facts: 1) larger MNP self-assemblies are formed in the presence of higher concentration/quantity of H1N1 nucleoprotein; 2) the Brownian motions of MNPs that are cross-linked in the self-assemblies are blocked (i.e., 10 larger hydrodynamic size as shown in Scheme 1(d)); 3) the magnetic moments of these 'trapped' MNPs fail to align with the external magnetic fields, thus, weaker dynamic magnetic responses are observed from these samples. The measured harmonic amplitudes arranged from lowest to highest are: I (4.42 µM) < II (2.21 µM) < III (884 nM) < IV (442 nM) < V (221 nM) < VI (88 nM) < VII (44 nM) < VIII (0 nM, MNP+Aby) < IX (0 nM, bare MNP). Figure 2(a):i -- iv and Figure 2(a):v -- viii highlight the amplitudes of the 3rd and the 5th harmonics at driving field frequencies of 1 kHz, 5 kHz, 10 kHz, and 20 kHz, respectively. These values are summarized in Tables S1 & S2 from Supporting Information. The differences in the harmonics from samples VIII and IX indicate that the rabbit IgG polyclonal antibodies have been successfully anchored onto MNPs. The conjugation of antibodies onto MNPs could slightly increase the hydrodynamic size of MNPs (see Scheme 1(d)), which is also evident in the DLS results in Figure 5(a) -- (e) where Figure 5(d) & (e) are the averaged hydrodynamic sizes are 48.4 nm and 46.3 nm for negative control groups, VIII (MNP+Aby) and IX (bare MNP), respectively. 2.6 Harmonic Ratio (R35) as MNP Quantity-Independent Metric for the Detection of H1N1 Nucleoprotein. As the harmonics are largely dependent on the number of MNPs from the sample, the MPS results could be biased by the deviations of MNP quantities from one sample to another, especially for the detection of very low concentration target biomarkers. Besides the metrics of the 3rd and the 5th harmonics for the characterization of H1N1 nucleoproteins. The harmonic ratio, R35, has also been proposed as MNP quantity-independent metric for the detection of biomarkers (see Notes S7 from Supporting Information).[26,31,56] Herein, the harmonic ratios R35 from samples I -- IX at various driving field frequencies are summarized in Figure 3(a). Again, a clear trend is noticed that at all driving field frequencies, the samples with higher concentrations of H1N1 nucleoprotein yield smaller R35 values. Figure 3(a):i -- iv highlights the R35 values from each sample at driving field frequencies of 1 kHz, 5 kHz, 10 kHz, and 20 kHz, respectively. 11 Figure 3. MPS measurements of the ratio of the 3rd over the 5th harmonics (R35) from samples I -- IX. (a) Harmonic ratios, R35, from samples I -- IX as we vary the driving field frequencies from 400 Hz to 20 kHz. Insets i) -- iv) highlight the R35 measured at 1 kHz, 5 kHz, 10 kHz and 20 kHz, respectively. (b) Boxplot of R35 from samples I -- IX. The first quartile (Q1, 25th percentile), median (Q2, 50th percentile), and third quartile (Q3, 75th percentile) values from samples I -- IX are summarized in Table 2. R35 heatmap of samples I -- IX is reported in Notes S8 from Supporting Information. 12 Figure 3(b) is the boxplot of R35 values from MNP samples at all driving field frequencies. The statistical first quartile (Q1), median (Q2), third quartile (Q3) R35 values from samples I -- IX are summarized in Table 2. Again, there is a clear trend that the R35 value decreases as the concentration of H1N1 nucleoprotein increases. Table 2. The 25th, 50th, and 75th percentiles R35 values of 9 samples in the boxplot from Figure 3(b) Sample Index I II III IV V VI VII VIII IX 25th Percentile (Q1) 1.541 1.549 1.559 1.568 1.573 1.580 1.585 1.595 1.610 50th Percentile (Q2) 1.531 1.542 1.553 1.560 1.566 1.571 1.578 1.588 1.603 75th Percentile (Q3) 1.495 1.508 1.523 1.533 1.534 1.540 1.550 1.559 1.576 2.7 Sensitivity of H1N1 Nucleoprotein Detection. The MPS responses of MNPs versus H1N1 nucleoprotein concentrations ranging from 44 nM to 4.42 µM (4.4 -- 442 pmole) are investigated in Figure 4. Overall, the harmonic amplitudes and the R35 values increase as the concentration of H1N1 nucleoprotein decreases. It should be noted that both the 3rd and the 5th harmonics show significantly weaker capabilities in detecting samples with low concentrations of H1N1 nucleoproteins (see Notes S9 from Supporting Information). Furthermore, neither the 3rd harmonic nor the 5th harmonic succeed in distinguishing samples I (4.42 µM) and II (2.21 µM) at driving field frequency of 20 kHz, where the concentrations of H1N1 nucleoprotein from these samples are extremely high. On the other hand, the harmonic ratio, R35, demonstrates comparable capabilities in distinguishing high and low concentrations of H1N1 nucleoprotein samples (see Figure S3(c) from Supporting Information). However, it fails to detect the differences in samples I (4.42 µM) and II (2.21 µM) at 20 kHz driving field frequency. All of the metrics reported in this paper, the 3rd harmonic, the 5th harmonic, and the harmonic ratio R35, achieved a detection limit of 44 nM (442 pmole) for H1N1 nucleoprotein. 13 Figure 4. (a) -- (f) are the amplitudes of the 3rd harmonics from samples I -- VIII at driving field frequencies of (a) 1 kHz; (b) 3 kHz; (c) 5 kHz; (d) 10 kHz; (e) 15 kHz; and (f) 20 kHz, respectively. (g) -- (l) are the amplitudes of the 5th harmonics from samples I -- VIII at driving field frequencies of (g) 1 kHz; (h) 3 kHz; (i) 5 kHz; (j) 10 kHz; (k) 15 kHz; and (l) 20 kHz, respectively. (m) -- (r) are the harmonic ratios, R35, from samples I -- VIII at driving field frequencies of (m) 1 kHz; (n) 3 kHz; (o) 5 kHz; (p) 10 kHz; (q) 15 kHz; and (r) 20 kHz, respectively. The error bar represents standard deviation. 2.8 Hydrodynamic Size Analysis on the MNP Self-Assemblies. The hydrodynamic sizes of the samples are characterized by Dynamic Light Scatter (DLS, Microtrac NanoFlex). Figure 5(a) -- (f) shows the hydrodynamic size distributions of samples II (2.21 µM), IV (442 nM), VI (88 nM), VIII (0 nM, MNP+Aby) and IX (0 nM, bare MNP) with mean values of 58.8 nm, 51.7 nm, 48.7 nm, 48.4 nm and 46.3 nm, respectively. The decreasing mean values from samples II through IX signifies a gradual decrease in the self-assembly degrees of MNPs. Although the primary peaks in the size distributions for all the samples are nearly similar, the main differences come from the peaks at the tails of the size distribution curves (sizes between 200 nm to 300 nm) which is highlighted in the blue area from Figure 5(a) -- (e) as well as the black square in Figure 14 5(f). The tail shows the largest peak in sample II which gradually diminishes in the consecutive samples with little to no peak in the sample IX. The black dashed lines in Figure 5(a) -- (e) are the log-normal curve fittings of the particle sizes from samples II, IV, VI, VIII, and IX. Figure 5(f) highlights the differences in the collected DLS size distribution curves from samples II and VIII. Sample II containing 100 µL, 2.21 µM H1N1 nucleoprotein shows the maximum self-assembly degree and hence the largest mean hydrodynamic size is observed. Decreasing concentration of H1N1 nucleoprotein in the remaining samples results in less degrees of MNP self-assembly, hence, smaller mean hydrodynamic size. DLS characteristics of these samples show good agreement with the MPS measurements discussed earlier. 15 16 Figure 5. Statistical distribution of the hydrodynamic sizes of samples (a) II, (b) IV, (c) VI, (d) VIII, and (e) IX as characterized by DLS. Dotted lines are the fitted log-normal distribution curves. The µ values represent the statistical mean of the hydrodynamic sizes of the samples. The peaks in the blue highlighted region at the tail show decreasing numbers of MNP self-assemblies from sample II through samples IV, VI, VIII and IX. (f) Comparison of the measured DLS size distribution curves between samples II (2.21 µM) and VIII (0 nM, MNP+Aby). The peak at the tail of DLS size distribution curve from sample II is highlighted in the black square. (g) -- (o) are the bright-field TEM images of samples II ((g) -- (i)), VI ((j) -- (l)), and IX ((m) -- (o)). (p) and (t) are the bright-field TEM images from sample VI, highlighting the different shapes of MNP assemblies in the sample. (q), (r), and (s) are the MNP self-assembly models: (q) the classic MNP self-assembly model (cluster); (r) the MNP open ring tetramer model, and (s) the MNP chain model representing the orange, blue, and green dotted contours in (p), respectively. (u), (v), (w), (x), and (y1) -- (y3) are the MNP pentamer, tetramer, trimer, dimer, and monomer models representing the purple, lime, yellow, red, and black solid contours in (t), respectively. Note that figures (q) -- (s) and (u) -- (y) are not drawn to scale. 2.9 Morphological Characterization on the MNP Self-Assemblies. The morphologies of the MNPs and MNP self-assemblies from Samples I -- IX are characterized by the transmission electron microscopy (TEM, FEI T12 120 kV). Each TEM sample is prepared by dipping a drop of sample fluid onto a TEM grid (TED PELLA, inc.). The droplet is air-dried at room temperature before taking the TEM characterization. As shown in Figure 5(g) -- (o), the bright-field TEM images are taken from samples II (2.21 µM), VI (88 nM), and IX (0 nM, bare MNP). MNPs show average magnetic core size of around 25 nm and narrow size distribution, which is critical to obtain consistent and comparable experimental results. Larger MNP self-assemblies are observed from samples with higher H1N1 nucleoprotein concentration like sample II, which is consistent with the DLS results that higher nucleoprotein concentration yields larger average hydrodynamic size. It has been mentioned earlier that, the H1N1 nucleoprotein molecule has many epitopes that allow the binding of multiple IgG polyclonal antibodies. In addition, the MNPs are surface-functionalized with multiple IgG polyclonal antibodies that would in turn favors the binding to multiple H1N1 nucleoprotein molecules. This cross- linking between MNPs and H1N1 nucleoproteins causes the formation of different degrees of MNP self- assemblies, which has been captured in the TEM images from sample VI as shown in Figure 5(p) & (t) and is schematically drawn in Figure 5(q) -- (y3). Figure 5(q), (r) & (s) shows the classic MNP self-assembly (cluster), an open ring tetramer and a chain tetramer highlighted in the dotted-orange, dotted-blue and dotted-green contours 17 in Figure 5(p), respectively. Figure 5(u), (v), (w) & (x) shows the formation of a pentamer, a tetramer, a trimer and a dimer highlighted in solid-purple, solid-lime, solid-yellow and solid-red contours in Figure 5(t), respectively. The solid-black contour in Figure 5(t) is a monomer and all possible schematic representations for the formation of monomers are drawn in Figure 5(y1) -- (y3). The possible reasons for monomer MNPs, as shown in TEM images, in the presence of H1N1 nucleoproteins are: 1) the MNP is conjugated with H1N1 nucleoproteins but is not cross-linked to a second MNP (see Figure 5(y1)); 2) the MNP is not bound to any H1N1 nucleoprotein molecules (see Figure 5(y2)); or 3) the MNP does not have enough IgG polyclonal antibodies anchored on its surface decreasing its ability to bind with H1N1 nucleoprotein molecules and assemble into a cluster (see Figure 5(y3)). 3. Conclusions In this paper, we have successfully demonstrated the feasibility of using a MPS system combined with the self- assembly of MNPs for rapid, sensitive, and wash-free detection of H1N1 nucleoprotein. The H1N1 nucleoprotein molecule has multiple epitopes that serve as binding sites for IgG polyclonal antibodies. Thus, each H1N1 nucleoprotein may be bound to more than one MNP, consequently assembling into clusters. In addition, each MNP functionalized with IgG antibodies specific to multiple epitopes of the H1N1 nucleoprotein enabling the particle to bind to multiple H1N1 nucleoprotein molecules. As a result, the cross-linking of MNPs and H1N1 nucleoproteins leads to different degrees of MNP self-assemblies depending on the number/concentration of H1N1 nucleoprotein molecules in the MNP ferrofluid. Noticeable changes in the macroscopic magnetic responses of MNPs are detected by the MPS system when these ferrofluid samples are subjected to external oscillating magnetic fields. Herein, we have reported the 3rd harmonic, the 5th harmonic, and the harmonic ratio R35 as metrics for the detection of H1N1 nucleoprotein. The detection scheme of self-assembly of MNPs along with the harmonic metrics used in this present study was able to detect concentrations as low as 44 nM (442 pmole) of H1N1 nucleoprotein. There are several advantages for using the MPS method: 1) Rapid: the testing time for detection is within 10 s; 2) Sensitive: detection limits are at the pmole level; 3) Wash-free, the as-prepared surface functionalized MNPs can be mixed directly with the fluidic sample without further wash steps and the analytical signals are detected directly from the entire sample volume, making the bioassays simple and fast; 4) Easy-to-use: measurement could be carried out on minimally processed biological samples by non-technicians with minimum technical training requirements; 5) Cost effective: each bioassay only uses nanogram amounts of iron oxide nanoparticles; 6) Portable: the entire MPS system could be assembled onto a single PCB board, allowing to configure a hand-held device for field testing. In short, the approach we report here is a rapid and versatile technique that is suitable for the quantitative detection of a wide range of biomarkers/analytes. MPS is a promising, new, and cheap bioassay platform that has the potential to be applied in many biomedical applications including both in vivo and in vitro diagnostics. 18 Furthermore, a test kit containing microliter volumes of surface functionalized MNPs along with a point-of-care MPS device opens the future to field-based bioassays in areas of nanomedicine, food safety, agriculture, and veterinary medicine. 4. Experimental Section Materials: The recombinant Influenza A virus (A/Puerto Rico/8/34/Mount Sinai (H1N1)) nucleoprotein (AAM75159.1) (Met1-Gly490) is purchased from Sino Biological Inc. (Catalog#11675-V08B). It comprises of 501 amino acids and has a predicted molecular mass of 56.6 kDa. The H1N1 nucleoprotein is provided as lyophilized powder from sterile 60 mM Tris, 500 mM NaCl, pH 7.4 and stored at -20C under sterile condition before usage. The lyophilized H1N1 nucleoprotein powder is reconstituted in DI water to prepare a stock solution of 0.25 mg/mL (4.42 M). The Influenza A nucleoprotein specific polyclonal IgG is produced in rabbits immunized with purified, recombinant Influenza A nucleoprotein (Catalog#11675-V08B; AAM75159.1; Met1- Gly490). This rabbit IgG polyclonal antibody purchased from Sino Biological Inc. (Catalog#11675-T62) is provided as liquid solution with measured concentration of 2 mg/mL, stored at -20C before usage. The rabbit polyclonal IgG is diluted in PBS to a desired concentration of 5 g/mL. The MNPs with average particle magnetic core size of 25 nm, concentration of 0.29 nmole/mL nanoparticles (5 mg/mL Fe), suspended in DI water containing 0.02% sodium azide, is purchased from Ocean NanoTech Inc. (Catalog#SHP-25). They are water- soluble iron oxide nanoparticles coated with a monolayer of oleic acid and a monolayer of amphiphilic polymer, their reactive group is carboxylic acid and their zeta potential is between -35 mV and -15 mV. The EDC (1-ethyl- 3-(3-dimethylaminopropyl)carbodiimide hydrochloride, C8H17N3·HCl) is purchased from Thermo Fisher Scientific (Catalog#22980). It is a water-soluble carbodiimide crosslinker that activates the carboxyl groups on MNPs for spontaneous reaction with primary amines from rabbit IgG polyclonal antibodies, enabling stable antibody conjugation. The EDC powder is dissolved in DI water to a desired concentration of 10 mg/mL before immediate use. The MES (2-(N-Morpholino)ethanesulfonic acid, C6H13NO4S, Prod. No. M3671), Trizma hydrochloride (Tris(hydroxymethyl)aminomethane hydrochloride, NH2C(CH2OH)3·HCl, Prod. No. T5941), and Phosphate Buffered Saline (PBS, Prod. No. 79378) are purchased from Sigma-Aldrich. The MES powder is dissolved in DI water to a desired concentration of 25 mM, pH 6.0. Trizma hydrochloride powder is dissolved in DI water to a desired concentration of 100 mM, pH 7.4. Magnetic Particle Spectroscopy (MPS) System Setups: The lab-based MPS system setups can be found from Supporting Information Note S1 and the signal chain is shown in Figure 1(d). A laptop installed with LabVIEW sends commands to the Data Acquisition Card (DAQ, NI USB-6289) to generate two sinusoidal waves, which are amplified by two Instrument Amplifiers (IA, HP 6824A). These amplified sinusoidal waves are sent to the primary and secondary coils to produce oscillating magnetic fields: one with frequency 𝑓𝐿 = 10 𝐻𝑧 and amplitude 𝐴𝐿 = 170 𝑂𝑒, the other with frequency 𝑓𝐻 varying from 400 Hz to 20 kHz and amplitude 𝐴𝐻 = 17 𝑂𝑒. One pair 19 of differently wound pick-up coils (600 windings clock-wise and 600 windings counter-clock-wise) sense the induced magnetic responses from MNPs and send back to DAQ. The response signals at combinatorial frequencies 𝑓𝐻 ± 2𝑓𝐿 (the 3rd harmonics) and 𝑓𝐻 ± 2𝑓𝐿 (the 5th harmonics) are extracted and analyzed. Supporting Information Magnetic Particle Spectroscopy System Setups; Magnetic Relaxation Time Models; Langevin Model of Magnetic Responses; The Phase Lag Model; The Induced Voltage in Pick-up Coils; The 3rd and the 5th Harmonics; The 3rd over the 5th Harmonic Ratio (R35) as MNP Quantity-Independent Metric; The R35 Heatmap; Capabilities of the 3rd Harmonic, the 5th Harmonic, and the Harmonic Ratio R35 as Metrics for Distinguishing H1N1 Nucleoprotein Samples with High and Low Concentrations; Table S1. The 3rd harmonic amplitudes (µV) from Figure 2(a):i -- iv; Table S2. The 5th harmonic amplitudes (µV) from Figure 2(b):v -- viii. This material is available in the supporting information. Acknowledgements This study was financially supported in part by the Institute of Engineering in Medicine of the University of Minnesota through FY18 IEM Seed Grant Funding Program and the College of Veterinary Medicine Emerging, Zoonotic and Infectious Diseases Signature Program, USDA-NIFA SAES General Agriculture research funds (MIN-62-097). Portions of this work were conducted in the Minnesota Nano Center, which is supported by the National Science Foundation through the National Nano Coordinated Infrastructure Network (NNCI) under Award Number ECCS-1542202. Portions of this work were carried out in the Characterization Facility, University of Minnesota, a member of the NSF-funded Materials Research Facilities Network (www.mrfn.org) via the MRSEC program. Conflict of Interest The authors declare no competing financial interest. Keywords magnetic particle spectroscopy, influenza a virus, H1N1, nucleoprotein, magnetic nanoparticle, self-assembly, wash-free, bioassay References [1] R. G. Webster, W. J. Bean, O. T. Gorman, T. M. Chambers, Y. Kawaoka, Microbiol Rev 1992, 56. [2] A. M. King, E. Lefkowitz, M. J. Adams, E. B. Carstens, Virus taxonomy: ninth report of the International Committee on Taxonomy of Viruses; Elsevier, 2011; Vol. 9. 20 [3] L. Ferguson, L. Eckard, W. B. Epperson, L.-P. Long, D. Smith, C. Huston, S. Genova, R. Webby, X.-F. Wan, Virology 2015, 486, 28. [4] J. McCauley, S. Hongo, N. Kaverin, G. Kochs, R. Lamb, M. Matrosovich, D. Perez, P. Palese, R. Presti, E. Rimstad, Virus Taxon. Ninth Rep. Int. Comm. Taxon. Viruses 2012, 749. [5] J. K. Taubenberger, D. M. Morens, Rev. Biomed. 2006, 17, 69. [6] B. Kumar, K. Asha, M. Khanna, L. Ronsard, C. A. Meseko, M. Sanicas, Arch. Virol. 2018, 163, 831. [7] C. Albo, A. Valencia, A. Portela, J. Virol. 1995, 69, 3799. [8] R. Arranz, R. Coloma, F. J. Chichón, J. J. Conesa, J. L. Carrascosa, J. M. Valpuesta, J. Ortín, J. Martín- Benito, Science 2012, 338, 1634. [9] A. Moeller, R. N. Kirchdoerfer, C. S. Potter, B. Carragher, I. A. Wilson, Science 2012, 338, 1631. [10] R. W. Ruigrok, F. Baudin, J. Gen. Virol. 1995, 76, 1009. [11] L. Shu, W. Bean, R. Webster, J. Virol. 1993, 67, 2723. [12] T. Ganzenmueller, J. Kluba, B. Hilfrich, W. Puppe, W. Verhagen, A. Heim, T. Schulz, C. Henke-Gendo, J. Med. Microbiol. 2010, 59, 713. [13] Z.-N. Li, K. M. Weber, R. A. Limmer, B. J. Horne, J. Stevens, J. Schwerzmann, J. Wrammert, M. McCausland, A. J. Phipps, K. Hancock, J. Virol. Methods 2017, 243, 61. [14] S. Faccini, A. De Mattia, C. Chiapponi, I. Barbieri, M. B. Boniotti, C. Rosignoli, G. Franzini, A. Moreno, E. Foni, A. D. Nigrelli, J. Virol. Methods 2017, 243, 31. [15] C. Chartrand, M. M. G. Leeflang, J. Minion, T. Brewer, M. Pai, Ann. Intern. Med. 2012, 156, 500. [16] J. Merckx, R. Wali, I. Schiller, C. Caya, G. C. Gore, C. Chartrand, N. Dendukuri, J. Papenburg, Ann. Intern. Med. 2017, 167, 394. [17] M. González-Del Vecchio, P. Catalán, V. de Egea, A. Rodríguez-Borlado, C. Martos, B. Padilla, B. Rodríguez-Sanchez, E. Bouza, Eur. J. Clin. Microbiol. Infect. Dis. 2015, 34, 1081. [18] A. T. Cruz, G. J. Demmler-Harrison, A. C. Caviness, G. J. Buffone, P. A. Revell, Pediatrics 2010, 125, e645. [19] K. Wu, T. Klein, V. D. Krishna, D. Su, A. M. Perez, J.-P. Wang, ACS Sens. 2017, 2, 1594. [20] V. D. Krishna, K. Wu, A. M. Perez, J. P. Wang, Front. Microbiol. 2016, 7, 8. [21] D. Su, K. Wu, V. Krishna, T. Klein, J. Liu, Y. Feng, A. M. Perez, M. C. Cheeran, J.-P. Wang, Front. Microbiol. 2019, 10, 1077. [22] K. Wu, D. Su, R. Saha, D. Wong, J.-P. Wang, J. Phys. Appl. Phys. 2019, 52, 173001. [23] H.-J. Krause, N. Wolters, Y. Zhang, A. Offenhäusser, P. Miethe, M. H. Meyer, M. Hartmann, M. Keusgen, J. Magn. Magn. Mater. 2007, 311, 436. [24] P. I. Nikitin, P. M. Vetoshko, T. I. Ksenevich, J. Magn. Magn. Mater. 2007, 311, 445. [25] K. Wu, K. Schliep, X. Zhang, J. Liu, B. Ma, J. Wang, Small 2017, 13, 1604135. 21 [26] X. Zhang, D. B. Reeves, I. M. Perreard, W. C. Kett, K. E. Griswold, B. Gimi, J. B. Weaver, Biosens. Bioelectron. 2013, 50, 441. [27] K. Wu, J. Liu, Y. Wang, C. Ye, Y. Feng, J.-P. Wang, Appl. Phys. Lett. 2015, 107, 053701. [28] A. M. Rauwerdink, E. W. Hansen, J. B. Weaver, Phys. Med. Biol. 2009, 54, L51. [29] I. Perreard, D. Reeves, X. Zhang, E. Kuehlert, E. Forauer, J. Weaver, Phys. Med. Biol. 2014, 59, 1109. [30] A. V. Orlov, J. A. Khodakova, M. P. Nikitin, A. O. Shepelyakovskaya, F. A. Brovko, A. G. Laman, E. V. Grishin, P. I. Nikitin, Anal. Chem. 2013, 85, 1154. [31] K. Wu, J. Liu, D. Su, R. Saha, J.-P. Wang, ACS Appl. Mater. Interfaces 2019, 11, 22979. [32] J. M. Perez, T. O'Loughin, F. J. Simeone, R. Weissleder, L. Josephson, J. Am. Chem. Soc. 2002, 124, 2856. [33] P. Miao, Y. Tang, L. Wang, ACS Appl. Mater. Interfaces 2017, 9, 3940. [34] A. Mezger, J. Fock, P. Antunes, F. W. Østerberg, A. Boisen, M. Nilsson, M. F. Hansen, A. Ahlford, M. Donolato, ACS Nano 2015, 9, 7374. [35] A. B. Chinen, C. M. Guan, J. R. Ferrer, S. N. Barnaby, T. J. Merkel, C. A. Mirkin, Chem. Rev. 2015, 115, 10530. [36] Y. Li, R. Lin, L. Wang, J. Huang, H. Wu, G. Cheng, Z. Zhou, T. MacDonald, L. Yang, H. Mao, J. Mater. Chem. B 2015, 3, 3591. [37] T. Klein, W. Wang, L. Yu, K. Wu, K. L. Boylan, R. I. Vogel, A. P. Skubitz, J.-P. Wang, Biosens. Bioelectron. 2019, 126, 301. [38] J. Liu, K. Wu, J.-P. Wang, AIP Adv. 2016, 6, 056126. [39] J. Liu, K. Schliep, S.-H. He, B. Ma, Y. Jing, D. J. Flannigan, J.-P. Wang, Phys. Rev. Mater. 5, 2, 054415. [40] N. T. Thanh, Clinical Applications of Magnetic Nanoparticles: From Fabrication to Clinical Applications; CRC Press, 2018. [41] A. V. Orlov, V. A. Bragina, M. P. Nikitin, P. I. Nikitin, Biosens. Bioelectron. 2016, 79, 423. [42] Y. Wang, G. Zhu, W. Qi, Y. Li, Y. Song, Biosens. Bioelectron. 2016, 85, 777. [43] B. Tian, X. Liao, P. Svedlindh, M. Strömberg, E. Wetterskog, ACS Sens. 2018, 3, 1093. [44] E. C. Alocilja, Y. Luo, J. Biol. Eng. 2017, 11, 14. [45] C. M. Castro, A. A. Ghazani, J. Chung, H. Shao, D. Issadore, T.-J. Yoon, R. Weissleder, H. Lee, Lab. Chip 2014, 14, 14. [46] Y. Zhao, Y. Li, K. Jiang, J. Wang, W. L. White, S. Yang, J. Lu, Food Control 2017, 71, 110. [47] L. M. Jackman, S. Sternhell, Application of Nuclear Magnetic Resonance Spectroscopy in Organic Chemistry: International Series in Organic Chemistry; Elsevier, 2013. [48] K. Park, T. Harrah, E. B. Goldberg, R. P. Guertin, S. Sonkusale, Nanotechnology 2011, 22, 085501. [49] J.-J. Chieh, K.-W. Huang, Y.-Y. Lee, W.-C. Wei, J. Nanobiotechnology 2015, 13, 1. [50] S. B. Trisnanto, Y. Kitamoto, Jpn. J. Appl. Phys. 2016, 55, 02BD02. 22 [51] S. Achtsnicht, A. M. Pourshahidi, A. Offenhäusser, H.-J. Krause, Sensors 2019, 19, 2599. [52] W. Wang, P. Ma, H. Dong, H.-J. Krause, Y. Zhang, D. Willbold, A. Offenhaeusser, Z. Gu, Biosens. Bioelectron. 2016, 80, 661. [53] M. Nikitin, A. Orlov, S. Znoyko, V. Bragina, B. Gorshkov, T. Ksenevich, V. Cherkasov, P. Nikitin, J. Magn. Magn. Mater. 2018, 459, 260. [54] M. P. Nikitin, A. Orlov, I. Sokolov, A. Minakov, P. Nikitin, J. Ding, S. Bader, E. Rozhkova, V. Novosad, Nanoscale 2018, 10, 11642. [55] H. Khurshid, Y. Shi, B. L. Berwin, J. B. Weaver, Med. Phys. 2018, 45, 3258. [56] A. M. Rauwerdink, J. B. Weaver, Med. Phys. 2010, 37, 2587. 23 Supporting Information Detection of Influenza A Virus Nucleoprotein Through the Self-Assembly of Nanoparticles in Magnetic Particle Spectroscopy-Based Bioassays: A Method for Rapid, Sensitive, and Wash-free Magnetic Immunoassays Kai Wu, Jinming Liu, Renata Saha, Diqing Su, Venkatramana D. Krishna, Maxim C-J Cheeran and Jian-Ping Wang* Dr. K. Wu, J. Liu, R. Saha, Prof. J.-P. Wang Department of Electrical and Computer Engineering University of Minnesota, Minneapolis, MN 55455, United States Email: [email protected] D. Su Department of Chemical Engineering and Material Science University of Minnesota, Minneapolis, MN 55455, United States Dr. V.D. Krishna, Prof. M.C-J. Cheeran Department of Veterinary Population Medicine University of Minnesota, St. Paul, Minnesota 55108, USA E-mail: [email protected] S-1 Note S1. Magnetic Particle Spectroscopy System Setups Figure S1 shows the photographs of lab-based MPS system used in this work. Figure S1. (a) Photograph of lab-based MPS system setups. i) Laptop with LabVIEW that sends commands and receives data; ii) The data acquisition card (DAQ, NI USB-6289); iii) Two instrument amplifiers (IA, HP 6824A). (b) Photograph of one set of coils and the plastic vials with maximum capacity of 300 µL. iv) one set of coils; v) cap of plastic vial; vi) plastic vial. (c) Photograph of primary, secondary, and pick-up coils. vii) primary coil with 1200 windings; viii) Secondary coil with 1200 windings; ix) One pair of pick-up coils with 600 windings clock- wise and 600 windings counter-clock-wise is placed in the center of primary and secondary coils. The plastic vial is inserted into the top half of the pick-up coils (as shown in Figure 1(d)) for the monitoring of MNP magnetic responses. The lower half of pick-up coils are designed to cancel out the voltage caused by the oscillating driving field from primary and secondary coils. As a result, the pick-up coils send back the induced voltage that is only due to the magnetic responses of MNPs. S-2 Note S2. Magnetic Relaxation Time Models When the MNPs are suspended in solution under oscillating magnetic fields, there are two mechanisms governing the rotation of magnetic moments in response to the magnetic fields (see Scheme 1(a) & (b)). One is the intrinsic Neel motion (rotating magnetic moment inside the stationary MNP) and the other is the extrinsic Brownian motion (rotating the entire MNP along with its magnetic moment). The relaxation dynamics of MNPs are dependent on the balance between the magnetic forces that tend to align the magnetic moments of MNPs along the external magnetic field direction and the thermal fluctuations that hinder the alignments. Relaxation time models are used to characterize these mechanisms. Zero-field Brownian relaxation time is expressed as:[1] 𝜏𝐵𝑧𝑒𝑟𝑜−𝑓𝑖𝑒𝑙𝑑 = 𝜏𝐵0 = 3𝜂𝑉ℎ 𝑘𝐵𝑇 (1), where 𝜂 is the viscosity of ferrofluid, 𝑘𝐵 is Boltzmann constant, 𝑇 is temperature, and 𝑉ℎ is the hydrodynamic volume of nanoparticles. In this paper, iron oxide nanoparticles with magnetic core diameters 𝐷 and surface chemical compounds thickness of 𝑐 are assumed. So, 𝑉ℎ = 𝜋(𝐷 + 2𝑐)3 6⁄ . Zero-field Néel relaxation time is expressed as:[2] where, 𝜏𝑁𝑧𝑒𝑟𝑜−𝑓𝑖𝑒𝑙𝑑 = √𝜋 2 𝜏𝑁0𝜎−3 2⁄ 𝑒𝜎 (2), 𝜏𝑁0 = 𝛽(1+𝛼′2)𝑀𝑠 2𝛾𝛼′ (3) 𝜎 = 𝛽 = −∆𝐸 𝑘𝐵𝑇 (4) 𝑉𝑐 𝑘𝐵𝑇 (5), where 𝛾 is electron gyromagnetic ratio (1.76 × 1011 𝑟𝑎𝑑 𝑠𝑇⁄ ), 𝛼′ is damping constant, which is typically on the order of 10−3~10−2 for bulk materials.[3] While fine particles have larger damping constants than bulk, usually 𝛼′ = 0.1 for iron oxide nanoparticles.[2] ∆𝐸 is the energy barrier a nanoparticle needs to overcome before flipping from one minimum energy state to the other minimum energy state. Magnetic core volume 𝑉𝑐 = 𝜋𝐷3 6⁄ . For MPS measurements, external magnetic fields are applied to the MNP ferrofluid systems. Hence, the zero- field relaxation time equations aforementioned do not apply in this case. These equations only give us the time for net magnetization to relax back to equilibrium after the removal of external magnetic fields.[2] As is reported before, the relaxation time is not only dependent on the externally applied magnetic field strength but also dependent on the field frequency.[4] Analytical expression for Brownian relaxation time at nonzero field is:[5] Analytical expression for Néel relaxation time at nonzero field is:[6] 𝜏𝐵 = 𝜏𝐵0 1+𝛼2−𝛼2𝑐𝑜𝑡ℎ2𝛼 𝛼𝑐𝑜𝑡ℎ𝛼−1 (6) S-3 𝜏𝑁 = 𝜏𝑁0 𝜎𝑒𝑓𝑓(1−ℎ2) √𝜎𝑒𝑓𝑓 𝜋⁄ 1+1 𝜎𝑒𝑓𝑓 ⁄ ( + 2−𝜎−1) × ( 1−ℎ 𝜎𝑒𝑓𝑓(1−ℎ)2 𝑒 −1 + 1+ℎ 𝜎𝑒𝑓𝑓(1+ℎ)2 𝑒 −1 −1 ) (7), −1 where 𝛼 and ℎ are time-varying parameters modulated by external magnetic field 𝐻(𝑡): 𝛼(𝑡) = 𝑀𝑠𝑉𝑐 𝑘𝐵𝑇 𝜇0𝐻(𝑡) (8) ℎ(𝑡) = 𝛼(𝑡) 2𝜎𝑒𝑓𝑓 (9) As we apply a periodically varying external magnetic field 𝐻(𝑡) to the ferrofluid system, the relaxation time 𝜏𝐵 and 𝜏𝑁 will also show a periodical change synchronously. The magnetization dynamics of MNPs are usually characterized by effective relaxation time 𝜏𝑒𝑓𝑓, which is dependent on Brownian relaxation time 𝜏𝐵 and Néel relaxation time 𝜏𝑁. Both relaxation processes are dependent on the frequency and amplitude of applied magnetic fields. The 𝜏𝑒𝑓𝑓 of a nanoparticle governs its ability to follow the external fields. The effective relaxation time 𝜏𝑒𝑓𝑓 is related to the Brownian and Néel relaxation times: 1 𝜏𝑒𝑓𝑓 = 1 𝜏𝐵 + 1 𝜏𝑁 (10) 𝜏𝑒𝑓𝑓 of a nanoparticle governs its ability to follow the alternating field. S-4 Note S3. Langevin Model of Magnetic Responses In the presence of oscillating magnetic fields, MNPs are magnetized and their magnetic moments tend to align with the fields. For a ferrofluid system of monodispersed, noninteracting MNPs, the magnetic response obeys Langevin function: where, 𝑀𝐷(𝑡) = 𝑚𝑠𝑐𝐿(𝜉) (11), 𝐿(𝜉) = coth 𝜉 − 1 𝜉 (12) 𝜉 = 𝑚𝑠𝐻(𝑡) 𝑘𝐵𝑇 (13) The MNPs are characterized by magnetic core diameter 𝐷, saturation magnetization 𝑀𝑠 and concentration 𝑐. Assuming MNPs are spherical without mutual interactions. The magnetic moment of each particle is 𝑚𝑠 = 𝑀𝑠𝑉𝑐 = 𝑀𝑠𝜋𝐷3 6⁄ , where 𝑉𝑐 is volume of the magnetic core, 𝜉 is the ratio of magnetic energy over thermal energy, 𝑘𝐵 is Boltzmann constant, and 𝑇 is the absolute temperature in Kelvin. 𝐻(𝑡) = 𝐴𝐻𝑐𝑜𝑠(2𝜋𝑓𝐻𝑡) + 𝐴𝐿𝑐𝑜𝑠(2𝜋𝑓𝐿𝑡) are the external magnetic fields, where 𝐴𝐻, 𝐴𝐿, 𝑓𝐻, 𝑓𝐿 are the amplitude and frequency of high and low frequency driving fields, respectively. It should be noted that the Langevin model is not able to describe the phase information of the harmonic signals. When the frequency of the driving magnetic field is low, the SPION's magnetic moment can follow the driving field tightly and the magnetic susceptibility 𝜒 is a real number. As the driving field frequency increases, a phase lag 𝜑 between the SPION's magnetic moment and the driving magnetic field is introduced, which leads to a complex magnetic susceptibility 𝜒 and the Debye model is used in this case. S-5 Note S4. The Phase Lag Model The phase lag is modulated by the low-frequency field and also can be monitored at the harmonic phase angles: where 𝜔 is the angular frequency, 𝜏𝑒𝑓𝑓 is modulated by the oscillating magnetic field. 𝜑(𝑡) = 𝑎𝑟𝑐𝑡𝑎𝑛(𝜔𝜏𝑒𝑓𝑓) (15) S-6 Note S5. The Induced Voltage in Pick-up Coils According to Faraday's law of Induction, the induced voltage in a pair of pick-up coils is expressed as: 𝑢(𝑡) = −𝑆0𝑉 𝑑 𝑑𝑡 𝑀𝐷(𝑡) (14) where 𝑉 is volume of MNP suspension. Pick-up coil sensitivity 𝑆0 equals to the external magnetic field strength divided by current. S-7 Note S6. The 3rd and the 5th Harmonics Taylor expansion of 𝑀𝐷(𝑡) from equation (11) shows the major frequency mixing components: 𝑀𝐷(𝑡) 𝑚𝑠𝑐 𝑚𝑠𝐻(𝑡) = 𝐿 ( 𝑘𝐵𝑇 ) = 1 3 ( 𝑚𝑠 𝑘𝐵𝑇 ) 𝐻(𝑡) − 1 45 ( 3 ) 𝐻(𝑡)3 + 2 945 5 ( 𝑚𝑠 𝑘𝐵𝑇 ) 𝐻(𝑡)5 + ⋯ 𝑚𝑠 𝑘𝐵𝑇 3 = ⋯ + [− 1 60 𝐴𝐻𝐴𝐿 2 ( 𝑚𝑠 𝑘𝐵𝑇 ) + ⋯ ] × 𝑐𝑜𝑠[2𝜋(𝑓𝐻 ± 2𝑓𝐿)𝑡 + 𝜑𝑓𝐻±2𝑓𝐿] 1 + [ 1512 𝐴𝐻𝐴𝐿 4 ( 5 𝑚𝑠 𝑘𝐵𝑇 ) + ⋯ ] × 𝑐𝑜𝑠[2𝜋(𝑓𝐻 ± 4𝑓𝐿)𝑡+𝜑𝑓𝐻±4𝑓𝐿] + ⋯ The mixing frequency components are found at odd harmonics exclusively: 𝑀𝐷(𝑡)3𝑟𝑑 ≈ −𝑚𝑠𝑐 60 𝐴𝐻𝐴𝐿 2 ( 3 𝑚𝑠 𝑘𝐵𝑇 ) 𝑀𝐷(𝑡)5𝑡ℎ ≈ 𝑚𝑠𝑐 1512 𝐴𝐻𝐴𝐿 4 ( 5 𝑚𝑠 𝑘𝐵𝑇 ) × 𝑐𝑜𝑠[2𝜋(𝑓𝐻 + 2𝑓𝐿)𝑡 + 𝜑𝑓𝐻±2𝑓𝐿] (17) × 𝑐𝑜𝑠[2𝜋(𝑓𝐻 + 4𝑓𝐿)𝑡+𝜑𝑓𝐻±4𝑓𝐿] (18) Amplitudes of induced voltages at the 3rd and 5th harmonics are expressed as: 𝑢3𝑟𝑑 = −𝑆0𝑉 𝑢5𝑡ℎ = −𝑆0𝑉 𝑑 𝑑𝑡 𝑑 𝑑𝑡 𝑀𝐷(𝑡)3𝑟𝑑 (19) 𝑀𝐷(𝑡)5𝑡ℎ (20) (16) S-8 Note S7. The 3rd over the 5th Harmonic Ratio (R35) as MNP Quantity-Independent Metric Harmonic ratio of the 3rd over the 5th harmonics is expressed as: 𝑅35 = 𝑢3𝑟𝑑 𝑢5𝑡ℎ = 𝑑 𝑑𝑡 𝑑 𝑑𝑡 𝑆0𝑉 𝑀𝐷(𝑡)3𝑟𝑑 𝑀𝐷(𝑡)3𝑟𝑑 𝑆0𝑉 𝑀𝐷(𝑡)5𝑡ℎ 𝑀𝐷(𝑡)5𝑡ℎ = 𝑑 𝑑𝑡 𝑑 𝑑𝑡 (21) Hence, one advantage of using this harmonic ratio to characterize magnetic properties of MNPs is that this parameter is independent of concentration/quantity of MNPs in the testing sample. S-9 Note S8. The R35 Heatmap Figure S2. The heatmap of R35 values at combinational H1N1 nucleoprotein concentrations and driving field frequencies. S-10 Note S9. Capabilities of the 3rd Harmonic, the 5th Harmonic, and the Harmonic Ratio R35 as Metrics for Distinguishing H1N1 Nucleoprotein Samples with High and Low Concentrations. It should be noted that both the 3rd and 5th harmonics show significantly weaker capabilities in detecting low concentrations of H1N1 nucleoprotein (see the red curves in Figure S3(a) & (b)). Under the driving field frequency of 1 kHz, the differences in 3rd harmonic amplitudes measured from samples VII (44 nM) and VIII (0 nM) is only 2.8%, this difference increases to 3.9% at a driving field frequency of 5 kHz (see Figure S3(a)). Similarly, the differences in 5th harmonic amplitudes from samples VII (44 nM) and VIII (0 nM) is only 2.6% at 1 KHz and increases to 3.1% at 5 kHz (see Figure S3(b)). However, neither the 3rd harmonic nor the 5th harmonic succeed in distinguishing samples I (4.42 µM) and II (2.21 µM) where the concentrations of H1N1 nucleoprotein are extremely high, at 20 kHz. Figure S3. (a) The 3rd harmonic, (b) the 5th harmonic, and (c) the harmonic ratio R35 as metrics for distinguishing samples with low and high H1N1 nucleoprotein concentrations. The grey area indicates that the metric fails to differentiate samples VII & VIII at 20 kHz. S-11 On the other hand, the harmonic ratio R35 demonstrates comparable capabilities in distinguishing high and low concentrations of H1N1 nucleoprotein samples (see Figure S3(c)). For the detection of samples with low concentrations of H1N1 nucleoproteins, the R35 differences from samples VII (44 nM) and VIII (0 nM) increase from 0.14% to 1.02% from 1 kHz to 5 kHz. At the higher end of H1N1 nucleoprotein concentrations, the harmonic ratio R35 shows stronger capability in distinguishing samples I (4.42 µM) and II (2.21 µM) with differences of 1.33% and 0.97% at 10 kHz and 15 kHz, respectively (see the red curve in Figure S3(c)). S-12 Table S1. The 3rd harmonic amplitudes (µV) from Figure 2(a):i -- iv. Sample Index 1 kHz 5 kHz 10 kHz 20 kHz I 138 219 II III IV V VI 141.2 148.4 150.75 162.6 168.6 VII 180 VIII IX 185 184.8 224.4 236.5 241.66 262 272.75 290 301.25 306.25 199.25 203.4 215 219.8 238 246.75 263.2 272.5 280.25 162.4 161.4 171.4 173.6 188 193.8 206.8 214.8 219 Table S2. The 5th harmonic amplitudes (µV) from Figure 2(b):v -- viii. Sample Index I II III IV V VI 89.728 91.441 95.878 96.711 104 107.4 1 kHz 5 kHz 10 kHz 20 kHz VII 114 VIII 116 IX 117 143 144.6 152.25 154.33 167 172.75 183 188.75 190 133.25 134.6 141 143 154.4 159.5 169.4 174.25 177.5 110 109.6 115.4 116.4 125.6 129.6 138 142.2 143.8 References [1] L. Tu, Y. Jing, Y. Li, J.-P. Wang, Applied Physics Letters 2011, 98, 213702; L. Tu, Y. Feng, T. Klein, W. Wang, J.-P. Wang, Magnetics, IEEE Transactions on 2012, 48, 3513; A. M. Rauwerdink, J. B. Weaver, Medical physics 2010, 37, 2587; K. Wu, A. Batra, S. Jain, C. Ye, J. Liu, J.-P. Wang, Applied Physics Letters 2015, 107, 173701; K. Wu, J. Liu, Y. Wang, C. Ye, Y. Feng, J.-P. Wang, Applied Physics Letters 2015, 107, 053701. R. J. Deissler, M. A. Martens, Magnetics, IEEE Transactions on 2015, 51, 1. T. Qu, R. Victora, Journal of Applied Physics 2014, 115, 17C506. J. Dieckhoff, D. Eberbeck, M. Schilling, F. Ludwig, Journal of Applied Physics 2016, 119, 043903; R. J. Deissler, Y. Wu, M. A. Martens, Medical physics 2014, 41, 012301; J. B. Weaver, A. M. Rauwerdink, C. R. Sullivan, I. Baker, Medical physics 2008, 35, 1988. J. Waldron, Y. P. Kalmykov, W. Coffey, Physical Review E 1994, 49, 3976. P. Cregg, D. Crothers, A. Wickstead, Journal of applied physics 1994, 76, 4900. S-13 [2] [3] [4] [5] [6]
1809.06197	1	1809	2018-09-17T13:42:57	Substrate patterning using regular macroporous block copolymer monoliths as sacrificial templates and as capillary microstamps	[ "physics.app-ph" ]	Polystyrene-block-poly(2-vinylpyridine) (PS-b-P2VP) monoliths containing regular arrays of macropores (diameter ~1.1 microns, depth ~700 nm) at their surfaces are used to pattern substrates by patterning modes going beyond the functionality of classical solid elastomer stamps. In a first exemplary application, the macroporous PS-b-P2VP monoliths are employed as sacrificial templates for the deposition NaCl nanocrystals and topographically patterned iridium films. One NaCl nanocrystal per macropore is formed by evaporation of NaCl solutions filling the macropores followed by iridium coating. Thermal PS-b-P2VP decomposition yields topographically patterned iridium films consisting of ordered arrays of hexagonal cells, each of which contains one NaCl nanocrystal. For the second exemplary application, spongy-continuous mesopore systems are generated in the macroporous PS-b-P2VP monoliths by selective-swelling induced pore generation. Infiltrating the spongy-continuous mesopore systems with ink allows capillary microstamping of continuous ink films with holes at the positions of the macropores onto glass slides compatible with advanced light microscopy. Capillary microstamping can be performed multiple times under ambient conditions without re-inking and without quality deterioration of the stamped patterns. The macroporous PS-b-P2VP monoliths are prepared by double replication of primary macroporous silicon molds via secondary polydimethylsiloxane molds.	physics.app-ph	physics	Substrate patterning using regular macroporous block copolymer monoliths as sacrificial templates and as capillary microstamps Leiming Guo(cid:1) Michael Philippi, Martin Steinhart(cid:1) Dr. L. Guo, M. Philippi, Prof. M. Steinhart Institut für Chemie neuer Materialien, Universität Osnabrück, Barbarastr. 7, 49069 Osnabrück, E-mail: [email protected]; [email protected] Keywords: Block copolymers, replication molding, stamping, porous materials, surface patterning 1 Abstract Polystyrene-block-poly(2-vinylpyridine) (PS-b-P2VP) monoliths containing regular arrays of macropores (diameter ~1.1 µm, depth ~0.7 µm) at their surfaces are used to pattern substrates by patterning modes going beyond the functionality of classical solid elastomer stamps. In a first exemplary application, the macroporous PS-b-P2VP monoliths are employed as sacrificial templates for the deposition NaCl nanocrystals and topographically patterned iridium films. One NaCl nanocrystal per macropore is formed by evaporation of NaCl solutions filling the macropores followed by iridium coating. Thermal PS-b-P2VP decomposition yields topographically patterned iridium films consisting of ordered arrays of hexagonal cells, each of which contains one NaCl nanocrystal. For the second exemplary application, spongy-continuous mesopore systems are generated in the macroporous PS-b-P2VP monoliths by selective-swelling induced pore generation. Infiltrating the spongy-continuous mesopore systems with ink allows capillary microstamping of continuous ink films with holes at the positions of the macropores onto glass slides compatible with advanced light microscopy. Capillary microstamping can be performed multiple times under ambient conditions without re-inking and without quality deterioration of the stamped patterns. The macroporous PS-b-P2VP monoliths are prepared by double replication of primary macroporous silicon molds via secondary polydimethylsiloxane molds. 2 Introduction Polymer monoliths containing macropores with diameters in the range from several 100 nm to a few µm are a versatile material platform for a diverse variety of applications including separation, catalysis, energy storage, energy conversion, drug delivery and tissue engineering. They may be employed as surfaces exhibiting specific wetting and adhesion properties or as microreactor arrays. Polymer monoliths with surfaces topographically patterned with macropore arrays are accessible by soft molding against polydimethylsiloxane (PDMS) molds[1] or by non-lithographic synthetic routes such as formation of breath figures,[2] spreading polymer solutions on non- solvents,[3] and macroscale spinodal decomposition.[4] Block copolymers (BCPs) consisting of at least two chemically distinct blocks offer additional synthetic pathways to macroporous polymer monoliths. Examples include electric-field-induced BCP patterning,[5] non-solvent-induced phase separation,[6] and breath figure formation on BCP solutions.[7] Of particular interest is the additional functionalization of macroporous polymer monoliths with continuous-spongy mesopore systems; BCPs already contain a mesoscopic fine structure that may serve as precursor for the mesoporous structure level. The chemically distinct blocks of the BCPs are typically incompatible so that they assemble into ordered nanoscopic domain structures.[8] Mesopores can be generated by selectively degrading one of the BCP components[9] or by selective-swelling induced pore generation.[10] For example, bioinspired anti-adhesive surfaces with hierarchical structure characterized by mesopore systems and macroscale topographic surface patterns were prepared by molding solutions of the BCP polystyrene-block-poly(2-vinylpyridine) (PS-b-P2VP) against PDMS molds derived from microsphere monolayers combined with selective-swelling induced pore generation.[10d] 3 Beyond the functionalities of conventional elastomeric stamps, the potential of macroporous polymer monoliths as a means to pattern substrates has by and large remained unexplored. Here, we demonstrate that PS-b-P2VP monoliths with surfaces topographically patterned with macroscopic monodomains of macropores allow modes of surface patterning not accessible by classical elastomeric specimens. First, we demonstrate that metal-coated macroporous PS-b- P2VP monoliths (such as macroporous monoliths of the corresponding homopolymers) containing one NaCl nanocrystal per macropore can be used as sacrificial template; thermal decomposition of the PS-b-P2VP yielded metal films topographically patterned with hexagonal arrays of cells replicating the macroporous structure of the sacrified macroporous PS-b-P2VP monoliths. Each cell contained one NaCl nanocrystal initially located in the macropore acting as precursor of the corresponding cell. In this work NaCl is used as a test system. However, the precise positioning of nanocrystals on surfaces demonstrated here may be a generic approach to two-dimensional nanocrystal arrays for electronic, optical, and sensor applications.[11] Furthermore, thin corrugated metal films may exhibit peculiar plasmonic properties potentially enabling peculiar light transmission modes.[12] Moreover, corrugated metal films may show anomalous reflection of light.[13] Finally, the cells of the corrugated metal films may be used as electrochemical reaction compartments. In the second application example, we generated spongy-continuous mesopore systems in macroporous PS-b-P2VP monoliths by selective- swelling induced pore generation and used the continuous-spongy mesopore systems as ink reservoirs for capillary microstamping. Transfer of the macroporous surface topography by capillary microstamping onto counterpart substrates yielded ink films containing holes at the position of the macropores; the use of the spongy-continuous mesopore system enabled multiple successive capillary microstamping cycles under ambient conditions without deterioration of the 4 quality of the stamped pattern. For example, hydrophilic-hydrophobic micropatterns consisting of a holey hydrophobic film containing arrays of hydrophilic dots may be generated in this way. Hydrophilic-hydrophobic micropatterns compatible with standard light microscopy, in which water drops located on the hydrophilic dots are immobilized by the surrounding hydrophobic films, may be promising lab-on-chip configurations for parallel chemical syntheses and screenings.[14] Macroporous PS-b-P2VP monoliths with a lattice constant of 1.5 µm and a macropore diameter of 1.1 µm were accessible by double replication of primary macroporous silicon molds.[15] Negative PDMS replicas of the macroporous silicon were used as secondary molds to form macroporous PS-b-P2VP monoliths as positive replicas of the primary macroporous silicon molds. 5 Figure 1. Preparation and use of ordered macroporous PS-b-P2VP monoliths. a) A PS-b-P2VP solution (light blue) is deposited onto a smooth silicon wafer (black). b) The PS-b-P2VP solution is imprinted with a secondary PDMS mold (green, negative replica of primary macroporous silicon mold). c) The PS-b-P2VP (blue) solidifies after evaporation of the solvent. d) A macroporous PS-b-P2VP monolith (blue, positive replica of the primary macroporous silicon mold and negative replica of the secondary PDMS mold) is obtained after detachment of the secondary PDMS mold. e)-h) In a first exemplary application, e) an aqueous NaCl solution (light red) is deposited onto the macroporous PS-b-P2VP monolith still attached to the underlying silicon wafer. f) After evaporation of the water, one NaCl nanocrystal (red) forms in each macropore. g) The macroporous PS-b-P2VP monolith and the NaCl nanocrystals are coated with a thin iridium layer (ochre). h) Heat treatment results in thermal decomposition of the PS-b-P2VP. The remaining iridium skin forms a cellular film on the underlying silicon wafer conserving the hexagonal array geometry of the macroporous PS-b-P2VP monolith; in each cell of the iridium skin one NaCl nanocrystal is located. i)-l) In the second application example, i) spongy-continuous mesopore systems are formed in the macroporous PS-b-P2VP monoliths by selective-swelling induced pore generation. j) The spongy-continuous mesopore system is filled with ink (orange). The ink-filled spongy macroporous PS-b-P2VP monolith is approached to a counterpart substrate, such as a glass slide (light grey). On the areas of the counterpart surface forming contact with the ink-filled spongy macroporous PS-b-P2VP monolith ink is deposited. l) After retraction of the spongy macroporous PS-b-P2VP monolith, the counterpart surface is patterned with a continuous ink film containing holes at the positions of the macropores. Steps i) -- l) can be repeated multiple times under ambient conditions without re- inking. 6 Results and discussion To prepare macroporous PS-b-P2VP monoliths, we used macroporous silicon with a macropore diameter of 1 µm and a macropore depth of 700 nm as primary mold. The macropores in the primary macoporous silicon mold formed extended hexagonal monodomains with a lattice constant of 1.5 µm. To obtain secondary PDMS molds (Supporting Figure S1) as negative replicas of the primary macroporous silicon molds, cross-linkable PDMS prepolymer formulations were molded against the macroporous silicon. The secondary PDMS molds were faithful negative replica of the primary macroporous silicon molds. To prepare the macroporous PS-b-P2VP monoliths, we dropped a solution containing 5 wt-% of asymmetric PS-b-P2VP (P2VP forms the minority domains) in tetrahydrofuran (THF) onto a silicon wafer (Figure 1a) and placed the secondary PDMS mold on the solution while a pressure of 3.9 kN/m2 was applied (Figure 1b). After evaporation of the THF for at least 15 min (Figure 1c) the secondary PDMS molds were nondestructively detached and macroporous PS-b-P2VP monoliths with an area of ~1 cm2 were obtained (Figure 1d). The macropores in the macroporous PS-b-P2VP monoliths formed hexagonal monodomains (Figure 2a; see Supporting Figure 2 for a large-field SEM image) replicating the macropore monodomains of the primary macroporous silicon molds. The macropores in the PS-b-P2VP monoliths had a diameter ~1.1 µm, whereas the pillar diameter of the secondary PDMS molds amounted to only ~1.0 µm (corresponding to the macropore diameter of the primary macroporous silicon molds). We assume that the secondary PDMS molds were swollen by THF during the molding of the PS-b-P2VP/THF solutions. The depth of the macropores in the PS-b-P2VP monoliths amounted to ~700 nm and corresponded to the heights of the pillars of the secondary PDMS molds (Supporting Figure S1). A ~1.5 µm thick 7 PS-b-P2VP layer separated the bottoms of the macropores in the PS-b-P2VP monoliths from the underlying Si wafer (Figure 2b). a) b) Figure 2. Scanning electron microscopy images of macroporous PS-b-P2VP monoliths corresponding to Figure 1d obtained with a solution of 5 wt-% PS-b-P2VP in THF by imprinting with a secondary PDMS mold (Figure S1) at a pressure of 3.9 kN/m2. a) Top view. b) Cross-section. The macroporous PS-b-P2VP monolith on top is connected to a Si wafer at the bottom. In a first example for substrate patterning with macroporous PS-b-P2VP monoliths we used the macroporous PS-b-P2VP monoliths as sacrificial templates. It should be noted that in this specific case macroporous monoliths consisting of the corresponding homopolymers could be used as well. At first, we applied the "discontinuous wetting" method[3, 16] to deposit one cubic NaCl nanocrystal with a diameter of ~320 nm into each macropore (Figure 1f and Figure 3a, for a large-field view see Supporting Figure S3). To this end, the macroporous PS-b-P2VP monoliths were immersed into an aqueous 2 M NaCl solution for 1 min (Figure 1e) and dried 8 under ambient conditions for 2 h. In the next step we sputter-coated the macroporous PS-b-P2VP monoliths containing the NaCl nanocrystals with a ~5 nm thick iridium layer (cf. Figure 1g). Then, the samples were heated to 540°C at a rate of 1K/min and kept at this temperature in air for 3 h. This thermal treatment results in thermal decomposition of the PS-b-P2VP.[17] After the heat treatment, the Si wafer was covered with a regularly corrugated iridium film containing a hexagonal pattern of cells separated by iridium rims (for large-field views see Supporting Figure S4a,b) with a heigth of ~160 nm (Supporting Figure S4c). While the cells mark the initial positions of the macropores in the macroporus PS-b-P2VP monoliths, the rims are a signature of the PS-b-P2VP walls that separated the macropores. Each cell contains one NaCl nanocrystal (Figure 3b,c and Supporting Figure S4b), as schematically displayed in Figure 1h. 9 a) b) c) Figure 3. Macroporous PS-b-P2VP monoliths as sacrificial templates. Scanning electron microscopy image of a) a macroporous PS-b-P2VP monolith containing one NaCl nanocrystal in each macropore corresponding to panel f) of Figure 1. b), c) Scanning electron microscopy images of the product obtained by annealing an iridium-coated macroporous PS-b-P2VP monolith containing one NaCl nanocrystal in each macropore at 540°C in air. Seen is the remaining iridium skin forming a hexagonal array of cells each of which contains one NaCl nanocrystal corresponding to panel h) of Figure 1. 10 The second exemplary application of macroporous PS-b-P2VP monoliths is the stamping of continuous ink films with holes at the position of the macropores onto counterpart substrates. We have recently reported capillary microstamping of arrays of discrete ink spots using spongy- mesoporous stamps with contact surfaces topographically patterned with rod-like contact elements.[18] While classical stamping procedures such as polymer pen lithography[19] rely on solid stamps requiring adsorption of non-volatile ink components onto the outer stamp surface, capillary microstamping with spongy mesoporous stamps allows either continuous ink supply to the contact surface of the stamps via the spongy mesopore system or the use of the spongy mesopore system as ink reservoir. Multiple capillary microstamping cycles can be carried out manually under ambient conditions without complex alignment procedures, without re-inking and without deterioration of the quality of the stamped pattern. So far, only arrays of discrete dots were accessible by capillary microstamping. It is desirable to expand the range of accessible surface patterns and to generate continuous ink films containing regular arrays of holes by capillary microstamping. For this purpose, we converted solid macroporous PS-b-P2VP monoliths into spongy macroporous PS-b-P2VP monoliths containing continuous mesopore systems as additional hierarchical structure level by selective-swelling induced pore generation.[10] Selective-swelling induced pore generation can, in principle, be applied to any asymmetric BCP on condition that a solvent selective to the minority blocks exists. In this way, mesoporous PS-b-P2VP specimens for membrane separation[20] as well as for metal[10a] and metal oxide deposition[17, 21] were obtained. However, ultrafiltration membranes were also obtained by subjecting polystyrene-b-poly(methyl methacrylate) to selective-swelling induced pore generation.[22] Here, the macroporous PS-b-P2VP monoliths were treated with ethanol at 70°C for 30 minutes. Osmotic pressure drives ethanol, a solvent selective to P2VP, into the 11 P2VP minority domains. The P2VP blocks adopt stretched conformations to maximize contact to the ethanol molecules. Hence, the P2VP minority domains swell, whereas the glassy PS matrix undergoes morphology reconstruction in order to accommodate the increasing volumes of the swelling P2VP minority domains that eventually form a continuous network. Evaporation of the ethanol results in entropic collapse of the stretched P2VP blocks that relax into their native coiled conformations while the glassy PS matrix fixates the reconstructed morphology. Thus, continuous mesopore systems form in lieu of the swollen P2VP minority domains, the walls of which consist of coiled P2VP blocks[23] (Figure 1i). The mesopore systems generated in the macroporous PS-b-P2VP monoliths were open to the environment and had pore sizes ranging from 25 to 60 nm (Figure 4a). They penetrated the entire PS-b-P2VP monoliths (Figure 4b). The regular arrays of macropores were retained during selective-swelling induced pore generation (for a large-field SEM image see Supporting Figure S5). The thickness of the spongy macroporous PS-b-P2VP monoliths and the depth of the macopores nearly doubled to 3.9 µm and 1.4 µm because of the volume expansion in the vertical direction during selective-swelling induced pore formation.[24] 12 a) b) c) 13 Figure 4. a), b) Scanning electron microscopy images of a spongy macroporous PS-b-P2VP monolith corresponding to panel i) of Figure 1. a) Top view; b) cross section showing the spongy macroporous PS-b-P2VP monolith on top and the underlying Si wafer at the bottom. c) TIRFM image of a glass slide patterned with a holey rhodamine B film by stamping an aqueous rhodamine B solution infiltrated into a spongy macroporous PS-b-P2VP monolith under ambient conditions. The displayed image shows the result of the 5th consecutive stamping cycle without reloading the spongy macroporous PS-b-P2VP monolith with ink. The displayed image field has an area of 53.91 μm × 53.91 μm. A spongy macroporous PS-b-P2VP monolith with a size of 0.6 cm × 0.6 cm still connected to a Si wafer was used as stamp. The underside of the Si wafer was glued onto a stamp holder with a mass of 27 g in such a way that the macropores of the PS-b-P2VP monolith were exposed. The pressure imposed by the mass of the stamp holder was estimated to be 7.35 kN/m2. In a first stamping experiment we used a solution of 0.05 wt-% of the dye rhodamine B in water as ink. We dropped 50 μL of the aqueous rhodamine B solution onto the macroporous surfaces of the PS-b-P2VP monoliths to fill the mesopore systems with ink (Figure 1j). After 1 min, residual ink was removed from the macroporous surface of the PS-b-P2VP monoliths with filter paper. Then, stamping on glass slides was instantly performed under ambient conditions with a contact time of 1 s per stamping cycle by bringing the macroporous surface of the spongy macroporous PS-b- P2VP monoliths into contact with the glass slides (Figure 1k). Supporting Figure S6 shows a total internal reflection fluorescence microscopy (TIRFM) image of the result of the first stamping cycle after inking, whereas Figure 4c displays the result of the 5th consecutive stamping cycle without re-inking. The patterns obtained in this way are faithful replicas of the macropore arrays of the spongy macroporous PS-b-P2VP monoliths (Figure 1l). The areas in which the walls surrounding the macropores contacted the underlying glass slides showed fluorescence, indicating that in these areas ink containing rhodamine B was transferred to the underlying glass slide. Conversely, at the positions of the macropores no rhodamine B was deposited onto the 14 glass slides and no fluorescence can be seen. As a result, a continuous rhodamine B film was formed that contained a hexagonal array of quadratic holes with edge lengths of ~950 nm and areas of ~0.9 µm2. Since the lattice constant of the hexagonal array amounts to 1.5 µm, the rhodamine B film covers 60 % of the surface of the glass slide. The comparison of Supporting Figure S6 and Figure 4c reveals that the quality of the stamped pattern did not deteriorate. a) b) 15 c) Figure 5. Stamping of 1-dodecanthiol films that contain holes at the positions of the macropores of the spongy macroporous PS-b-P2VP monoliths used as stamps onto gold-coated glass slides, followed by partial etching of the gold exposed in the holes not protected by 1-dodecanetiol. a) AFM topography image of a sample produced by the 1st stamping cycle; b) AFM topography image of a sample produced by the 5th consecutive stamping cycle. In panels a) and b) the dark dots represent the unprotected (partially etched) gold surface at the positions of the macropores of the spongy macroporous PS-b-P2VP monoliths, whereas 1-dodecanethiol was printed onto the continuous elevated areas separating the dark dots. c) TIRFM image of a sample prepared like that shown in panel a) after immersion into an aqueous solution of 0.05 wt-% rhodamine B for 30 s. The image field of panel c) has a width of 38.54 µm and a height of 36.12 µm. In a second stamping experiment we stamped 1-dodecanethiol films with regular arrays of holes at the positions of the macropores of the spongy macroporous PS-b-P2VP monoliths used as stamps onto gold-coated glass slides. The spongy macroporous PS-b-P2VP monoliths were infiltrated with 20 µL of a 3 mM ethanolic 1-dodecanethiol solution After 30 s we removed residual solution from the macroporous surface of the spongy macroporous PS-b-P2VP monoliths with filter paper and started stamping with a contact time of 1 s per cycle immediately thereafter. The exposed gold not protected by adsorbed 1-dodecanethiol was partially etched following procedures reported elsewhere [18b, 25] to enhance contrast in the following microscopic 16 investigations. As revealed by AFM measurements (Figure 5a,b), the thickness of the gold layer was reduced from initially 30 nm to 10-15 nm. As obvious from the AFM height images, the gold was etched in square-shaped dot-like areas with edge lengths of ~1.0 µm forming hexagonal arrays with a lattice constant of ~1.5 µm. Hence, no 1-dodecanethiol was deposited in the square- shaped dot-like areas that replicated the positions of the macropores of the spongy macroporous PS-b-P2VP monoliths. On the other hand, the continuous 1-dodecanethiol films surrounding the square-shaped dot-like areas covered ~56 % of the surface, which is consistent with the results obtained by stamping aqueous rhodamine B ink (Figure 4c). Immersion of thus-treated surfaces into an aqueous rhodamine B solution resulted in selective adsorption of the rhodamine B onto the exposed gold surfaces. Therefore, strong rhodamine B fluorescence occurred in the discrete dot-like areas not covered by 1-dodecanethiol at the positions of the macropores of the spongy macroporous PS-b-P2VP monoliths (Figure 5c). Conclusion We have explored non-elastomeric monoliths consisting of the block copolymer PS-b-P2VP having surfaces topographically patterned with regular arrays of micron-sized macropores as means to pattern substrates. The macroporous PS-b-P2VP monoliths were prepared by a double replication process. Ordered macroporous silicon containing wafer-scale macropore monodomains was used as primary mold. PDMS secondary molds with pillars as negative replicas of the macropores of the primary macroporous silicon molds were in turn molded against PS-b-P2VP solutions to obtain the macroporous PS-b-P2VP monoliths. The macroporous PS-b-P2VP monoliths were, therefore, positive replicas of the primary macroporous silicon molds and negative replicas of the secondary PDMS molds. Macroporous PS-b-P2VP monoliths 17 enable modes of substrate patterning that go beyond the functionality of solid elastomeric stamps. In a first application example (also possible with macroporous monoliths consisting of the corresponding homopolymers) in each macropore of the PS-b-P2VP monoliths attached to underlying silicon wafers one NaCl nanocrystal was deposited. After coating the macroporous PS-b-P2VP monoliths with thin iridium layers the PS-b-P2VP was decomposed by heat treatment. The remaining iridium skins formed corrugated films on the silicon wafers that exhibited regular arrays of hexagonal cells replicating the macropores of the destroyed macroporous PS-b-P2VP monoliths. Exactly one NaCl crystal was located in each cell. In a second model application the macroporous PS-b-P2VP monoliths were further functionalized with a continuous spongy mespore system by selective-swelling induced pore generation. Using the spongy mesopore system as ink reservoir enabled capillary microstamping that yielded continuous ink films with holes at the positions of the macropores. Multiple capillary microstamping cycles could be carried out under ambient conditions without quality deterioration of the stamped patterns. These examples demonstrate that substrate patterning with non-elastomeric polymer monoliths -- including the deposition of nanocrystals -- may be employed to generate preconcentration sensors, devices for data storage and energy conversion, as well as surfaces with specific adhesion, wettability and tailored interactions with biological cells. Experimental Section Materials: Macroporous silicon with a pore diameter of 1 µm, a pore depth of 700 nm and a lattice constant of 1.5 µm was obtained from SmartMembranes (Germany). Cylinder-forming PS-b-P2VP (Mn (PS) = 101 kg/mol, Mn (P2VP) = 29 kg/mol, polydispersity = 1.6) was purchased 18 from Polymer Source Inc. (Canada). Tetrahydrofuran (THF, ≥99.9%), ethanol (≥99.9%), rhodamine B (≥95%) and 1-dodecanethiol (≥98%) were obtained from Sigma Aldrich. PDMS prepolymer formulation (Sylgard® 184) was supplied by Dow Corning Corporation. K2S2O3, KOH, K3Fe(CN)6 and K4Fe(CN)6 were obtained from Alfa Aesar; 1-octanol was purchased from Merck. All chemicals were used without further purification. Silicon wafers as the substrates to prepare the macroporous PS-b-P2VP monoliths were cut into the size of 2 cm × 2 cm, ultrasonicated at least three times in ethanol and dried before use. Preparation of macroporous PS-b-P2VP monoliths: To prepare PDMS secondary molds, base and curing agent of the PDMS prepolymer formulation were mixed at a weight ratio of 10:1 under stirring. After degassing for 1 h under ambient conditions at room temperature, the PDMS prepolymer mixture was poured onto the macroporous silicon primary molds and cured at 70 ºC for 12 h. The PDMS secondary molds with a thickness of ~1 mm were detached from the macroporous silicon primary molds and cut into pieces extending 1 cm × 1 cm. A portion of 20 µL of a solution containing 5 wt-% PS-b-P2VP in THF was drop-cast on smooth silicon wafers at room temperature and instantly covered with the PDMS secondary molds under a pressure of 3.9 kN/m2. After a wait time of 15 min the PDMS secondary molds were detached from the macroporous PS-b-P2VP monoliths obtained in this way. Use of macroporous PS-b-P2VP monoliths as sacrificial templates: The macroporous PS-b- P2VP monoliths were immersed into aqueous solutions containing 2 mol/L NaCl for 1 min followed by drying for 2 h under ambient conditions at room temperature. The macroporous PS- b-P2VP monoliths containing one NaCl nanocrystal per macropore were sputter-coated with a ~5 nm thick iridium layer (Sputter Angle Emitech K575X, UK). The thickness of the iridium layer was determined by AFM (see below). Thermal decomposition of the PS-b-P2VP was carried out 19 by heating the samples to 540°C at a rate of 1 K/min followed by annealing at this temperature for 3 h in air. Capillary microstamping: To fabricate spongy macroporous PS-b-P2VP monoliths, solid macroporous PS-b-P2VP monoliths still attached to underlying Si wafers were immersed into ethanol at 70 ºC for 30 min followed by drying under ambient conditions. The spongy macroporous BCP monoliths along with the underlying Si wafers were cut into pieces extending 0.6 cm × 0.6 cm and glued onto cylindrical sample holders with a weight of 27 g in such a way that the macropores of the spongy macroporous PS-b-P2VP monoliths were exposed. To stamp rhodamine B, 50 µL portions of aqueous solutions containing 0.05 wt-% rhodamine B were drop-cast onto the spongy macroporous PS-b-P2VP monoliths within 1 min. Residual solution was removed using filter paper. Then, the spongy macroporous PS-b-P2VP monoliths were instantly brought into contact with glass slides for ~1 s. To stamp 1-dodecanethiol, glass slides were coated with a ~7 nm thick titanium layer and a ~30 nm thick gold layer using a Balzers BAE 120 evaporator applying recipes reported elsewhere.[18b] 20 µL portions of a 3 mM 1- dodecanethiol solution in ethanol were infiltrated into the spongy macroporous PS-b-P2VP stamps. Residual solution was removed using filter paper after 30 s. Stamping onto the gold- coated glass slides was performed immediately thereafter with a contact time of ~1 s. To partially etch the gold layer, 0.248 g K2S2O3, 0.261 g KOH, 0.032 g K3Fe(CN)6, 0.0042 g K4Fe(CN)6 and 3.18 µL 1-octanol were mixed with 10 mL deionized water. Gold-coated glass slides onto which holey 1-dodecanethiol films had been stamped were immersed into the thus- obtained solution for 6 min followed by rinsing with deionized water and ethanol. To adsorb rhodamine B onto samples treated in this way, the samples were immersed into an aqueous solution of 0.05 wt-% rhodamine B for 30 s. 20 Characterization: Scanning electron microscopy was carried out using a Zeiss Auriga device operated at an accelerating voltage of 7 kV. Before SEM investigations the samples were sputter- coated with a ~5 nm thick iridium layer. AFM was carried out using a NTEGRA microscope (NT-MDT). TIRF-Microscopy was carried out with an inverted microscope (Olympus IX71) equipped with a 4-Line TIRF condenser (Olympus TIRF 4-Line). A 150-fold oil immersion Objective (Olympus UAPON 150x TIRF, NA 1.45) was used in combination with a 561 nm diode-pumped solid state laser (Cobolt Jive 561). A TIRF pentaband polychroic beamsplitter (Semrock zt405/488/561/640/730rpc) and a pentabandpass emitter (BrightLine HC 440/521/607/694/809) were used. Images were acquired by an electron multiplying CCD camera (Andor iXon Ultra 897) and processed using Olympus CellSens Software. Supporting Information Supporting Information is available from the author. Acknowledgements The authors thank the European Research Council (ERC-CoG-2014, project 646742 INCANA) for funding. REFERENCES [1] [2] [3] a) Y. Xia, E. Kim, X. M. Zhao, J. A. Rogers, M. Prentiss, G. M. Whitesides, Science 1996, 273, 347; b) Y. Xia, G. M. Whitesides, Angew. Chem. Int. Ed. 1998, 37, 550; c) K. Y. Suh, Y. S. Kim, H. H. Lee, Adv. Mater. 2001, 13, 1386; d) Y. S. Kim, K. Y. Suh, H. H. Lee, Appl. Phys. Lett. 2001, 79, 2285; e) C. M. Bruinink, M. Peter, P. A. Maury, M. De Boer, L. Kuipers, J. Huskens, D. N. Reinhoudt, Adv. Funct. Mater. 2006, 16, 1555; f) K. Y. Suh, M. C. Park, P. Kim, Adv. Funct. Mater. 2009, 19, 2699. a) U. H. F. Bunz, Adv. Mater. 2006, 18, 973; b) L. S. Wan, L. W. Zhu, Y. Ou, Z. K. Xu, Chem. Commun. 2014, 50, 4024. Y. Wang, Z. M. Liu, B. X. Han, Z. Y. Sun, J. L. Zhang, D. H. Sun, Adv. Funct. Mater. 2005, 15, 655. 21 [4] [5] [6] [7] [8] [9] [10] H. Sai, K. W. Tan, K. Hur, E. Asenath-Smith, R. Hovden, Y. Jiang, M. Riccio, D. A. Muller, V. Elser, L. A. Estroff, S. M. Gruner, U. Wiesner, Science 2013, 341, 530. H. Q. Xiang, Y. Lin, T. P. Russell, Macromolecules 2004, 37, 5358. K. V. Peinemann, V. Abetz, P. F. Simon, Nat. Mater. 2007, 6, 992. A. Munoz-Bonilla, E. Ibarboure, E. Papon, J. Rodriguez-Hernandez, Langmuir 2009, 25, 6493. V. Abetz, P. F. W. Simon, Adv. Polym. Sci. 2005, 189, 125. a) A. S. Zalusky, R. Olayo-Valles, J. H. Wolf, M. A. Hillmyer, J. Am. Chem. Soc. 2002, 124, 12761; b) M. A. Hillmyer, Adv. Polym. Sci. 2005, 190, 137. a) Y. Wang, C. He, W. Xing, F. Li, L. Tong, Z. Chen, X. Liao, M. Steinhart, Adv. Mater. 2010, 22, 2068; b) Y. Wang, F. Li, Adv. Mater. 2011, 23, 2134; c) Y. Wang, Acc. Chem. Res. 2016, 49, 1401; d) A. Eichler-Volf, L. Xue, G. Dornberg, H. Chen, A. Kovalev, D. Enke, Y. Wang, E. V. Gorb, S. N. Gorb, M. Steinhart, ACS Appl. Mater. Interfaces 2016, 8, 22593. [11] A. N. Shipway, E. Katz, I. Willner, Chemphyschem 2000, 1, 18. [12] a) G. A. Farias, A. A. Maradudin, V. Celli, Surf. Sci. 1983, 129, 9; b) R. W. Gruhlke, W. R. Holland, D. G. Hall, Phys. Rev. Lett. 1986, 56, 2838; c) S. Wedge, I. R. Hooper, I. Sage, W. L. Barnes, Phys. Rev. B 2004, 69; d) J. Feng, T. Okamoto, Opt. Lett. 2005, 30, 2302. I. F. Salakhutdinov, V. A. Sychugov, A. V. Tishchenko, B. A. Usievich, O. Parriaux, F. A. Pudonin, Quantum Electron. 1997, 27, 795. [13] [14] E. Ueda, P. A. Levkin, Adv. Mater. 2013, 25, 1234. [15] a) V. Lehmann, H. Föll, J. Electrochem. Soc. 1990, 137, 653; b) A. Birner, U. Grüning, S. Ottow, A. Schneider, F. Müller, V. Lehmann, H. Föll, U. Gösele, Phys. Status Solidi A 1998, 165, 111. [19] [16] V. R. Thalladi, G. M. Whitesides, J. Am. Chem. Soc. 2002, 124, 3520. [17] F. Li, X. Yao, Z. Wang, W. Xing, W. Jin, J. Huang, Y. Wang, Nano Lett. 2012, 12, 5033. a) W. Han, P. Hou, S. Sadaf, H. Schäfer, L. Walder, M. Steinhart, ACS Appl. Mater. [18] Interfaces 2018, 10, 7451; b) M. Schmidt, M. Philippi, M. Münzner, J. M. Stangl, R. Wieczorek, W. Harneit, K. Müller-Buschbaum, D. Enke, M. Steinhart, Adv. Funct. Mater. 2018, 28, 1800700. a) F. Huo, Z. Zheng, G. Zheng, L. R. Giam, H. Zhang, C. A. Mirkin, Science 2008, 321, 1658; b) D. J. Eichelsdoerfer, X. Liao, M. D. Cabezas, W. Morris, B. Radha, K. A. Brown, L. R. Giam, A. B. Braunschweig, C. A. Mirkin, Nat. Protoc. 2013, 8, 2548. a) X. Yao, L. Guo, X. Chen, J. Huang, M. Steinhart, Y. Wang, ACS Appl. Mater. Interfaces 2015, 7, 6974; b) L. Guo, Y. Wang, Chem. Commun. 2014, 50, 12022. J. Yang, L. Tong, Y. Yang, X. Q. Chen, J. Huang, R. Z. Chen, Y. Wang, J. Mater. Chem. C 2013, 1, 5133. [20] [21] [22] H. Ahn, S. Park, S. W. Kim, P. J. Yoo, D. Y. Ryu, T. P. Russell, ACS Nano 2014, 8, 11745. [23] Y. Wang, U. Gösele, M. Steinhart, Nano Lett. 2008, 8, 3548. [24] X. Y. Jun Yin, Jiun-You Liou, Wei Sun, Ya-Sen Sun, Yong Wang, ACS Nano 2013, 7, 9961. [25] Y. N. Xia, X. M. Zhao, E. Kim, G. M. Whitesides, Chem. Mater. 1995, 7, 2332. 22
1801.00833	1	1801	2017-12-29T14:11:08	The Effect of Metal Thickness on Si Wire to Plasmonic Slot Waveguide Mode Conversion	[ "physics.app-ph", "physics.optics" ]	We investigate mode converters for Si wire to plasmonic slot waveguides at 1550 nm telecom wavelength. The structures are based on a taper geometry. We provide optimal dimensions with more than 90% power transmission for a range of metal (Au) thicknesses between 30-250 nm. We provide details on how to differentiate between the total power and the power in the main mode of the plasmonic slot waveguide. Our analysis is based on the orthogonality of modes of the slot waveguide subject to a suitable inner product definition. Our results are relevant for lowering the insertion loss and the bit error rate of plasmonic modulators.	physics.app-ph	physics	The Effect of Metal Thickness on Si Wire to Plasmonic Slot Waveguide Mode Conversion S¸ ukru Ekin Kocabas¸ 1 7 1 0 2 c e D 9 2 ] h p - p p a . s c i s y h p [ 1 v 3 3 8 0 0 . 1 0 8 1 : v i X r a Abstract-We investigate mode converters for Si wire to plasmonic slot waveguides at 1550 nm telecom wavelength. The structures are based on a taper geometry. We provide optimal dimensions with more than 90% power transmission for a range of metal (Au) thicknesses between 30-250 nm. We provide details on how to differentiate between the total power and the power in the main mode of the plasmonic slot waveguide. Our analysis is based on the orthogonality of modes of the slot waveguide subject to a suitable inner product definition. Our results are relevant for lowering the insertion loss and the bit error rate of plasmonic modulators. Index Terms-Si photonics, plasmonic slot waveguides, mode converters I. INTRODUCTION The practice of using waveguides to transfer information be- tween different points in space with high bandwidth photonic links is gaining popularity. Fiber optics have replaced electrical wiring for a range of distances from thousands of kilometers coast to coast long-haul connections, to meter scaled rack to rack communications in data centers. The burgeoning field of silicon (Si) photonics has made it possible to build integrated opto-electronic components for a more intimate and efficient coupling of electronics for data processing and optics for data transfer. Si photonic links are based on the Si wire waveguides which are typically fabricated on silicon on insulator (SOI) wafers [1]. In addition to waveguides, photonic links also require light sources, modulators for electrical to photonic and photode- tectors for photonic to electrical conversion of information. Mach-Zehnder interferometers and resonant microring cavities have been used in Si modulators [2]. Both technologies have relatively large power consumption. In order to have energy efficient and small footprint modulators, recent designs incorporated metallic (i.e. plasmonic) and Si wire waveguides [3], [4]. The coupling rate from Si wire to plasmonic slot waveguides is an important parameter that contributes to the insertion loss and hence the bit error rate of the modulators [5]. Previous studies on Si wire to plasmonic slot waveguide transitions focused mostly on relatively thick, 180-250 nm Au layers [6], [7], [8], [9]. A recent study focused on layers of 20-30 nm thickness [10]. The propagation length of the modes in plasmonic slot waveguides changes as a function Department of Electrical and Electronics Engineering at Koc¸ University, Istanbul, Turkey. [email protected] Preprint version of the online copy available at http://dx.doi.org/10.19113/ sdufbed.88342 for the "Suleyman Demirel Universitesi Fen Bilimleri Enstitusu Dergisi." of the slot dimensions [11]. There is a trade-off between the amount of field enhancement due to the small size of the slots and the propagation distance of the ensuing modes. It would be advantageous to choose the necessary field enhancement based on the light propagation constraints of the application at hand. Field enhancement and metal layer thickness are two closely coupled variables. In this work, we cover a range of Au layer thicknesses between 30-250 nm and provide blueprints for coupling geometries that work at 1550 nm telecom wave- length range with more than 90% coupling efficiency. We use numerical simulations to verify our designs and we give details of our modeling techniques which use both simple power extraction as well as modal decomposition of scattered fields for estimating the transmission efficiency of the couplers. II. MATERIAL AND METHOD The mode coupler parameters that we employ are illustrated in Fig. 1. The aim is to convert the Si wire mode on the left to the mode of the plasmonic slot waveguide on the right, through the taper region at the center. As shown in Fig. 1(b), the Si wire and the slot waveguide are centered with respect to one another in the y direction. The ambient environment is SiO2. Such vertical alignment is possible with clean room fabrication techniques as discussed in [12]. We obtain the optical properties of Si from the Sellmeier- type dispersion formula quoted in [13]. The refractive index of glass is obtained from Eq. (20) in [14]. The permittivity of Au is obtained from the supplemental data provided in [15]. Since we will be working at a wavelength of λ = 1550 nm, we quote the relevant optical constants of Si, SiO2 and Au in Table I. We use the exp(+iωt) convention, thus Im(εAu) < 0. PERMITTIVITIES OF SI, SIO2 AND AU AT 1550 NM. TABLE I Material Si SiO2 Au Relative Permittivity (ε/ε0) 12.085 2.0852 −126.80− 5.3664i We use the finite element method implementation of the COMSOL package (v5.1) to solve for the waveguide modes of the Si wire and plasmonic slot waveguides as a function of the height of Si (hSi) and Au (hAu) for fixed values of Si width (wSi) and slot width (wslot). Typical mode profiles for (cid:107)E(cid:107) are plotted in Fig. 2. The Si wire and plasmonic slot waveguide modes are TE-like with the E-field primarily in the x direction [11], [1]. The plasmonic slot waveguide concentrates the fields 2 Fig. 3. Effective index (neff) at λ = 1550 nm as a function of waveguide height for (a) plasmonic slot waveguide with wslot = 30,50,100,220 nm and (b) Si wire waveguide with wSi = 400 nm. Fig. 1. Geometry and the definitions of the parameters of the mode coupler. a) Top view, b) Cross section view superposing the cuts at the A− A(cid:48) and B− B(cid:48) planes. within the slot region. The Si wire guides light primarily within the Si region, with evanescent waves leaking to the SiO2 region. Fig. 4. Mode intensity propagation length (Lp) at λ = 1550 nm as a function of waveguide height for plasmonic slot waveguide with wslot = 30,50,100,220 nm. and perfect magnetic conductor (PMC) boundary conditions at planes that vertically and horizontally bisect the waveguides, respectively [see Fig. 1(b)]. These boundary conditions enable us to reduce the number of unknowns by 4-fold, speeding up the simulations. Fig. 2. Plots of (cid:107)E(cid:107) for (a) Si wire and (b) plasmonic slot waveguide modes. The effective index (neff) of the modes are plotted in Fig. 3 as a function of waveguide height. At large heights, the plasmonic slot waveguide modes approach the 2D metal- insulator-metal modes [11] which have larger neff for smaller slot widths. The price to pay for the large neff is a reduced propagation length (Lp) as shown in Fig. 4. Lp is for the mode −1 2Im(kz) where kz is the wave vector intensity and is given by of the mode in the direction of propagation. For the Si wire waveguide, the mode approaches the 2D slab waveguide mode as the height increases. We use COMSOL to simulate the 3D geometry as depicted in Fig. 1. We surround the simulation volume by perfectly matched layers. We source the Si wire waveguide mode from the left and solve for the fields at λ = 1550 nm, as we vary various geometric variables. We take advantage of the symmetry of the waveguide modes and of the geometry in the x−y plane, by putting perfect electric conductor (PEC) We record the tangential (Ex,Ey,Hx,Hy) fields along the straight section of the plasmonic slot waveguide, at different z cuts, starting from z = 0 as shown in Fig. 1(a). We calculate the time averaged total power by numerically integrating 2 Re(E × H∗)z = the z component of the Poynting vector, 1 2 Re(ExH∗ 1 x ) at different z cuts, as shown by the blue '×' symbols in Fig. 5. Fitting an exponential to these data points (red curve) give us a propagation length LTotal = 5565 nm, which is smaller than Lp = 6105 nm for a slot waveguide of hAu = wslot = 30 nm. y − EyH∗ The fields along different z cuts of the slot waveguide in- clude non-bound scattered fields as well. In order to determine the power in the bound mode, we make a modal expansion of the fields. The total electric and magnetic fields at different z cuts, E,H can be expressed as E =αE0 +∑ H =αH0 +∑ n n βnEn, βnHn, (1) (2) where E0,H0 are the waveguide modes of the slot waveguide as illustrated in Fig. 2(b) and En,Hn are a decomposition of wgapstartgapslotextrawSiwslotwend(cid:96)taperAuSixzAA(cid:48)BB(cid:48)z=0(a)(b)yxpecpmcwSiwslothAuhSiSiO2(a)SiO2Si(b)SiO2Au0501001502002501.61.822.22.42.6wslot=30nm50100220(a)hAu(nm)neff01503004506001.522.53wSi=400nm(b)hSi(nm)05010015020025015101520253035×1033050100220hAu(nm)Lp(nm)the scattered fields. The coefficient α is what we are after. We can obtain the power in the bound mode by calculating (cid:21) 3 nm thick Au layers, respectively. We searched around these design points, varied hSi and other dimensions to come up with the optimized parameters in Table II. (cid:20)(cid:90)(cid:90) 1 Re 2 = α2 1 2 (cid:20)(cid:90)(cid:90) (αE0 × α∗H∗ (Ex0H∗ Re 0)z dxdy y0 − Ey0H∗ (cid:21) x0) dxdy . (3) OPTIMAL DIMENSIONS FOR hAU = 30 & 250 NM TABLE II Parameter Name hAu hSi wSi wslot (cid:96)taper wend wgap start gap slot extra Set 1 (nm) 30 300 400 30 600 0 20 200 200 Set 2 (nm) 250 725 400 250 1700 0 75 200 200 The norm of the electric field on the central plane that cuts through the structure is plotted in Fig. 6. The hAu = 30 nm case has a shorter (cid:96)taper, and the field intensities in the slot region are higher due to the 30 nm width of the slot. This case has a transmission factor of ∼88%. The hAu = 250 nm case has a longer taper and a larger slot width of 250 nm with a transmission factor of ∼95%. It is noteworthy that both sets have relatively thick Si layers, hSi (cid:29) hAu. (cid:90)(cid:90) In order to obtain the α coefficient, we make use of the orthogonality property of the bound modes that can be proved through the use of the Lorentz theorem. We define an inner product between two sets of fields (E1,H1) and (E2,H2) similar to the one in [16] as (E1,H1)· (E2,H2) ≡ 1 2 With this definition, we can take the inner product of the fields from a fixed z cut in the 3D simulation (1)-(2) with the bound modes calculated before, (E0,H0), and use the fact that the bound modes are orthogonal to the non-bound modes, i.e. (E0,H0)· (En,Hn) = 0 to get (E1 × H2 + E2 × H1)z dxdy. (4) α = (E,H)· (E0,H0) (E0,H0)· (E0,H0) . (5) We calculate α by numerically integrating the relevant fields in (5). We calculate the power in the bound modes from (3), and plot them as magenta '+' symbols in Fig. 5. As expected, the power in the bound mode of the slot waveguide is less than the total power in a given z-cut. When we fit an exponential to the power in bound modes (cyan curve), we get a decay length of LMode = 6283 nm, a result much closer to Lp. We estimate the power conversion efficiency of a given taper geometry by calculating the power at the z = 0 cut. We use the total power at z = 1100 nm, and estimate the power at z = 0 by using the Lp value from mode analysis, see the black dashed lines in Fig. 5. This is an approximate method, albeit a practical one, which gives a close estimate of the total power in the bound mode of the waveguide at z = 0 (slightly less than 88% in this example). Fig. 5. Power transmission factor as a function of z for a slot waveguide of hAu = wslot = 30 nm. Different calculation methods are explained in the text. III. RESULTS We used the reported values in [8] and [10] for mode conversion to plasmonic slot waveguides with 250 nm and 30 Fig. 6. Plot of (cid:107)E(cid:107) through the central plane. After we had the optimal values for two different Au thicknesses, we linearly interpolated all the geometry variables in between the two sets listed in Table II and calculated the transmission factor of the resulting mode converter structures corresponding to Au thicknesses ranging from 30 to 250 nm. The results are provided in Fig. 7. We measured the power at the z = 1100 nm cut, and back propagated to z = 0 by multiplying the results with exp(1100/Lp) where Lp values are calculated separately for each (hAu,wslot) pair similar to Fig. 4. 02004006008001,0000.740.760.780.80.820.840.860.88LTotal=5565nmLMode=6283nmLp=6105nmz(nm)TransmissionTotalModeFitTotalFitModeFitApp.(a)Set1forhAu=30nm(b)Set2forhAu=250nm4 [8] Chen, C.-T., Xu, X., Hosseini, A., Pan, Z., Subbaraman, H., Zhang, X., Chen, R. T. 2015. Design of highly efficient hybrid Si-Au taper for dielectric strip waveguide to plasmonic slot waveguide mode converter. Journal of Lightwave Technology, 33(2015), 535–540. [9] Zhu, B. Q., Tsang, H. K. 2016. High coupling efficiency silicon waveguide to metal–insulator–metal waveguide mode converter. Journal of Lightwave Technology, 34(2016), 2467–2472. [10] Ono, M., Taniyama, H., Xu, H., Tsunekawa, M., Kuramochi, E., Nozaki, K., Notomi, M. 2016. Deep-subwavelength plasmonic mode converter with large size reduction for Si-wire waveguide. Optica, 3(2016), 999– 1005. [11] Veronis, G., Fan, S. H. 2007. Modes of subwavelength plasmonic slot waveguides. Journal of Lightwave Technology, 25(2007), 2511–2521. [12] Kewes, G., Schoengen, M., Neitzke, O., Lombardi, P., Schonfeld, R.- S., Mazzamuto, G., Schell, A. W., Probst, J., Wolters, J., Lochel, B., Toninelli, C., Benson, O. 2016. A realistic fabrication and design concept for quantum gates based on single emitters integrated in plasmonic- dielectric waveguide structures. Scientific Reports, 6(2016), 28877. [13] Palik, E. D. 1985. Handbook of optical constants of solids. Vol. 1. Academic Press, London, 804s. [14] Kitamura, R., Pilon, L., Jonasz, M. 2007. Optical constants of silica glass from extreme ultraviolet to far infrared at near room temperature. Applied Optics, 46(2007), 8118–8133. [15] McPeak, K. M., Jayanti, S. V., Kress, S. J. P., Meyer, S., Iotti, S., Rossinelli, A., Norris, D. J. 2015. Plasmonic films can easily be better: Rules and recipes. ACS Photonics, 2(2015), 326–333. [16] Yang, J., Hugonin, J.-P., Lalanne, P. 2016. Near-to-far field transforma- tions for radiative and guided waves. ACS Photonics, 3(2016), 395–402. [17] Ji, X., Barbosa, F. A. S., Roberts, S. P., Dutt, A., Cardenas, J., Okawachi, Y., Bryant, A., Gaeta, A. L., Lipson, M. 2017. Ultra-low-loss on-chip resonators with sub-milliwatt parametric oscillation threshold. Optica, 4(2017), 619–624. [18] Veronis G., Fan, S. 2007. Theoretical investigation of compact cou- plers between dielectric slab waveguides and two-dimensional metal- dielectric-metal plasmonic waveguides. Optics Express, 15(2007), 1211– 1221. Fig. 7. Power transmission factor as a function of Au thickness. IV. DISCUSSION AND CONCLUSION We investigated optimal structures for coupling the mode of a Si wire waveguide to the mode of a plasmonic slot waveguide. We concentrated on the taper geometry and came up with designs that have over 90% power transfer efficiency (approaching 95% in some instances) for Au thicknesses ranging from 30–250 nm. The results that we quote can find applications in compact plasmonic modulator designs with minimal insertion loss and low bit error rates. Although we focused on Si wire waveguides in this study, the techniques that we present can easily be applied to Si nitride waveguides that have been shown to be highly advan- tageous for non-linear applications [17]. Lastly, although our focus has been on taper structures, resonant stub-like elements are another route for designing mode converters as has been demonstrated in 2D structures [18]. REFERENCES [1] Yamada, K. 2011 Silicon Photonic Wire Waveguides: Fundamentals and Applications. ss 1–29. Lockwood, D. J., Pavesi L., ed. 2011. Silicon Photonics II: Components and Integration, Springer, Berlin, 253s. [2] Thomson, D., Zilkie, A., Bowers, J. E., Komljenovic, T., Reed, G. T., Vivien, L., Marris-Morini, D., Cassan, E., Virot, L., F´ed´eli, J.-M., Hartmann, J.-M., Schmid, J. H., Xu, D.-X., Boeuf, F., O'Brien, P., Mashanovich, G. Z., Nedeljkovic, M. 2016. Roadmap on silicon pho- tonics. Journal of Optics, 18(2016), 073003. [3] Melikyan, A., Alloatti, L., Muslija, A., Hillerkuss, D., Schindler, P.C., Li, J., Palmer, R., Korn, D., Muehlbrandt, S., Thourhout, D. V., Chen, B., Dinu, R., Sommer, M., Koos, C., Kohl, M., Freude, W., Leuthold J. 2014. High-speed plasmonic phase modulators. Nature Photonics, 8(2014), 229–233. [4] Haffner, C., Heni, W., Fedoryshyn, Y., Josten, A., Baeuerle, B., Hoess- bacher, C., Salamin, Y., Koch, U., Dordevic, N., Mousel, P., Bonjour, R., Emboras, A., Hillerkuss, D., Leuchtmann, P., Elder, D. L., Dalton, L. R., Hafner, C., Leuthold, J. 2016. Plasmonic organic hybrid modulators- scaling highest speed photonics to the microscale. Proceedings of the IEEE, 104(2016), 2362–2379. [5] Hoessbacher, C., Josten, A., Baeuerle, B., Fedoryshyn, Y., Hettrich, H., Salamin, Y., Heni, W., Haffner, C., Kaiser, C., Schmid, R., Elder, D. L., Hillerkuss, D., Moller, M., Dalton, L. R., Leuthold, J. 2017. Plasmonic modulator with ¿170 GHz bandwidth demonstrated at 100 GBd NRZ. Optics Express, 25(2017), 1762–1768. [6] Tian, J., Yu, S., Yan, W., Qiu, M. 2009. Broadband high-efficiency surface-plasmon-polariton coupler with silicon-metal interface. Applied Physics Letters, 95(2009), 013504. [7] Han, Z., Elezzabi, A. Y., Van, V. 2010. Experimental realization of subwavelength plasmonic slot waveguides on a silicon platform. Optics Letters, 35(2010), 502–504. 30801301802500.750.80.850.90.95hAu(nm)Transmissionz=0nmz=1100nm
1907.07545	2	1907	2019-12-23T17:19:09	Design and analysis of high electron mobility transistor (HEMT) inspired III-V electro-optic modulator topologies	[ "physics.app-ph", "physics.optics" ]	III-V heterostructure based high electron mobility transistors (HEMTs) offer superior performance as compared to CMOS silicon transistors owing to the high mobility in the 2D electron gas (2DEG) channel at the heterostructure interface. Gallium nitride (GaN) based HEMTs are also suitable for high power and high temperature applications. GaN has a rich offering of material properties spanning domains of nonlinear optics, piezoelectric micro-electro-mechanical systems (MEMS), and monolithic microwave integrated circuits (MMICs). In this paper, we propose HEMT inspired III-V electro-optic modulator topologies that could potentially outperform silicon photonic modulators. We analyze the electroabsorption and electrorefraction on account of the 2DEG interaction with light and present a design framework to selectively leverage the desired mechanism of modulation. Our analysis suggests that modulation index of electrorefractive modulation in a HEMT-like structure is comparable to silicon photonic modulators, albeit with much higher electron mobility and thereby much higher modulation rates.	physics.app-ph	physics	Design and analysis of high electron mobility transistor (HEMT) inspired III-V electro-optic modulator topologies PALLABI DAS,1 TIAN-LI WU,2 AND SIDDHARTH TALLUR1,* 1Department of Electrical Engineering, Indian Institute of Technology (IIT) Bombay, Mumbai 400076, India 2International College of Semiconductor Technology, National Chiao Tung University, 30010 Hsinchu, Taiwan *[email protected] Abstract: III-V heterostructure based high electron mobility transistors (HEMTs) offer superior performance as compared to CMOS silicon transistors owing to the high mobility in the 2D electron gas (2DEG) channel at the heterostructure interface. Gallium nitride (GaN) based HEMTs are also suitable for high power and high temperature applications. GaN has a rich offering of material properties spanning domains of nonlinear optics, piezoelectric micro-electro- mechanical systems (MEMS), and monolithic microwave integrated circuits (MMICs). In this paper, we propose HEMT inspired III-V electro-optic modulator topologies that could potentially outperform silicon photonic modulators. We analyze the electroabsorption and electrorefraction on account of the 2DEG interaction with light and present a design framework to selectively leverage the desired mechanism of modulation. Our analysis suggests that modulation index of electrorefractive modulation in a HEMT-like structure is comparable to silicon photonic modulators, albeit with much higher electron mobility and thereby much higher modulation rates. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement Introduction 1. In the last few decades, fiber optic communication has become the technology of choice for high speed communication links with capacities of several hundreds of gigabits per second (Gbps) and low signal losses over link lengths of thousands of kilometers. While fiber optic communication forms a large part of the backbone for communication systems, data rates achievable in short- reach communication networks remain a critical bottleneck, limited by interconnects. Parasitic capacitance in electrical traces limits maximum data rates in conventional electrical interconnects to < 10Gbps [1, 2]. Silicon photonics has drawn immense interest in recent years owing to electronic-photonic integration and CMOS-compatible processes and signal levels that it offers for optical information transfer and signal processing [2 -- 5]. Encoding information generated by on-chip CMOS integrated circuits onto an optical carrier signal requires an electro-optic (EO) modulator. Although silicon is not an electro-optically active material, it can used as an EO modulator by exploiting carrier injection or plasma dispersion effect, when used in p-i-n (p-type/intrinsic/n-type layers) device configuration [2,6,7]. However the carrier recombination lifetime of free charge carriers limits the operating speeds to ≈ 50 − 60Gbps in such devices [8], and higher bandwidth applications typically employ wavelength division multiplexing [9]. Alternative modulation mechanisms utilizing other materials compatible with silicon technology e.g. germanium, have been demonstrated to achieve more efficient modulation [10]. Si-Ge multi-quantum well (MQW) electro-absorption modulators exploiting the quantum confined stark effect (QCSE) and Franz-Keldysh effect have also been reported [2, 11, 12]. QCSE has also been explored in III-V intersubband opto electronic devices [13,14], despite the processing challenge it imposes in terms of incorporating the multi-quantum well structure into the device layer stack. Another alternative technology platform gaining momentum in recent years is lithium niobate (LiNbO3) [15,16]. Unlike silicon, LiNbO3 has strong EO coefficient and lithium niobate electro-optic modulators have recently been demonstrated operating at frequencies exceeding 100Gbps [17] albeit with inadequate bit error rates for practical applications. While most commercial optical modulator products also employ lithium niobate, the material compatibility with other platforms fundamentally limits the scope of on-chip interconnect applications of this technology. Another limitation for high speed on-chip communication arises due to gain-bandwidth performance limits of transistors in any CMOS technology node. For any transistor technology node, the RF performance depends on the cut-off frequency ( fT) and the maximum operating frequency of the device ( fmax). Compared to the highest speed silicon CMOS transistors, gallium nitride (GaN) based High Electron Mobility Transistors (HEMTs) are emerging as a commercially viable technology for high-speed and high-power applications [18] owing to the rich material properties in GaN, such as large band gap, large breakdown field strength, high saturation velocity and high electron mobility [13, 19 -- 21]. A large driver of the performance superiority in GaN HEMTs is the electrically tunable, highly localized sheet of charge (two dimensional electron gas: 2DEG) formed at the III-V heterostructure interface due to spontaneous and piezoelectric polarization in the constituent materials [13]. In this paper, we propose and analyze electro-optic modulator design topologies inspired by III-V HEMTs in a GaN technology platform. Such structures display two types of electro-optic interactions mediated by the 2DEG, namely electroabsorption (due to quantum confinement of electrons in the quantum well formed at the III-V heterostructure interface), and electrorefraction (due to plasma dispersion effect of the 2DEG). We present a variety of design topologies to selectively harness the desired mechanism and compare their performance parameters. Our analysis suggests that electrorefraction is the dominant modulation mechanism due to 2DEG. Coupled with the large electron mobility in HEMT-like devices, this could be leveraged to realize high speed electrorefractive optical modulators. We present the simulation framework used in this work, and the results and observations below. 2. Simulation setup Silicon EO modulators typically operate based on electrorefraction (ER) in p-i-n Si slot waveguide structures, where the electric field driven free carrier concentration can be modulated in the intrinsic region by electric field introduced by applying suitable potential difference across the p and n regions. Modulating the free carrier concentration results in modulation of the refractive index of the material due to plasma dispersion effect [2]. Free carrier electroabsorption and electrorefraction in bulk GaN has been studied by Soltani et al. [22], and demonstrated in bulk GaN EO modulators by several groups [23]. Electroabsorption modulator topologies based on intersubband (ISB) transition of electrons in multi-quantum well (MQW) heterostructures have been explored by several groups [24,25]. However, electroabsorption and electrorefraction due to 2DEG in III-V heterostructures, and their suitability for designing EO modulators have not previously been studied. To simulate electroabsorption due to intersubband transition, we need to solve self-consistent 1-D Schrödinger-Poisson equation to compute the sub-band energies in the potential well formed at the heterostructure interface. This is achieved through TCAD (Technology computer-aided design) simulations performed using Silvaco ATLAS interactive tool. A simulation model of polarization charge analysis for various barrier thickness and Al concentration values in an AlGaN/GaN HEMT device has already been implemented as a built-in example in ATLAS [26]. We adapt this example to our device design to compute the 2DEG carrier concentration density for various values of gate voltage applied to the HEMT. The single mode condition for the waveguide structure is essential for any integrated opto-electronic device to sustain the fundamental mode of propagation for strong optical confinement and efficient coupling with external optical fibers without intermode interference. A hybrid numerical-analytical design methodology for obtaining single mode condition in ridge waveguides with geometries larger than the wavelength of guided light has been presented in our previous work [27]. We design a ridge waveguide structure following our design methodology and perform finite difference time domain (FDTD) simulations in OptiFDTD 32-bit tool to validate the design. The optimized geometry is then simulated in Silvaco ATLAS to quantify the degree of electroabsorption and electrorefraction in the structure. 3. Results and Discussion 3.1. Electrorefraction in free-space modulator topology The device manifestation that we consider is conceptually similar to a AlGaN/GaN HEMT where free-space light beam interacts with the 2DEG formed at the heterostructure. The light beam is incident normal to the wafer, reflects from the bottom surface of the wafer and thereby interacts with the 2DEG upon incidence as well as post reflection. The interaction can be enhanced by embedding such a device in a Fabry-Perot cavity as shown in Figure 1(a). We analyze the effect of plasma dispersion effect due to the 2DEG carrier density modulated via external gate voltage applied to a depleted Schottky diode formed by the gate electrode and channel [28,29]. Intersubband (ISB) absorption due to TM polarized light in a quantum well is forbidden by selection rules at normal incidence [30]. Computation of the absorbtion spectra due to ISB transition in a triangular potential well formed at the AlGaN/GaN interface for light incident at an angle to the normal is presented in the next section. Since there is no electric field applied across the GaN film, the linear electro-optic effect and quadratic electro-optic effect will not contribute to optical modulation in such a structure. Therefore electrorefration due to plasma dispersion effect of the 2DEG will by the sole contributor to electro-optic modulation of the free space light beam. The plasma dispersion effect results in modulation of the effective refractive index (and thereby, the phase of the carrier light beam) at the AlGaN/GaN interface due to modulation of the 2DEG carrier concentration, by varying the applied gate voltage. Applying a sufficiently large reverse bias voltage to the Schottky contact on the gate electrode completely depletes the 2DEG (OFF state of modulator), and a sufficiently large gate voltage can be applied to introduce the 2DEG (ON state of modulator). The phase of the light beam can be modulated by an amount (∆φ) expressed as ∆φ = λ (2t × ∆n2DEG), where λ is the wavelength of light, t is the thickness of the 2DEG, and 2π ∆n2DEG is the change in refractive index due to the 2DEG plasma dispersion effect. The expression accounts for the interaction of light with 2DEG both upon incidence and reflection, as mentioned in the previous paragraph. For an efficient modulator, a phase shift of π radians (OFF state) should be achievable using low switching voltages. To enhance the phase shift, the device could be embedded in a photonic resonator or a high finesse Fabry-Perot cavity. The Drude plasma model captures the dependence of complex refractive index (n + ik) of a material on the free carrier concentration in the semiconductor, where n and k denote the real and imaginary part of the refractive index respectively. The analytical expressions for the plasma dispersion effect (electrorefraction) and free-carrier absorption (electroabsorption) are given by [22] ∆n = −e2λ2 respectively, where, 8π2c2εo ns ∆N and ∆P are the free-carrier densities, µe and µh are the mobilities and m∗ ce and m∗ ch are the conductivity effective masses for electrons and holes respectively, e is the electron charge, c is the speed of light, o is the permittivity of vacuum, and ns is the real part of refractive index of the undoped semiconductor. HEMTs typically employ undoped films, and the contribution to refractive index modulation from free-carriers in the AlGaN and GaN films is therefore negligible. The refractive index modulation is thereby dominated by modulation of the 2DEG charge density on account of the gate voltage. As previously stated, the structure topology and ∆k = + ∆P µh m∗ 16π3c3εo ns ∆N m∗ c e + ∆P m∗ c h e3λ3 ∆N µe m∗ c e (cid:19) c h (cid:18) (cid:19) (cid:18) Fig. 1. (a) Light interaction with 2DEG in AlGaN/GaN HEMTs can be realized in a free-space manifestation (left) embedded in a Fabry Perot cavity or a multipass waveguide manifestation (right). (b) Illustration of the simulation process flow followed in this work to determine the modulation index for electrorefractive modulation due to 2DEG. (c) Variation of 2DEG charge density (N2DEG) and slope of (N2DEG) − Vg curve with applied gate bias (Vg). (d) The variation of refractive index with applied gate voltage is well correlated to the variation in sheet charge density of the 2DEG. (cid:16) ∂N2D E G (cid:17) ∂Vg forbids electro-absorption. The modified Drude model accounting solely for electrorefraction i.e. ∆n2DEG due to change in 2DEG charge density (∆N2DEG) is given by equation (1): (cid:18) ∆N2DEG (cid:19) ∆n2DEG = −e2λ2 8π2c2εonGaN m∗ ce (1) Here nGaN denotes the refractive index of the GaN film. Figure 1(b) schematically illustrates the computational analysis flow. To validate the expression in equation (1), we first simulate the variation of N2DEG for various gate bias voltages (varied from −8V to +2V) in Silvaco ATLAS. The source and drain region contacts are connected to ground and the DC gate voltage (Vg) is swept in a range covering pinch-off (2DEG fully depleted) to enhancement (maximum 2DEG concentration). The analytical expression for N2DEG dependence on gate voltage (Vg) has been extensively studied in literature [31 -- 34] (for details see supplementary information in appendix A), and these models are used to adjust the polarization charge factor in the simulation to reproduce the 2DEG sheet charge density (≈ 1014cm−2). Self-consistent solutions of the one-dimensional (1D) Schrödinger-Poisson equations are solved to evaluate the wavefunctions of the first two energy sub-bands (Figure 2). The contribution of the fundamental energy level dominates the contribution of the second energy level to N2DEG, and therefore only the fundamental energy level contribution is considered for subsequent analysis. We use two approaches to evaluate the dependence of ∆n2DEG on Vg. The first approach utilizes a hybrid TCAD-analytical computation, wherein we obtain the relation of N2DEG to Vg from TCAD simulations, which is then used to calculate the change in refractive index using equation (1). It is important to note that the 2DEG concentration is typically specified as a surface charge density. However, the notation used in equation (1) denotes the volume charge density of ∂Vg c,2DEG the 2DEG. The TCAD simulations provide 2DEG concentration in terms of volume charge density at user-specified depth from the interface. The 2DEG surface charge concentration (in cm−2) is obtained by integrating the volume charge density (in units of cm−3) over the thickness of the GaN film. To evaluate the hybrid-TCAD analytical model we substitute the electron concentration (in units of cm−3) at a depth of 1nm from the interface in equation (1). Figure 1(c) shows the 2DEG charge density variation with gate voltage (Vg) for typical HEMT dimensions with 25nm thick Al0.3Ga0.7N on 300nm thick GaN substrate obtained with TCAD simulations performed in Silvaco ATLAS, and the derivative of the curve. The derivative ( ∂N2D E G ) of the curve signifies the sensitivity and is computed in Origin Lab. Considering the peak-peak amplitude of the modulating gate voltage to be Vdrive = 1Vp−p and operating wavelength λ = 1.55µm, we obtain the change in volume charge density of 2DEG (∆N2DEG) ≈ 3.5 × 1018cm−3 from TCAD simulations, and the corresponding change in refractive index as computed using the modified Drude model presented in equation (1) is ∆n2DEG ≈ −7.39 × 10−3. The values of other = 0.22 × me [35], where me is the electron parameters are chosen as nGaN = 2.31 and m∗ rest mass ≈ 9.1 × 10−31kg. This results in a phase shift ∆φ2DEG ≈ −5.99 × 10−5 radians. As mentioned earlier, the phase shift can be enhanced by embedding the device in a high finesse Fabry-Perot cavity to increase the number of round-trips of light through the 2DEG interface. The second approach we follow relies on utilizing the built-in Drude model in Silvaco ATLAS (details of implementation of the model are presented in supplementary information in appendix B). We record the effective refractive index at the interface for various values of Vg. The real part of the effective refractive index as a function of applied gate voltage is plotted in Figure 1(d). The pinch-off voltage (Vo f f ) for complete depletion of the 2DEG is obtained from capacitance-voltage (C-V) profile simulation in Silvaco ATLAS. We observe that the knee point for the real part of the effective refractive is the same as Vo f f , thus cross-validating the simulations. We also observe that the 2DEG sheet charge reduces to a negligible value below Vo f f (−6V in this example). The maximum slope of the curve (modulation efficiency) is noticed for Vg in range of −6V to −1V. The change in effective refractive index as obtained from the TCAD simulation is ≈ −3.0 × 10−3, which is comparable to the solution obtained from the hybrid TCAD-analytical approach. The magnitude of the change in refractive index is also comparable to conventional silicon plasma dispersion modulators [2], and thus the proposed architecture holds great promise for ultra-fast modulation of free-space light in a practical device realization. The limitation of the AlGaN/GaN HEMT free space modulator is the limited interaction length of light with 2DEG, which will result in diminished modulation depth. This can be improved in a waveguided implementation of this structure, where the heterostructure interface can be designed to overlap with the core of the waveguide that has the largest optical mode intensity. The interaction can be additionally enhanced in an AlGaN/GaN/AlGaN double quantum well structure. However such a structure requires significantly larger reverse bias voltages to deplete the 2DEG charge carriers in the well and may not be practical. Moreover the ISB absorption emerges as the dominant modulation mechanism in this structure. The detailed analysis of ISB absorption in a multi quantum well (MQW) based AlGaN/GaN electroabsorption modulator is described in the next subsection. 3.2. Electroabsorption in guided wave modulator topology Incorporating the heterostructure into a waveguide enables enhanced overlapping of the optical mode with the 2DEG charge carriers, thereby boosting the modulation index in such a device. This can be achieved by incorporating a number of quantum wells in the waveguide core. A variety of waveguide based optical devices based on ISB transitions in III-V quantum wells (QWs) have been reported [25,36,37]. ISB absorption occurs only if the electric field is oriented normal to the plane of the layers comprising the quantum wells. Therefore ISB absorption due to normal incidence of TM polarized light is prohibited. Alternatively, a multi-pass ridge waveguide with wedged facets can leverage electrorefraction as well as ISB in the triangular potential well Fig. 2. TCAD simulation results for (a) eigen-energy levels and (b) wavefunctions for the 2DEG electrons in the triangular potential well formed at the AlGaN/GaN interface. The two lowest subbands lie below the Fermi level in the triangular potential well. (c) The absorption spectrum (at 0V gate voltage) obtained from TCAD simulation shows that the electroabsorption due to intersubband transitions in the triangular potential well is maximum at a wavelength of 800nm and negligible at 1.55µm wavelength. at the interface of a single AlGaN/GaN HEMT structure as shown in Figure 1(a). We perform TCAD simulations to obtain the electroabsorption spectra for Al0.3Ga0.7N/GaN HEMT for TM polarized light incident at an angle of 30o to the normal. The device geometry we consider here is an Al0.3Ga0.7N/GaN/GaN structure, where a 10nm thick undoped GaN region is defined as the quantum well, sandwiched between undoped 25nm thick Al0.3Ga0.7N and 300nm thick GaN layers. The energy band diagram obtained through TCAD simulation shows presence of two subbands below the Fermi level in the triangular potential well (Figure 2(a)). The corresponding wavefunctions are shown in Figure 2(b). The absorption spectrum obtained from the simulation (Figure 2(c)) shows that the optical absorption due to ISB transition in the triangular potential well is maximum at near infrared (NIR) wavelengths in vicinity of 800nm and negligible for 1.5µm wavelength light. To increase the electroabsorption at C-band and L-band wavelengths for telecommunication, we investigate the performance of multi-quantum well (MQW) structures. Electroabsorption in MQW structures can also be exploited in a waveguide structure, with large overlap of the optical mode with the MQWs. Here we analyze electroabsorption due to ISB transitions in an Al0.5Ga0.5N/GaN MQW structure. The device structure consists of five periods of GaN wells and AlGaN barriers, with a thickness of 1.6nm each, sandwiched between 500nm thick AlGaN cladding, and terminated by 300nm thick GaN layer as shown in Figure 3(a). We perform FDTD simulations to identify optimal geometry for a single-mode AlGaN/GaN MQW ridge waveguide. Sapphire is used as substrate in the simulation in FDTD. The width and height of the AlGaN cladding layer in the ridge geometry are 1µm and 317.6nm respectively. The position of the MQWs is chosen to overlap with the region with largest light intensity in the optical mode. Fig. 3. (a) Indicative geometry of an AlGaN/GaN ridge waveguide with embedded AlGaN/GaN multi quantum wells (MQWs) designed for single mode condition. The optical mode profile simulated using OptiFDTD 32 bit software shows maximal overlap with the MQWs. (b) Energy band diagram for 5-period Al0.5Ga0.5N/GaN MQWs obtained from TCAD simulation. (c) The absorption spectra obtained from TCAD simulations for the 5-period Al0.5Ga0.5N/GaN MQWs for various values of gate bias voltage. The absorption peak is a strong function of the gate bias voltage and hence the MQW makes it possible to realize an electroabsorption modulator for 1.55µm wavelength light. (d) The absorption coefficient of Al0.5Ga0.5N/GaN MQW structure is two orders of magnitude lower than that of a conventional AlN/GaN MQW, thus indicating that the latter is a more suitable choice for electroabsorption modulator as compared to the former. (cid:20) −2 2mi (cid:21) δz2 + V(z) ⊥ δ2 ψi(z) = Eiψi(z), where mi To analyze the electroabsorption in the AlGaN/GaN MQW ridge waveguide, we perform TCAD simulation using Silvaco ATLAS for the geometry obtained above. The 1-D Schrödinger equation for an electron residing in any of the discrete energy states in a quantum well is given ⊥ is the effective mass of the electrons in the by ith subband, and Ei and ψi are the corresponding energy level and wavefunction. FermiâĂŹs golden rule is used to calculate the probability of a transition from electron from ith state to f th state in Silvaco ATLAS. The multiband spatially dependent Kronig-Penney (K-P) model is used for self consistency. The electron effective mass (m∗) for both AlGaN and GaN is assumed to be m∗ = 0.22me. The wells are n-doped, with dopant density 1019cm−3. We study the lowest two energy states in the TCAD simulation. Piezoelectric and spontaneous polarization of the nitride semiconductor is taken into consideration which causes the band bending at the interfaces, resulting in the energy band structure shown in Figure 3(b). The gate contact is placed atop the AlGaN barrier and bottom contact is made to the bulk GaN substrate. The simulation is performed for TM polarized light incident at an angle of 30◦ to the normal. The ISB gain is computed as a function of photon energy in the MQWs in TCAD [38]. Negative value of the gain is indicative of electroabsorption in the quantum wells. The absorption spectra obtained from TCAD simulations for the 5-period Al0.5Ga0.5N/GaN quantum well structure is shown in Figure 3(c). We observe an absorption peak at 1.558µm that arises due to the ISB transition of electrons from the lower subband to the upper subband in the conduction band of the bottom-most GaN QW. Upon application of a gate bias voltage at the top AlGaN layer with respect to the GaN bulk substrate, we observe a shift in the wavelength as well as magnitude of the absorption peak. Insertion of capping layer can impact the infrared absorption in MQW structures , and has been analyzed in detail by Beeler et al. in an AlN/GaN MQW structure [39]. We have compared the absorption coefficient of Al0.5Ga0.5N/GaN MQW structure with that of a conventional AlN/GaN MQW with well and barrier thickness of 1.9nm thickness as shown in Figure 3(d). We observe that the absorption coefficient for AlN/GaN is larger than the Al0.5Ga0.5N/GaN MQW by several orders of magnitude for the same number of quantum wells. Therefore AlN/GaN MQWs are more suitable for electroabsorption based modulator architecture, while the HEMT-like structure studied in this paper is better suited for electrorefraction based modulator presented in the previous subsection. 4. Conclusion In summary, we have presented and thoroughly analyzed two electro-optic modulator topologies in an AlGaN/GaN platform, leveraging the plasma dispersion effect of the highly confined 2DEG at the heterostructure interface. A modified Drude model is presented for quantifying the plasma dispersion effect of the 2DEG and supplemented with Silvaco ATLAS based TCAD simulations. FDTD and TCAD co-simulations are presented to solve for optimal device performance. The device primarily analyzed here is a free-space electrorefractive modulator with a structure identical to an AlGaN/GaN HEMT. We have analyzed electroabsorption in the AlGaN/GaN heterostructure as well as in III-V MQW structure and observe negligible electroabsorption in an AlGaN/GaN platform as compared to conventional AlN/GaN MQWs of comparable geometry for 1.55µm wavelength light. The electrorefraction effect is found to be dominant over electroabsorption at 1.55µm wavelength in a free-space optical modulator with TM polarized light incident normal to the plane of the device. Our future work will focus on fabrication and characterization of the HEMT electrorefractive modulator presented here, including studying performance parameters that cannot be adequately captured through simulations, such as modulation depth and modulation rate limit in such devices. Acknowledgments The authors thank Prof. Swaroop Ganguly and Prof. Dipankar Saha at IIT Bombay for their valuable inputs on TCAD simulations. The authors also thank Microelectronics Computation Lab (MCL) at IIT Bombay for providing access to Silvaco ATLAS. Disclosures The authors declare no competing interests. A. Analytical model and TCAD simulation of 2DEG concentration in HEMTs To derive an expression for the change in refractive index (∆n2DEG) as a function of applied gate voltage (Vg), we first simulate the variation of the 2DEG sheet charge density (N2DEG) for different bias conditions in Silvaco ATLAS. The source and drain region contacts are connected to ground and the DC gate voltage is swept in a range covering pinch off to enhancement. The N2DEG variation with Vg for different molar concentration of Al (x in AlxGa1−xN) is obtained from the simulation, as shown in Figure 4(a). For obtaining higher 2DEG concentration, we have selected 30% Al molar concentration in AlxGa1−xN for simulation in all subsequent discussions in this paper, unless otherwise mentioned. The analytical expression for N2DEG as function of Vg has been extensively studied in literature [31 -- 34,40]. These analytical models are only valid for the linear region of operation in a HEMT. Our proposed modulator relies on both depletion as well as linear regions of operation, and hence these models are not adequate. Hence we obtain the expression by solving Poisson's equation [40] for our device. Assuming the AlGaN layer is completely ionized and solving Poisson's equation for metal/i-AlGaN/i-GaN layer, we obtain: N2DEG = Cd q [(Vg − Vo f f ) − Vc − E f ] (2) where Cd is the effective capacitance per unit area between the gate electrode and 2DEG, Vc is the voltage at drain/source contact and E f is the Fermi level. The term Vo f f denotes the pinch-off voltage, below which the 2DEG is completely depleted. The complexity in modeling the 2DEG charge density arises due to the fact that the Fermi level E f in equation (2) is also a function of N2DEG and can be expressed as a transcendental equation [33]: (cid:20) (cid:18) E f − E0 (cid:19) kT/q (cid:21) (cid:20) (cid:18) E f − E1 (cid:19) kT/q (cid:21) + 1 N2DEG = DVtln exp + 1 + ln exp (3) where D is the density of states, k is the Boltzmann constant, T is the temperature, q is the charge on an electron, and E0 and E1 are energies of the two lowest subbands in the triangular potential well. In contrast to the charge control model [40], a precise analytical expression for N2DEG as a function of Vg has been provided by Dasgupta et al. [34] for AlGaAs/GaAs HEMTs. We have used similar methodology to derive the analytical model of N2DEG vs Vg for AlGaN/GaN HEMT structure. We represent E f as a function of N2DEG as a polynomial expressed in equation (4) [34] and then employ a curve fitting technique to obtain the relation of N2DEG and Vg. Substituting equation (4) in equation (2), we obtain: N2DEG = where (cid:48) 3 = k3 + k qdAlGaN E f = k1 + k2N1/2 2DEG + k3N2DEG (cid:26)[−k2 + [k2 2 + 4k3 (cid:48)(Vg − Vo f f − k1)]1/2] 2k3 (cid:48) (cid:27)2 (4) (5) (6) (cid:48) (cid:48) Here dAlGaN is the thickness and is the permitivitty of the AlGaN layer. The values for 3 that appear in equation (5) are obtained from curve fitting: k1 = −0.10101V , k1, k2, and k k2 = 4 × 10−9V − cm , k3 = 3 × 10−15V − cm2. Figure 4(b) confirms that the solution obtained from simulation and the derived analytical model show good agreement with each other. The pinch off voltage (Vo f f ) observed for 30% Al molar concentration in AlxGa1−xN from the simulation (shown in Figure 4(a)) is in agreement with experimental data reported by Cordier et al. [41]. It is to be noted that the simulation parameters are obtained for an ideal AlGaN/GaN HEMT structure with no interface traps, dislocations or any other defects. The polarization charge can be calibrated in the TCAD simulation in Silvaco ATLAS using polar.scale variable. The value of Vo f f is corroborated through capacitance-voltage (C-V) profile simulation of the heterostructure. We consider a depletion mode i-AlGaN/i-GaN HEMT structure consisting of a metal Schottky contact, and Ohmic drain and source contact. Consider the case where the Schottky contact is biased at the edge of depletion region (Vo f f ). The thin depleted AlGaN layer between the top Schottky and the bottom 2DEG sheet charge can be considered as a parallel plate capacitance (Cj), for gate voltage Vg > Vo f f (as shown in Figure 4(c)). This capacitance can be expressed as Cj = r o A , where o and r are the permittivity of free space and relative d Al G a N permittivity of the AlGaN layer respectively, A is the area under the gate electrode and dAlGaN is thickness of the AlGaN layer. Upon application of a reverse bias voltage, the 2DEG is depleted C j +C2D E G and for gate voltage Vg < Vo f f the heterostructure presents an additional depletion capacitance (C2DEG). The net capacitance thus reduces at pinch-off. The total capacitance (Ctot) in the pinch off regime is given by [2,28] Ctot = C j×C2D E G . The modulator is turned off by reducing the gate voltage below Vo f f , and is turned on by increasing the gate voltage sufficiently above Vo f f to have large 2DEG concentration in the channel. The TCAD simulation of C-V profile is setup by applying an ac voltage of frequency 1MHz in Silvaco ATLAS, and the parameters used for the simulation are listed in Table 1. The simulations are carried out for AlGaN/GaN HEMT with Al molar concentration of 30% and a barrier thickness of 25nm. Figure 4(d) shows the simulated C-V profile of the device. The value of Vo f f for the device parameters specified in Table 1 is obtained from simulation to be −6V, which can be feasibly achieved in a practical implementation. Geometry parameters used for electrical simulation Parameters Material Thickness Length Gate electrode Schottky metal 500nm 4µm Barrier layer Al0.3Ga0.7N 25nm 10µm Channel layer GaN 1.475 µm 10 µm Table 1. Parameters used for TCAD simulation in Silvaco ATLAS to obtain C-V profile of the heterostructure and N2DEG variation with Vg. B. Electro-optic TCAD simulation setup in Silvaco ATLAS TCAD simulation for Si optical modulator in a p-i-n waveguide configuration is provided as a built-in example in ATLAS [42]. The example model implements free carrier plasma dispersion effect using ABS.FCARRIER model. The model also simulates the change in effective refractive index with applied bias voltage using WCGD.REFR parameter and WAVEGUIDE statement to define geometry and physical models for a stand-alone optical mode simulation. The parameters of the free carrier Plasma dispersion model can be modified for different materials using FC.AN, FC.AP, FC.RN, FC.RP etc. parameters in the MATERIAL section. For detailed explanation of the terms and glossary, readers are encouraged to read the Atlas User's Manual [38]. Drude plasma model based on the free carrier plasma dispersion effect has also been reported in literature for simulation of an optical modulator with a MOS junction using Silvaco Atlas [43], where free carrier elecrto-refraction (∆n) and electro-absorption (∆k) models have been implemented using ABS.DRUDE parameter. We adapt this model for our device simulation. Various material properties such as electron effective mass, mobility and refractive index for different materials can be overridden manually by using DRUDE.ME, DRUDE.MUE, REAL.INDEX and IMAG.INDEX respectively in the MATERIAL statement. The parameters and their values used for this simulation are listed in Table 2. We use POLARIZATION and CALC.STRAIN in the MODEL statement to specify the piezoelectric and spontaneous polarization respectively. POLAR.SCALE is the polarization scale factor (with a default value = 1), whose value is set to 1.1 in order to obtain polarization charge of the order of ≈ 1014cm−2. For simulating the complex refractive index, we set the parameter INDEX.MODEL = 1, and numerical analysis is performed using 'Newton TRAP' method. We use the PROBE statement to record the effective refractive index at the interface for each value of the gate bias voltage. Fig. 4. (a) TCAD simulation showing 2DEG sheet charge concentration for various Al molar concentration (x) in an AlxGa1−xN/GaN heterostructure. (b) Variation of N2DEG with gate bias voltage (Vg) showing comparison of simulation result (solid line) and analytical model (dotted line). Both curves show good agreement with each other. (c) (i) Junction capacitance formed by i-AlGaN region sandwiched between the Schottky metal gate and the 2DEG sheet charge; (ii) Upon application of a larger reverse bias voltage, the 2DEG is pinched off by the extension of the depletion region in to the i-GaN region, in turn producing a 2DEG capacitance that appears in series with the junction capacitance.(d) Capacitance-voltage profile of the AlGaN/GaN heterostructure obtained from TCAD simulation performed in Silvaco ATLAS. References 1. D. A. B. Miller, "Optical interconnects to electronic chips," Appl. Opt. 49, F59 -- F70 (2010). 2. G. T. Reed, G. Mashanovich, F. Y. Gardes, and D. J. Thomson, "Silicon optical modulators," Nat. Photonics 4, 518 (2010). 3. A. E.-J. Lim, J. Song, Q. Fang, C. Li, X. Tu, N. Duan, K. K. Chen, R. P.-C. Tern, and T.-Y. Liow, "Review of silicon photonics foundry efforts," IEEE J. Sel. Top. Quantum Electron. 20, 405 -- 416 (2013). 4. A. Liu, R. Jones, L. Liao, D. Samara-Rubio, D. Rubin, O. Cohen, R. Nicolaescu, and M. Paniccia, "A high-speed silicon optical modulator based on a metal -- oxide -- semiconductor capacitor," Nature 427, 615 (2004). 5. M. J. Heck, J. F. Bauters, M. L. Davenport, D. T. Spencer, and J. E. Bowers, "Ultra-low loss waveguide platform and its integration with silicon photonics," Laser & Photonics Rev. 8, 667 -- 686 (2014). 6. F. Gardes, G. Mashanovich, and G. Reed, "Evolution of optical modulation in silicon-on-insulator devices," SPIE Newsroom, Dec 27 (2007). 7. L. Zhou and A. W. Poon, "Silicon electro-optic modulators using pin diodes embedded 10-micron-diameter microdisk resonators," Opt. express 14, 6851 -- 6857 (2006). 8. L. Liao, A. Liu, D. Rubin, J. Basak, Y. Chetrit, H. Nguyen, R. Cohen, N. Izhaky, and M. Paniccia, "40 Gbit/s silicon optical modulator for high-speed applications," Electron. letters 43, 1196 -- 1197 (2007). 9. S. Dwivedi, P. De Heyn, P. Absil, J. Van Campenhout, and W. Bogaerts, "Coarse wavelength division multiplexer on silicon-on-insulator for 100 GbE," in 2015 IEEE 12th International Conference on Group IV Photonics (GFP), (IEEE, 2015), pp. 9 -- 10. 10. P. Chaisakul, D. Marris-Morini, J. Frigerio, D. Chrastina, M.-S. Rouifed, S. Cecchi, P. Crozat, G. Isella, and L. Vivien, "Integrated germanium optical interconnects on silicon substrates," Nat. Photonics 8, 482 (2014). 11. Y.-H. Kuo, Y. K. Lee, Y. Ge, S. Ren, J. E. Roth, T. I. Kamins, D. A. Miller, and J. S. Harris, "Strong quantum-confined Stark effect in germanium quantum-well structures on silicon," Nature 437, 1334 (2005). Description Electron effective mass Electron mobility Parameter DRUDE.ME DRUDE.MUE Value 0.22 1000m2/(V − s) Refractive index of GaN Refractive of AlxGa1−xN Polarization charge index REAL.INDEX REAL.INDEX polar.scale 2.31 2.15 1.1 = 0.22 × me Remarks m∗ c,2DEG The value of the conduction band electron mobility is overridden by 2DEG mobility in Drude model For wavelength λ = 1.55µm For wavelength λ = 1.55µm and x = 0.3 To modify the sign and magni- tude of polarization charge (de- fault = 1). Numerical method Newton TRAP Complex Eigen value solver index.model Maxtrap = 20 Maxtrap specifies the number of times the trap (bias step) will be repeated in case of divergence (default = 4) Specifies whether the simple (IN- refractive index model DEX.MODEL = 0) or the complex refractive index (IN- DEX.MODEL = 1) is used 1 Table 2. Parameters used for optical TCAD simulation in Silvaco ATLAS. A detailed glossary is available in the user manual [38]. 12. Y. Rong, Y. Ge, Y. Huo, M. Fiorentino, M. R. Tan, T. I. Kamins, T. J. Ochalski, G. Huyet, and J. S. Harris Jr, "Quantum-confined Stark effect in Ge/SiGe quantum wells on Si," IEEE J. selected topics quantum electronics 16, 85 -- 92 (2009). 13. D. J. Piprek, ed., Nitride Semiconductor Devices: Principles and Simulation (John Wiley & Sons, Ltd, 2007). 14. H. Machhadani, P. Kandaswamy, S. Sakr, A. Vardi, A. Wirtmüller, L. Nevou, F. Guillot, G. Pozzovivo, M. Tchernycheva, A. Lupu et al., "GaN/AlGaN intersubband optoelectronic devices," New J. Phys. 11, 125023 (2009). 15. M. He, M. Xu, Y. Ren, J. Jian, Z. Ruan, Y. Xu, S. Gao, S. Sun, X. Wen, L. Zhou, L. Liu, C. Guo, H. Chen, S. Yu, L. Liu, and X. Cai, "High-performance hybrid silicon and lithium niobate machâĂŞzehnder modulators for 100 gbps and beyond," Nat. Photonics 13, 1 (2019). 16. L. Chen, Q. Xu, M. G. Wood, and R. M. Reano, "Hybrid silicon and lithium niobate electro-optical ring modulator," 17. C. Wang, M. Zhang, B. Stern, M. Lipson, and M. Lončar, "Nanophotonic lithium niobate electro-optic modulators," Optica 1, 112 -- 118 (2014). Opt. Express 26, 1547 -- 1555 (2018). Device Lett. 31, 195 -- 197 (2010). 18. J. W. Chung, W. E. Hoke, E. M. Chumbes, and T. Palacios, "AlGaN/GaN HEMT With 300-GHz fmax," IEEE Electron 19. M. Shur, "GaN based transistors for high power applications," Solid-State Electron. 42, 2131 -- 2138 (1998). 20. X. Huang, Z. Liu, Q. Li, and F. C. Lee, "Evaluation and application of 600 v gan hemt in cascode structure," IEEE Transactions on Power Electron. 29, 2453 -- 2461 (2014). 21. C. Yang, X. Luo, T. Sun, A. Zhang, D. Ouyang, S. Deng, J. Wei, and B. Zhang, "High breakdown voltage and low dynamic on-resistance algan/gan hemt with fluorine ion implantation in sin x passivation layer," Nanoscale research letters 14, 191 (2019). 22. M. Soltani and R. Soref, "Free-carrier electrorefraction and electroabsorption in wurtzite GaN," Opt. Express 23, 24984 -- 24990 (2015). 23. C.-K. Kao, A. Bhattacharyya, C. Thomidis, R. Paiella, and T. D. Moustakas, "Electroabsorption modulators based on bulk gan films and gan/algan multiple quantum wells," J. Appl. Phys. 109, 083102 (2011). 24. J. Heber, C. Gmachl, H. Ng, and A. Cho, "Comparative study of ultrafast intersubband electron scattering times at 1.55 µ m wavelength in GaN/AlGaN heterostructures," Appl. physics letters 81, 1237 -- 1239 (2002). 25. H. Machhadani, P. Kandaswamy, S. Sakr, A. Vardi, A. Wirtmüller, L. Nevou, F. Guillot, G. Pozzovivo, M. Tchernycheva, A. Lupu et al., "GaN/AlGaN intersubband optoelectronic devices," New J. Phys. 11, 125023 (2009). 26. "Polarization Charge Analysis: TCAD example," URLhttps://www.silvaco.com/examples/tcad/section20/example18 27. P. Roy, P. Das, and S. Tallur, "Hybrid Numerical-Analytical Effective Index Method for Designing Large Geometry Ridge Waveguides," in 2018 IEEE Photonics Conference (IPC), (IEEE, 2018), pp. 1 -- 2. 28. A. Ansari and M. Rais-Zadeh, "A Thickness-Mode AlGaN/GaN Resonant Body High Electron Mobility Transistor," IEEE Transactions on Electron Devices 61, 1006 -- 1013 (2014). 29. A. Ansari and M. Rais-Zadeh, "Depletion-mediated piezoelectric AlGaN/GaN resonators," Phys. Status Solidi (A) 213, 3007 -- 3013 (2016). /index.html. 30. R. Yang, J. Xu, and M. Sweeny, "Selection rules of intersubband transitions in conduction-band quantum wells," Phys. review. B, Condens. matter 50, 7474 -- 7482 (1994). 31. O. Ambacher, J. Smart, J. Shealy, N. Weimann, K. Chu, M. Murphy, W. Schaff, L. Eastman, R. Dimitrov, L. Wittmer et al., "Two-dimensional electron gases induced by spontaneous and piezoelectric polarization charges in N-and Ga-face AlGaN/GaN heterostructures," J. applied physics 85, 3222 -- 3233 (1999). 32. J. Zhang, B. Syamal, X. Zhou, S. Arulkumaran, and G. I. Ng, "A compact model for generic MIS-HEMTs based on the unified 2DEG density expression," IEEE Transactions on Electron Devices 61, 314 -- 323 (2014). 33. S. Khandelwal, N. Goyal, and T. A. Fjeldly, "A physics-based analytical model for 2DEG charge density in AlGaN/GaN HEMT devices," IEEE Transactions on Electron Devices 58, 3622 -- 3625 (2011). 34. N. DasGupta and A. DasGupta, "An analytical expression for sheet carrier concentration vs gate voltage for HEMT modelling," Solid-state electronics 36, 201 -- 203 (1993). 35. A. Kurakin, S. Vitusevich, S. Danylyuk, H. Hardtdegen, N. Klein, Z. Bougrioua, A. Naumov, and A. Belyaev, "Quantum confinement effect on the effective mass in two-dimensional electron gas of AlGaN/GaN heterostructures," J. applied physics 105, 073703 (2009). 36. E. Baumann, F. R. Giorgetta, D. Hofstetter, S. Leconte, F. Guillot, E. Bellet-Amalric, and E. Monroy, "Electrically adjustable intersubband absorption of a GaN/AlN superlattice grown on a transistorlike structure," Appl. physics letters 89, 101121 (2006). 37. D. Hofstetter, L. Diehl, J. Faist, W. J. Schaff, J. Hwang, L. F. Eastman, and C. Zellweger, "Midinfrared intersubband absorption on AlGaN/GaN-based high-electron-mobility transistors," Appl. physics letters 80, 2991 -- 2993 (2002). 38. "Atlas Users Manual: TCAD publications," . 39. M. Beeler, E. Trichas, and E. Monroy, "III-nitride semiconductors for intersubband optoelectronics:a review," Semicond. Sci. Technol. 28, 074022 (2013). 40. S. Khandelwal, N. Goyal, and T. A. Fjeldly, "A physics-based analytical model for 2DEG charge density in AlGaN/GaN HEMT devices," IEEE Transactions on Electron Devices 58, 3622 -- 3625 (2011). 41. Y. Cordier, J.-C. Moreno, N. Baron, E. Frayssinet, S. Chenot, B. Damilano, and F. Semond, "Demonstration of algan/gan high-electron-mobility transistors grown by molecular beam epitaxy on si (110)," IEEE Electron Device Lett. 29, 1187 -- 1189 (2008). 42. "Silicon p-i-n Waveguide Modulator: https://www.silvaco.com/examples/tcad/section26/example12/index.html. 43. "Simulation of Si Optical Modulator with a MOS Junction: https://www.silvaco.com/tech_lib_TCAD/simulationstandard/2015/jan_feb_mar/a1/a1.html. TCAD example," URL- TCAD publications," URL-
1912.13506	1	1912	2019-12-20T12:47:47	Integrated Sensor System to Control the Temperature Effects and the Hysteresis on Adaptive Fluid-Membrane Piezoelectric Lenses	[ "physics.app-ph" ]	We present in this paper an integrated sensor system for closed-loop control of the temperature effects and the hysteresis on the refractive power of an adaptive fluid-membrane piezoelectric lens. The piezoelectric hysteresis and the fluid thermal expansion contribute to a nonlinear response of the lens refractive power defined as a function of the actuation electric field. Hence, a pressure sensor and a temperature sensor are integrated inside the lens to monitor and define the lens refractive power as a function of both the internal fluid pressure and the temperature, thus, allowing for the closed-loop control of the refractive power. The adaptive lens has a refractive power range varying from -16 m-1 to +17 m-1 at 25{\deg}C and from -15 m-1 to +28 m 1 at 75{\deg}C.	physics.app-ph	physics	Integrated Sensor System to Control the Temperature Effects and the Hysteresis on Adaptive Fluid-Membrane Piezoelectric Lenses Hitesh G. B. Gowda and Ulrike Wallrabe Laboratory for Microactuators, Department for Microsystems Engineering - IMTEK, University [email protected]; [email protected] of Freiburg, Germany. ABSTRACT We present in this paper an integrated sensor system for closed-loop control of the temperature effects and the hysteresis on the refractive power of an adaptive fluid-membrane piezoelectric lens. The piezoelectric hysteresis and the fluid thermal expansion contribute to a nonlinear response of the lens refractive power defined as a function of the actuation electric field. Hence, a pressure sensor and a temperature sensor are integrated inside the lens to monitor and define the lens refractive power as a function of both the internal fluid pressure and the temperature, thus, allowing for the closed-loop control of the refractive power. The adaptive lens has a refractive power range varying from -16 m-1 to +17 m-1 at 25°C and from -15 m-1 to +28 m-1 at 75°C. Keywords: Adaptive lens; Fluid-Membrane lens; Hysteresis compensation; Temperature characterization; Integrated sensors; Piezoelectric devices; 1. INTRODUCTION The adaptive optics technology was initially used in astronomical applications to modify the phase of the incoming light by using a deformable mirror [1]. Eventually, the adaptive optics technology was used in microscopes for correcting aberrations [2], optical communication systems for coupling light [3], and in optical imaging systems for varying focal power [4]. In the adaptive optics technology, the surface of the optical component is deformed by a microactuator to modify the characteristics of the incident light. The microactuator contributes to a smaller size, lower power consumption, and a faster response system [5]. An imaging system implemented with an adaptive optics lens eliminates the conventional mechanical movement of lenses to focus an image and contributes to the faster response and longer lifespan of the system. The focal power of the adaptive lens, which is dependent on the surface curvature, is varied by deforming the flexible surface of the lens with the use of a microactuator. One such adaptive lens utilizing a piezoelectric microactuator to deform a flexible fluid-membrane interface was developed in the Laboratory for Microactuators, Department of Microsystems Engineering- IMTEK, University of Freiburg, Germany. The developed adaptive lens working principle relies on the interaction between the integrated annular piezoelectric actuator and the fluid chamber bounded by a flexible membrane [6]. The piezoelectric actuator deforms the fluid chamber and varies the internal fluid pressure. The varied fluid pressure acts on the flexible membrane and deforms it into an aspherical surface with varied surface curvature. Hence, a change in the internal fluid pressure results in a change in the refractive power of the lens. The refractive power defined as a function of the actuation electric field at the piezo exhibits a non-linear hysteresis. Hence, Draheim et al. integrated a relative pressure sensor inside the lens to define and control the refractive power as a function of the internal fluid pressure [7]. However, at higher temperatures, the thermal expansion of the fluid affects the fluid pressure as well and results in a deviation of the defined refractive power. In this paper, we present a further integration of a temperature sensor inside the lens to account for the temperature effects and to define the refractive power as a function of both the internal fluid pressure and the temperature. Besides, we have adapted our process to mount the lens chamber directly on a PCB substrate. 2. DESIGN The adaptive lens presented by Draheim et al. [6] consists of a piezoelectric actuator, a fluid chamber, and a flexible transparent PDMS membrane to form an active lens chamber. The piezoelectric actuator is manufactured by cutting the piezoelectric ceramics into annular discs using a UV laser and gluing them using a two-component epoxy in antiparallel polarization configuration. Further, the transparent membrane and the fluid chamber are molded onto the actuator using micromolding techniques. In our new PCB based design, the sensors and the electrical connectors are integrated on the PCB, which is also equipped with a central glass window. The PCB is commercially fabricated and is modified using a CNC machine to form slots for the lens chamber, the glass window, and the pressure sensor, as shown in Figure 1 (a) and (b). A PTC temperature sensor is soldered directly on the PCB, and a membrane-based relative pressure sensor is glued inside the previously machined slot and wire bonded for electrical connections (Figure 1 (b)). The separately fabricated lens chamber is glued with PDMS on the PCB and primed with an optical oil. The adaptive lens has a clear aperture of 10 mm at an outer diameter of 20 mm, as shown in Figure 1 (d). Figure 1 (a) Integrated PCB with sensors and connectors, (b) enlarged view to show temperature and pressure sensor, (c) adaptive lens mounted on the PCB substrate, and (d) 2D cross-sectional view of the adaptive lens showing the integrated temperature and pressure sensor inside the fluid chamber. In our bimorph actuator, the piezoceramics are configured with anti-parallel polarization. Under the application of an electric field on the piezoelectric bimorph, one piezo expands and the other contracts resulting in a bending deflection. The actuator bending deflection will deform the fluid chamber and vary the internal fluid pressure. The varied fluid pressure acts on the flexible PDMS membrane and deforms it in the opposite direction to that of the actuator deflection. Due to the circular design of the adaptive lens, the membrane deforms into an aspherical surface. By varying the magnitude and direction of the applied electric field on the piezoelectric actuator, a plano-convex lens (Figure 2 (a)), or a plano-concave lens (Figure 2 (b)) with a variable refractive power can be achieved. Figure 2 Adaptive lens under applied electric field at the piezo to form (a) a plano-convex lens or (b) a plano-concave lens. 3. CHARACTERIZATION The adaptive lens was characterized for the refractive power at different actuation levels and higher temperatures. For the characterization of the adaptive lens at higher temperatures, the adaptive lens (Figure 3 (b)) was mounted on a resistive heater (Figure 3 (a)) and heated to required temperatures. The membrane surface profile was measured using a profilometer and a confocal displacement sensor, providing a resolution of 110 nm (Figure 3 (c)). A sinusoidal electric field of 1.3 kV/mm at 1 Hz was applied to the piezo actuator through an electrical driver. The electrical driver was used to divide the electric field for the two piezoceramics in the bimorph and also to limit the negative electric field to 1/3 of the piezoelectric coercive field to avoid depolarization [8]. The refractive power was calculated from the radius of curvature of the measured membrane profile. The characterization was repeated with the adaptive lens heated to higher temperatures. Figure 3 The characterization setup showing (a) heating element, (b) adaptive lens mounted on the heater, and (c) confocal sensor to measure membrane surface. Figure 4 (a) shows the refractive power as a function of the applied electric fields at different temperatures. A hysteresis, as well as a temperature drift, are obvious. During the characterization, the outputs from the pressure sensor and the temperature sensor were measured synchronously with the membrane deformation. The measured sensor outputs and membrane deformation were used to define the refractive power as a function of both the internal fluid pressure and the temperature, as shown in Figure 4 (b). The integrated sensors were used to compensate for the hysteresis and the temperature drift and control the refractive power linearly. The refractive power ranges from -16 m-1 to +17 m-1 at 25°C and from -15 m-1 to +28 m-1 at 75°C. For the refractive power range at 25°C, the fluid pressure varies from - 270 Pa to + 270 Pa. Figure 4 Temperature effect on the refractive power of the lens defined (a) as a function of the actuated voltage showing hysteresis response and (b) as a function of the internal fluid pressure. The drift in refractive power seems stronger for positive than for negative electric fields; this can be explained by two effects: Firstly, the thermal expansion of the fluid results in an increased volume, which produces a positive membrane pre-deflection and hence a positive refractive power. Secondly, the increase in piezoelectric d31 coefficient at higher temperatures [8] increases the net actuator deflection, which corresponds to higher negative and higher positive membrane deflections, thus higher positive and negative refractive powers. The superposed two effects result in a net higher positive refractive power. 4. SIMULATION The asymmetric temperature drift of the refractive power observed in the experiment was underlined by a 2D axisymmetric model of the adaptive lens (Figure 5 (a)) simulated in COMSOL Multiphysics® [9]. The simulation model was used to analyze the combined effects of the fluid thermal expansion and the increased piezoelectric deflection at higher temperatures. Figure 5 (b) compares the simulation and experimental results that show a similar temperature drift of the refractive power. At low voltage, i.e., negative pressure, the temperature effect diminishes. Figure 5 (a) The 2D axisymmetric model of the adaptive lens simulated in COMSOL Multiphysics® with the surface plot showing the deformation of the membrane and change in internal fluid pressure and (b) comparison of experiment and simulation results showing similar refractive power drift at a higher temperature. 5. CLOSED-LOOP CONTROL The integrated sensors were used in a closed-loop arrangement (Figure 6 (a)), to define the refractive power as a function of the fluid pressure at different temperatures, shown in the continuous line plots in fig 5 (b). In contrast, with the knowledge of fluid pressure and temperature, any required refractive power in the range was able to be achieved, shown with the discrete dots in Figure 6 (b), which as well proves the repeatability and precise controllability of the lens. Figure 6 (a) Block diagram of the closed-loop system to control the refractive power and (b) continuous and discrete control of the refractive power defined as a function of fluid pressure and temperature. 6. SUMMARY The adaptive lens with the integrated piezoelectric actuator and sensors allows for closed-loop control of the refractive power with a peak to peak range from 33 m-1 at 25°C to 43 m-1 at 75°C. The integrated pressure and temperature sensor address the non-linear hysteresis and thermal expansion effects to control the refractive power linearly. The increase in refractive power range at higher temperatures is due to the increased actuator deflection and the thermal expansion of the fluid. Acknowledgment This work was financed by the Baden-Württemberg Stiftung gGmbH under the project VISIR2 (Variable Intelligent Sensors, Integrated and Robust for Visible and IR light). REFERENCES [1] [2] [3] [4] [5] J. M. Beckers, "Adaptive Optics for Astronomy: Principles, Performance, and Applications," Annu. Rev. Astron. Astrophys., vol. 31, no. 1, pp. 13 -- 62, Sep. 1993. M. J. Booth, "Adaptive optics in microscopy," Philos. Trans. R. Soc. A Math. Phys. Eng. Sci., vol. 365, no. 1861, pp. 2829 -- 2843, Dec. 2007. T. Weyrauch, M. A. Vorontsov, J. Gowens, and T. G. Bifano, "Fiber coupling with adaptive optics for free-space optical communication," 2002, pp. 177 -- 184. F. Schneider, J. Draheim, R. Kamberger, P. Waibel, and U. Wallrabe, "Optical characterization of adaptive fluidic silicone-membrane lenses," Opt. Express, vol. 17, no. 14, p. 11813, Jul. 2009. E. Thielicke and E. Obermeier, "Microactuators and their technologies," Mechatronics, vol. 10, no. 4, pp. 431 -- 455, Jun. 2000. [6] [7] [8] [9] J. Draheim, F. Schneider, R. Kamberger, C. Mueller, and U. Wallrabe, "Fabrication of a fluidic membrane lens system," J. Micromechanics Microengineering, vol. 19, no. 9, p. 095013, Sep. 2009. J. Draheim, T. Burger, R. Kamberger, and U. Wallrabe, "Closed-loop pressure control of an adaptive single chamber membrane lens with integrated actuation," in 16th International Conference on Optical MEMS and Nanophotonics, 2011, pp. 47 -- 48. B. P. Bruno, A. R. Fahmy, M. Stürmer, U. Wallrabe, and M. C. Wapler, "Properties of piezoceramic materials in high electric field actuator applications," Smart Mater. Struct., vol. 28, no. 1, p. 015029, Jan. 2019. H. G. B. Gowda and U. Wallrabe, "Simulation of an Adaptive Fluid-Membrane Piezoelectric Lens," Preprints, Oct. 2019.
1706.05262	3	1706	2017-11-24T09:53:24	Low-Energy Truly Random Number Generation with Superparamagnetic Tunnel Junctions for Unconventional Computing	[ "physics.app-ph", "cond-mat.mes-hall" ]	Low-energy random number generation is critical for many emerging computing schemes proposed to complement or replace von Neumann architectures. However, current random number generators are always associated with an energy cost that is prohibitive for these computing schemes. In this paper, we introduce random number bit generation based on specific nanodevices: superparamagnetic tunnel junctions. We experimentally demonstrate high quality random bit generation that represents orders-of-magnitude improvements in energy efficiency compared to current solutions. We show that the random generation speed improves with nanodevice scaling, and investigate the impact of temperature, magnetic field and crosstalk. Finally, we show how alternative computing schemes can be implemented using superparamagentic tunnel junctions as random number generators. These results open the way for fabricating efficient hardware computing devices leveraging stochasticity, and highlight a novel use for emerging nanodevices.	physics.app-ph	physics	Low-Energy Truly Random Number Generation with Superparamagnetic Tunnel Junctions for Unconventional Computing D. Vodenicarevic,1 N. Locatelli,1 A. Mizrahi,1, 2 J. S. Friedman,3 A. F. Vincent,1 M. Romera,2 A. Fukushima,4 K. Yakushiji,4 H. Kubota,4 S. Yuasa,4 S. Tiwari,5 J. Grollier,2 and D. Querlioz1, ∗ 1Centre for Nanoscience and Nanotechnology, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, C2N Orsay, 91405 Orsay, France 2Unit´e Mixte de Physique CNRS, Thales, Univ. Paris-Sud, Universit´e Paris-Saclay, 91767 Palaiseau, France 3University of Texas at Dallas, 800 West Campbell Road Richardson, TX 75080, USA 4AIST Tsukuba, 1-1-1 Higashi, Tsukuba, Ibaraki 305-8561, Japan 5School of ECE, Cornell University, Ithaca, New York 14850, USA (Dated: November 27, 2017) Low-energy random number generation is critical for many emerging computing schemes proposed to com- plement or replace von Neumann architectures. However, current random number generators are always asso- ciated with an energy cost that is prohibitive for these computing schemes. In this paper, we introduce random number bit generation based on specific nanodevices: superparamagnetic tunnel junctions. We experimentally demonstrate high quality random bit generation that represents orders-of-magnitude improvements in energy efficiency compared to current solutions. We show that the random generation speed improves with nanode- vice scaling, and investigate the impact of temperature, magnetic field and crosstalk. Finally, we show how alternative computing schemes can be implemented using superparamagentic tunnel junctions as random num- ber generators. These results open the way for fabricating efficient hardware computing devices leveraging stochasticity, and highlight a novel use for emerging nanodevices. I. INTRODUCTION With conventional transistor technology reaching its scala- bility limits1, significant effort is involved in the investigation of alternative computing schemes for microelectronics. Many of these emerging ideas, such as stochastic computing2–6 and some brain-inspired (or neuromorphic) schemes7–9, require a large quantity of random numbers. However, the circuit area and the energy required to generate these random numbers are major limitations of such computing schemes. For example, in the popular neuromorphic TrueNorth system7, one third of the neuron area is dedicated to perform random number gen- eration. Indeed, one million random bits are required, at each integration step of the system. More concerning, in stochastic computing architectures, random number generation is typi- cally the dominant source of energy consumption, as the logic performed using the random bits is generally quite simple and efficient by principle. Many practical stochastic computing schemes therefore try to limit the reliance on expensive inde- pendent random bits using various techniques, including the sharing or reuse of random bits10–12. However, such tricks limit the capabilities of stochastic computing to small tasks, as they introduce correlations between signals. Most of the aforementioned unconventional computing cir- cuits use pseudo-random number generators. But these ei- ther lead to low quality random numbers or are highly en- ergy and area-consuming. A preferable solution would be to rely on "true" random number generators that generate ran- dom bits based on physical phenomena that are intrinsically random. However, this is also difficult to realize with min- imal energy consumption. This difficulty is due to the fact that most true random number generators function by trigger- ing events whose outcome is intrinsically random. Triggering these events comes with a non-negligible energy cost. The most energy-efficient example uses a bistable CMOS circuit forced into in a meta-stable state which then randomly falls into one of the two stable states, generating one random bit13. It consumes 3pJ/bit and a circuit area of 4000µm2. In order to reduce this large area footprint, recent proposals suggest to leverage the inherent stochastic programming prop- erties that arise in many of the bi-stable nano-devices devel- oped for memory applications14. This approach was investi- gated with oxide-based resistive memory devices15–18, phase- change memory devices19,20, magnetic memory devices21–23, as well as with straintronic memory devices24. However, these approaches are based on repeated, energy-intensive program- ming operations, and still require high energy for random bit generation. For instance, it requires dozens of pJ/bit to in- duce a stochastic switch of magnetization in magnetic tunnel junctions with two stable states, as proposed in the "Spin- Dice" concept, due to the high energy barrier between the magnetic states. Optimized schemes have been proposed25–27, predicting further reduction of the energy cost per bit, but are still bounded by the need of a costly perturb operation. While proposing high quality random number with high throughput, such strategies are no fit for emerging neuro-inspired comput- ing applications like stochastic computing architectures. A more natural approach would be to extract random num- bers directly from thermal noise, as it provides randomness at no energy cost. Unfortunately, this approach requires large circuits to amplify thermal noise into a large signal of random bits, and has never been shown to be more energy efficient than the first approach until now. The lowest energy solution today is to use jitter as a way to efficiently amplify the noise present in CMOS ring oscillators. The most energy efficient implementation28 requires 23pJ/bit and 375µm2. In the present work, we propose the use of a nanomagnetic device that intrinsically amplifies thermal noise without exter- nal energy supply: superparamagnetic tunnel junctions. These 2 The devices we measured were fabricated by sputtering, with a standard magnetic tunnel junction process, with the CMOS- compatible stack detailed in Fig. 1(b). E-beam lithography patterning was then performed to produce 50× 150nm2 ellip- tic pillars. The free magnet has two stable states, parallel (P) and antiparallel (AP) relatively to the pinned layer (Fig. 1(c)). Through the tunnel magneto-resistance effect30, the electri- cal resistance of the junction in the AP state RAP is higher than the resistance in the P state RP . This effect is tradi- tionally measured through the TMR coefficient defined by RAP /RP = 1 + TMR. The lateral dimensions of the device are chosen so that the effective energy barrier between the two stable states is not very high compared to kBT . Unlike the case of MRAMs, for which the magnetization direction of the free magnet is highly stable and can only be switched by proper external action, the magnetization direction of the superparamagnetic free magnet spontaneously switches between its two stable states, due to low stability relative to thermal fluctuations (Fig. 1(c))31,32. Here, no bias or perturb scheme is required to provoke these random fluctuations, but only temperature. Resistance versus time measurements were done on junc- tions by applying a small 10µA constant current through the junction. Such a small current amplitude was chosen to have negligible influence on the magnetic behavior of the device33 and to maximize its lifetime while providing a clear signal. Fig. 1(d) shows a sample from the time evolution of the elec- trical resistance of a junction measured at room temperature, as well as a binarized version, obtained by thresholding. We see that the resistance follows two-state fluctuations analo- gous to a random telegraph signal. The mean frequency of fluctuations is strongly related to the shape and material prop- erties of the junction34. Fig. 2(a) shows the histograms of the dwell times in the '1' (AP) and '0' (P) states, obtained through measurement of a su- perparamagnetic tunnel junction over a 10 second period. We see that these histograms can be fitted by an exponential law, which is characteristic of a Poisson process. Fig. 2(b) presents the power spectrum density of the same signal, superimposed with the expected power spectrum density of a random tele- graph signal based on a Poisson process. Excellent agreement between the measured results and the hypothesis of a Poisson process is seen. Random bits can be extracted by sampling the voltage across the device at a constant frequency. The voltage was initially sampled at 100kHz, and bitstreams with slower sam- pling rates were obtained by subsampling the initial bitstream. To evaluate the quality of the obtained random bits, the device was measured for over 2.5 days, producing 21.2 gigabits. No external magnetic field was applied during the measurement. Figure 1. Structure and behavior of superparamagnetic tunnel junctions. (a) Basic structure of the measured superparamagnetic tunnel junctions and readout setup. (b) Detailed stack of the junc- (c) Representation of the two stable magnetic states, and tions. the associated energy barrier. (d) Experimental resistance trace and thresholding operation. bi-stable magnetic tunnel junctions are reminiscent of the ones used for Magnetic Random Access Memories (MRAMs)29. However, contrarily to MRAMs cells, the energy barrier be- tween the two magnetic states is very low, and thermal fluc- tuations induce repeated and stochastic magnetization switch- ing between the two states at room temperature. Therefore, no write operations are required and a low-energy readout of the device state naturally produces random bits. We show that these devices permit the generation of high quality random numbers at 20fJ/bit using less than 2µm2, which is orders of magnitude more efficient in terms of energy and area than current solutions. We first show experimentally that superparamagnetic tun- nel junctions allow the generation of high-quality random bits with minimal readout circuitry and that their behavior can be predicted by existing physical models. We then use the model to investigate the influence of device scaling and environmen- tal factors on random bit quality and speed. Circuit simulation enables an estimation of the energy efficiency of random bit generation. Finally, we demonstrate the potential of these de- vices for unconventional computing through the example task of email messages classification using random bits extracted from the experimental data, and show that they are particularly adapted to computing schemes trading off speed for ultra low energy consumption. II. EXPLOITING THE STOCHASTIC BEHAVIOR OF SUPERPARAMAGNETIC TUNNEL JUNCTIONS Superparamagnetic tunnel junctions are bistable spintronic nanodevices composed of a high stability pinned nanomagnet and a low-stability "free" nanomagnet, separated by a tunnel oxide layer (Fig. 1(a)). Their structure is highly similar to the magnetic tunnel junctions used as the basic cells of MRAMs. III. OPTIMIZING THE QUALITY OF RANDOM BITS The sampling frequency needs to be chosen carefully rel- ative to the mean switching frequency of the junction, de- fined as FMTJ = 1/(τ1 + τ0), where τ1 and τ0 are the mean 3 Figure 2. Statistics of the experimental superparamagnetic tunnel junction signal. (a) Experimental histograms of the dwell times in Anti-parallel (AP, top Figure, high resistance) and Parallel (P, bottom Figure, low resistance) states, for a superparamagnetic magnetic tunnel junction measured during 10s. (b) Experimental power spectrum density of the resistance signal. (c) Autocorrelation of the experimental resistance signal as a function of signal sampling period. dwell times in state 1 and 0, respectively. FMTJ was mea- sured to be 1.66kHz (τ1 + τ0 ≈ 604µs). Fig. 2(c) presents the correlation of consecutive bits extracted at different sam- pling rates. This result is superimposed with the one the- oretically expected from a Poisson process. At high sam- pling frequency, subsequent bits are naturally autocorrelated (at Fsampling = 100kHz, correlation reaches 92.8%), and can therefore not be used for applications. This correlation de- creases exponentially with the sampling period, which can therefore be chosen based on the correlation requirements on the random numbers. As observed, in Fig. 2(a), the AP and P states possess an asymmetric stability: the device spends more time on aver- age in the P state than in the AP state, which corresponds to a mean state (mean of the binarized signal) of 60.5%. This asymmetry can be connected to the stray field induced by the pinned magnet layer structure, which is present in all mag- netic tunnel junctions35. This biasing field offsets the junction mean state from the ideal 50% value required for most appli- cations, and is subject to device-to-device variations. In order to eliminate this bias and any residual bit correla- tion, a "whitening" of the random bits is therefore required. To achieve this operation, we make use of a standard technique: combining several bitstreams into a single one using XOR gates. It can be shown (mathematical derivation available in supplementary information S10) that the auto-correlation after XOR whitening is the product of the individual auto- correlations of the combined signals. It therefore decreases exponentially with the number of combined MTJ bitstreams, and is always lower than the auto-correlation of any of the combined signals. In the same way, the mean state of the whitened bitstream gets exponentially closer to 50% with the number of XOR-combined bitstreams, and stays always closer to perfect balance than any of the bitstreams being combined. As a reference, a more advanced but heavy stateful whitening technique ("Blum"36) was also applied to the raw measure- ments. As an illustration, we consider bits extracted at a frequency of 5kHz. The bitstream was then divided into chunks of equal length which were used as independent signals and XOR- combined bit per bit for the XOR whitening process. We plot in Fig. 3(a) the consecutive bit correlation and the mean state of the whitened bitstream as functions of the number of signals combined by XOR. The correlation and the mean value bias decrease with the number of XOR-combined sig- nals. With 4 bitstreams (XOR4), the resulting consecutive bit correlation drops under 1% and the mean value reaches 49.9%. For 8 bitstreams (XOR8), the auto-correlation is be- low 0.06% and the mean state reaches 50% with a standard deviation of 0.5%. These results suggest that XOR whitening can correct correlation and mean value issues. However, in order to fully evaluate the quality of a whitened bitstream, signal autocorrelation and mean state are not suffi- cient metrics. We therefore used the standardized National Institute of Standards and Technology Statistical Test Suite (NIST STS)37, which evaluates the quality of the random bit- stream against 188 tests. The NIST STS computes the statis- tics of bitstreams, such as mean value, auto-correlation, stan- dard deviation, estimated entropy or pattern occurrence fre- quencies, and checks weather they are consistent with perfect randomness. It also looks for the presence of repeated struc- tures, linear dependencies, and other behaviors unexpected in a perfectly random bitstream. To perform the NIST STS tests, the bitstream to be tested, measured during 2.5 days, is divided into 1 Mbits sequences. Each chunk is then tested independently, and the pass rate (percentage of one million bits sequences passing the test) was computed for each of the 188 tests. Fig. 3(b) and (c) show the results for XOR4 and XOR8 whitened bitstreams respectively. For a bitstream to be consistent with cryptographic quality, the pass rates of all tests should lie in the green region37, corre- sponding to the expected minimal pass rate provided by the NIST STS, dependent on the number of tested chunks. We can see that bits extracted by XOR8 whitening pass this re- quirement (this was also the case with the reference Blum technique), while with XOR4 whitening only a fraction of the tests are consistent with cryptographic quality of the random bits38. 4 Figure 3. Whitened experimental random bitstream quality assessment. (a) Mean state and consecutive bit autocorrelation as functions of the number of independent superparamagnetic tunnel junction signals combined by XOR. NIST STS randomness quality test results on experimental data whitened by XOR4 (b) and XOR8 (c) at a Fsampling = 5kHz sampling frequency. When all test results are in the green area, the bitstream is consistent with cryptographic quality. Fsampling Fsampling/FMTJ Raw XOR2 XOR4 XOR8 10.1 100 kHz 12.2 20 kHz 9.1 kHz 88.3 100 5.9 kHz 100 5 kHz 100 1.9 kHz 100 0.9 kHz 0.7 kHz 100 100 0.5 kHz 60.4 12.1 5.5 3.6 3.0 1.1 0.54 0.42 0.30 10.1 0.5 10.6 1.1 1.1 14.4 14.4 16.0 16.0 10.1 10.6 10.6 16.5 72.9 97.9 98.4 97.9 98.4 0 0.5 1.1 1.1 1.1 1.1 1.1 1.1 1.1 Table I. NIST Statistical Test Suite results for the whitened ex- perimental random bitstream. Percentage of NIST STS tests sat- isfying cryptographic quality requirements for different numbers of combined bitstreams, and different sampling frequencies. Table I presents more comprehensive results: the propor- tion of tests in the green area is given for XOR-whitened bitstreams at different sampling frequencies and numbers of XOR-combined bitstreams. The results confirm that the qual- ity of the whitened bitstream increases for lower sampling frequencies (less correlation) and higher numbers of XOR- combined bitstreams (less correlation and lower bias). Higher numbers of XOR-combined bitstreams therefore allow further increasing the sampling rate while still passing all the NIST statistical tests, at the expense of more circuit area and energy consumption. XOR8 at Fsampling/FMTJ = 3.0 appears to be an optimal choice, with 100% of the tests consistent with cryptographic quality and the highest sampling frequency. A more comprehensive analysis of the impact of the number of XORed bitstreams is presented in supplementary information S1. Consistent results (presented in supplementary information S2) were observed on a second sample, measured during 1.5 days, producing 8.96 gigabits. IV. SCALING CAPABILITIES OF THE RANDOM NUMBER GENERATORS IN TERMS OF SPEED AND ENERGY CONSUMPTION Further studying the potential of superparamagnetic tunnel junctions for random number generation requires a realistic model of the device. In the literature, at low electric current, magnetic tunnel junctions switching is usually described by an Arrhenius-N´eel two-states analysis, modeling a thermally activated magnetic switching39. The mean switching rates in each state are then described by: r0→1 = 1/τ0 = f0 exp r1→0 = 1/τ1 = f0 exp (cid:16)− ∆E0→1 (cid:16)− ∆E1→0 kB T kB T (cid:17) (cid:17) (1) where f0 = 1GHz is the magnetic attempt frequency, ∆E0→1 and ∆E1→0 are the energy barriers associated with each tran- sition (see Fig. 1(c)). Our experimental results suggesting that superparamagnetic tunnel junction switching is a Poisson pro- cess are consistent with this model. The superparamagnetic tunnel junctions that we character- ized experimentally in this study are slow devices. They can be used to generate random bits at kHz frequencies, sufficient for real-time brain-inspired systems like7, but not for high per- In our 50 × 150nm superparamag- formance applications. netic tunnel junctions, we identified that the switching occurs through nucleation and propagation of a magnetic domain, probably seeded by fluctuations in a subset of grains within it31 (supplementary information S3). By contrast, recent ex- periments on perpendicular magnetic anisotropy (PMA) mag- netic tunnel junctions have shown that aggressively scaled de- vices (diameters smaller than 35nm) switch at the scale of the whole volume34. Therefore, in the context of random num- ber generators, extreme scaling of the nanodevices appears as providential, as smaller volumes and areas are directly linked to a lower magnetization stability of the free magnet40, in- creasing random bit generation speed exponentially. This is in sharp contrast with MRAMs, where conservation of stabil- ity with extreme scaling is an important challenge41. From the study described in previous part, we observe that a 25% correlation between consecutive bits can be efficiently 5 Figure 4. Sampling rate and readout circuitry. (a) Effect of scaling the energy barrier on the ideal sampling frequency, based on the device model. The inset shows the energy barrier as a function of the junction diameter for PMA-MTJs. (b) Precharge sense amplifier (PCSA) circuit for reading the state of a superparamagnetic tunnel junction. (c) PCSA reading energy as a function of the superparamagnetic tunnel junction P state resistance RP . whitened out by XOR8 and allow generated random numbers to pass all the NIST STS tests. This consideration, together with the model, allows us to evaluate quantitatively the speed of scaled random bit generators based on superparamagnetic tunnel junctions, by evaluating the maximum sampling fre- (cid:46) 25% (details in quency to keep the correlation ρc supplementary information S4): sampling ≈ 3FMTJ = F max f0 exp (cid:18) (cid:19) X,X+1 3 2 − ∆E kBT (2) where ∆E is the energy barrier separating the two states. ∆E = Keff (D)π D2 4 t is derived as a function of the device diameter D, where t = 1.6nm is the free magnet thickness and the effective anisotropy Kef f (D) is derived considering interfacial anisotropy and bulk anisotropies, using experimen- tal values from34. Fig. 4(a), based on this derivation, shows that random bits could be generated at up to tens of MHz for energy barriers below 5kBT , corresponding to a diameter of 8nm. In addition, in a final system, specialized transistor-based electronics needs to be associated to the superparamagnetic tunnel junctions to read their states without interfering with the random bit generation quality. Here, we considered a precharge sense amplifier circuit (PCSA, Fig. 4(b)), a CMOS circuit originally proposed as an MRAM read circuit42. We simulated this circuit using standard integrated circuit design software (Cadence tools) and the transistor models of a 28nm commercial technology. The superparamagnetic tunnel junc- tions were modeled using a compact (VerilogA-based) model implementing the Arrhenius-N´eel model. The results of cir- cuit simulation (Fig. 4(c)) show that the read energy is rel- atively independent from superparamagnetic tunnel junction resistance, and very low (≈ 2fJ/bit). We also evaluated the read disturb effect of the PCSA. Reading the state of a junction can potentially affect random bit generation through the spin torque effect. Based on the spin torque model of Ref. 39, its impact on the mean state is around 10−6 for junctions such as the one we characterized experimentally. It would stay below 0.1% for ultrascaled junctions functioning at high frequencies, as shown in supplementary material S11. This small effect would therefore be corrected by whitening. Evaluating the energy consumption of random bit genera- tion requires taking into account the whitening process. As XOR whitening combines multiple junction states per gener- ated bit, it requires multiple read operations per generated bit. XOR8 reads 8 junctions to generate a bit, and requires 20fJ/bit in average (including the XOR gate operation). In terms of area, in a 28 nm technology, the layout of a full XOR8 ran- dom bit generator takes less than 2µm2. XOR4 whitening would require 9.8fJ/bit and a 1µm2 area. These results show the potential of superparamagnetic tun- nel junctions for state of the art low-energy random number generation. V. SENSITIVITY OF THE RANDOM NUMBER GENERATORS TO PERTURBATIONS Although superparamagnetic tunnel junctions allow ran- dom number generation with minimal energy, their sensitivity to external perturbations must be carefully evaluated. First, as the stochastic switching of superparamagnetic tun- nel junctions is thermally activated, temperature directly af- fects their switching rates. Fig. 5(a), based on the model intro- duced in the previous section, shows the temperature depen- dence of the maximum sampling frequency for several values of the effective barrier. Higher temperatures produce better random numbers: as temperature increases, the superparam- agnetic tunnel junction switching rates increase accordingly, thus allowing faster sampling frequencies. Devices should therefore be sized based on their lowest operation tempera- ture. Superparamagnetic tunnel junctions are also sensitive to magnetic fields. Fig. 5(b) shows the experimental mean state of a superparamagnetic junction as the function of external magnetic field. Fields of a few Oe shift the mean state to a level that cannot be corrected by XOR8 whitening. Magnetic shielding is therefore necessary for applications. Such tech- nology (based on mu metals) has already been developed for MRAM. 6 Figure 5. External perturbations and crosstalk effects. (a) Theoretical curve of the maximum sampling frequency for high quality random bit generation, as a function of temperature, for different junction stabilities (the black curve corresponds to the junction characterized in Figs. 2 and 3). (b) Black symbols: experimental mean state of the junction (up ratio) as a function of the applied magnetic field (red dotted line: theory). (c) Theoretical minimal distance between superparamagnetic tunnel junctions allowed to prevent crosstalk, as a function of the superparamagnetic tunnel junction diameter. Finally, a challenge regarding scalability and integration is that closely packed superparamagnetic tunnel junctions can interact by dipolar interaction, which could lead to corre- lations in random numbers. In the case of perpendicularly magnetized superparamagnetic tunnel junctions, using the previously introduced model, we evaluated that the critical center-to-center distance between two superparamagnetic tun- nel junctions guaranteeing negligible crosstalk43, correspond- ing to less than ρc = 0.1% cross-correlation, is given by (de- tails in supplementary information S5): (cid:18) dc = (cid:19)1/3 µ0(MSV )2 4πkBT tanh −1(ρc) (3) Fig. 5(c) shows the evolution of this critical distance at room temperature as the diameter of the junctions is scaled down. It falls below 100nm for ultimately scaled 10nm diameter devices, which constitutes a layout design rule, and which would be naturally respected if the junctions are associated with PCSA circuits. VI. USING SUPERPARAMAGNETIC TUNNEL JUNCTIONS FOR UNCONVENTIONAL COMPUTING To illustrate the potential of superparamagnetic tunnel junc- tions for unconventional computing, we use the experimental whitened random bitstreams as inputs for a modern stochastic circuit (Fig. 6(a) and Ref. 3) that performs Bayesian infer- ence as a non-Turing machine. As a pedagogical task, we use this circuit to classify email messages as spam or not spam (sample messages are presented in Fig. 6(a)), as recently in- troduced in Ref. 3. The approach uses a dictionary of known words with their associated occurrence rates in spam and non-spam messages. Each word of the dictionary has an associated probabilistic binary generator whose probability of drawing a 1 is set to different values depending on the presence (or absence) of the word in the presented sentence. As our random bit generators provide bitstreams with mean values of 0.5, multiple random bit generators are needed to create a probabilistic binary gen- erator (see "RNG" block in Fig. 6(a), detailed in supplemen- tary information Fig. S6). The outputs of these generators are then combined using C-Elements to perform an approximate Bayesian inference3. The time average of the output gives the probability of the presented message being spam. Fig. 6(b) gives the spam probability inferred using XOR4- whitened bitstreams and shows that the more random bit gen- erators are used per word, the more precisely the probabilistic binary generator can be tuned and the better the prediction is. Also, the longer the output averaging time, the more accurate the answer of the system is. A trade-off to keep a low energy consumption is found for 8 random bit generators / word and averaging over 2000 samples (supplementary information S7 and S8). Because of its reliance on multiple stages of binary bit- stream combination, and fine generator probability tuning, this circuit is sensitive to the quality of the underlying ran- dom number generator. We tested the circuit using raw 5kHz- sampled experimental bitstreams, as well as its XOR4 and XOR8 whitened versions. When the bits are not whitened, the circuit does not perform satisfactorily (Fig. 6(c) and sup- plementary information S7). Using bits whitened with XOR8, the circuit performs as well as the reference Blum whitener, successfully classifying all messages. Furthermore, XOR4, which does not pass all NIST STS tests, also provides perfect classification while requiring less energy. These results highlight the potential of the approach for low-energy applications. Using the results of the previous sec- tion, circuit simulation with 8 random bit generators / word and 2,000 clock cycles shows that a message can be classified using only nJ energy (the exact value depends on the num- ber of words in the dictionary, see supplementary information S9). This simple study shows that superparamagnetic tunnel junctions can be used for efficient random number generation for low-power probabilistic computing. 7 Figure 6. Email classification with stochastic computing using whitened experimental random bitstreams. (a) Stochastic email classi- fication circuit, and email messages to classify. One "RNG" block includes several random bit generators to provide bits with controllable probability. (b) Resulting spam probability as a function of the number of random bits per word using XOR4-whitened experimental 5kHz data over 2000 iterations. (c) Spam classification success rates for different whitening techniques for 5kHz sampling, using 8bits/word and 2000 iterations. VII. CONCLUSION In this work, we have shown that the natural dynamics of superparamagnetic tunnel junctions produces random tele- graph signals that can be read and turned into high quality random bitstreams with minimal energy and circuit overhead, while staying fully compatible with standard CMOS fabrica- tion processes. The whitening process turning these measurements into us- able random bitstreams implies energy and area overhead. But while reference Blum whitening would add important CMOS overhead, XOR adds very little. XOR8 and Blum both provide high random bit quality consistent with cryptographic require- ments, but XOR8 fits better to low energy applications, as it typically requires only 20fJ/bit and 2µm2, orders of magni- tudes less than current state of the art. This efficiency comes at the cost of speed. Scaled superparamagnetic tunnel junc- tions could generate random bits at speeds of dozens of MHz, which is slower than higher energy random bit generators, but sufficient for many unconventional computing schemes in very low power consumption contexts such as the Internet of Things. This efficiency also comes at the cost of a certain sen- sitivity of random bit generation to the environment, making it prone to attacks. Random bit generation based on super- paramagnetic tunnel junctions is therefore much better suited for unconventional computing than for cryptographic applica- tions. The evaluation of the probabilistic email classifier circuit also suggests that in many alternative computing schemes, lower-quality whitening can be used successfully to achieve extreme energy efficiency without degrading performance. At design time, a balance between random number quality, gen- eration speed, and energy consumption can be freely chosen to suit the target application. This is especially important in the context of modern Bayesian inference systems44,45, but also for embedded circuits and Internet of Things applications that are designed to work at low frequencies and low energies. This study shows, through the example of superparam- agnetic tunnel junctions acting as natural noise amplifiers, that emerging nanodevices could be used as highly efficient sources of true randomness for a wide range of applications. ACKNOWLEDGMENTS This work is supported by the European Research Coun- cil Starting Grant NANOINFER (reference: 715872), by the BAMBI EU collaborative FET Project grant (FP7-ICT-2013- C, project number 618024), by a public grant overseen by the French National Research Agency (ANR) as part of the Investissements d'Avenir program (Labex NanoSaclay, refer- ence: ANR-10-LABX-0035), by the ANR grant CogniSpin (reference: ANR-13-JS03-0004), and by the French Minist`ere de l'´ecologie, du d´eveloppement durable et de l'´energie. The authors thank J. Droulez and P. Bessi`ere for fruitful dis- cussion. SUPPLEMENTARY MATERIALS S1: Effect of whitening on random bitstream quality. S2: NIST STS results on a second junction. S3: Discussion on superparamagnetic junction switching process. S4: Sampling frequency and correlation between consecutive samples. S5: Crosstalk through dipolar interaction. S6: Spam detector: Email classifier architecture. S7: Effect of whitening on spam detection. S8: Effect of the number of bits per word on spam detection. S9: Energy consumption of spam detection. S10: Effects of XOR whitening on bitstream probability and auto-correlation. S11: Read disturb effect. ∗ [email protected] 1 R. Courtland, IEEE Spectrum: Technology, Engineering, and Sci- 2 A. Alaghi and J. P. Hayes, ACM Trans. Embed. Comput. Syst. 12, ence News (2016). 92:1 (2013). 8 24 M. Barangi, J. S. Chang, and P. Mazumder, IEEE Transactions CAS) (2015) pp. 1–4. on Magnetics 52, 1 (2016). 25 Y. Kim, X. Fong, and K. Roy, IEEE Magnetics Letters 6, 1 (2015). 26 A. Sengupta, A. Jaiswal, and K. Roy, in 2016 74th Annual Device Research Conference (DRC) (2016) pp. 1–2. 27 H. Lee, F. Ebrahimi, P. K. Amiri, and K. L. Wang, AIP Advances 7, 055934 (2017). 28 K. Yang, D. Fick, M. B. Henry, Y. Lee, D. Blaauw, and D. Sylvester, in 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC) (2014) pp. 280– 281. 29 D. Apalkov, A. Khvalkovskiy, S. Watts, V. Nikitin, X. Tang, D. Lottis, K. Moon, X. Luo, E. Chen, A. Ong, A. Driskill-Smith, and M. Krounbi, J. Emerg. Technol. Comput. Syst. 9, 13:1 (2013). 30 J. Z. Sun and D. C. Ralph, Journal of Magnetism and Magnetic 31 W. Rippard, R. Heindl, M. Pufall, S. Russek, and A. Kos, Physical Materials 320, 1227 (2008). Review B 84, 064439 (2011). 32 A. Mizrahi, N. Locatelli, R. Lebrun, V. Cros, A. Fukushima, H. Kubota, S. Yuasa, D. Querlioz, and J. Grollier, Scientific Re- ports 6 (2016). 33 A. F. Vincent, N. Locatelli, J.-O. Klein, W. S. Zhao, S. Galdin- Retailleau, and D. Querlioz, IEEE Transactions on Electron De- vices 62, 164 (2015). 34 H. Sato, E. C. I. Enobio, M. Yamanouchi, S. Ikeda, S. Fukami, S. Kanai, F. Matsukura, and H. Ohno, Applied Physics Letters 105, 062403 (2014). 35 J. Hayakawa, S. Ikeda, Y. M. Lee, R. Sasaki, T. Meguro, F. Mat- sukura, H. Takahashi, and H. Ohno, Japanese Journal of Applied Physics 45, L1057 (2006). 36 M. Blum, Combinatorica 6, 97 (1986). 37 J. Soto, in Proceedings of the 22nd National Information Systems Security Conference, Vol. 10 (NIST Gaithersburg, MD, 1999) p. 12. 38 The NIST tests also include a uniformity condition on the distri- bution of P-Values among tested sequences37. This condition was passed for all tests for sequences processed by Blum and XOR8. 39 A. Mizrahi, N. Locatelli, R. Matsumoto, A. Fukushima, H. Kub- ota, S. Yuasa, V. Cros, J. Kim, J. Grollier, and D. Querlioz, in 2015 IEEE Magnetics Conference (INTERMAG) (2015) pp. 1–1. 40 H. Sato, T. Yamamoto, M. Yamanouchi, S. Ikeda, S. Fukami, K. Kinoshita, F. Matsukura, N. Kasai, and H. Ohno, in 2013 IEEE International Electron Devices Meeting (2013) pp. 3.2.1–3.2.4. 41 T. M. Maffitt, J. K. DeBrosse, J. Gabric, E. T. Gow, M. C. Lam- orey, J. S. Parenteau, D. R. Willmott, M. A. Wood, and W. J. Gal- lagher, IBM Journal of Research and Development 50, 25 (2006). 42 W. Zhao, C. Chappert, V. Javerliac, and J. P. Noziere, IEEE Trans- actions on Magnetics 45, 3784 (2009). Collins, Physical Review E 60, 284 (1999). 44 A. Coninx, P. Bessi`ere, E. Mazer, J. Droulez, R. Laurent, M. A. Aslam, and J. Lobo, in Rebooting Computing (ICRC), IEEE In- ternational Conference on (IEEE, 2016) pp. 1–8. 45 M. Faix, R. Laurent, P. Bessiere, E. Mazer, and J. Droulez, IEEE Transactions on Emerging Topics in Computing (2016). 3 J. S. Friedman, L. E. Calvet, P. Bessiere, J. Droulez, and D. Quer- lioz, IEEE Transactions on Circuits and Systems I: Regular Papers 63, 895 (2016). 4 A. Morro, V. Canals, A. Oliver, M. L. Alomar, and J. L. Rossello, PloS one 10, e0124176 (2015). 5 T. J. Hamilton, S. Afshar, A. van Schaik, and J. Tapson, Proceed- ings of the IEEE 102, 843 (2014). 6 C. Winstead, V. C. Gaudet, A. Rapley, and C. Schlegel, in Infor- mation Theory, 2005. ISIT 2005. Proceedings. International Sym- posium on (IEEE, 2005) pp. 1116–1120. 7 P. A. Merolla, J. V. Arthur, R. Alvarez-Icaza, A. S. Cassidy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y. Naka- mura, B. Brezzo, I. Vo, S. K. Esser, R. Appuswamy, B. Taba, A. Amir, M. D. Flickner, W. P. Risk, R. Manohar, and D. S. Modha, Science 345, 668 (2014). 8 W. Maass, Proceedings of the IEEE 102, 860 (2014). 9 M. Suri, D. Querlioz, O. Bichler, G. Palma, E. Vianello, D. Vuil- laume, C. Gamrat, and B. DeSalvo, IEEE Transactions on Elec- tron Devices 60, 2402 (2013). 10 B. D. Brown and H. C. Card, IEEE Transactions on Computers 50, 891 (2001). 11 A. Morro, V. Canals, A. Oliver, M. L. Alomar, and J. L. Rossello, PLOS ONE 10, e0124176 (2015). 12 S. S. Tehrani, W. J. Gross, and S. Mannor, IEEE Communications Letters 10, 716 (2006). 13 S. K. Mathew, S. Srinivasan, M. A. Anders, H. Kaul, S. K. Hsu, F. Sheikh, A. Agarwal, S. Satpathy, and R. K. Krishnamurthy, IEEE Journal of Solid-State Circuits 47, 2807 (2012). 14 J. Rajendran, R. Karri, J. B. Wendt, M. Potkonjak, N. McDonald, G. S. Rose, and B. Wysocki, Proceedings of the IEEE 103, 829 (2015). 15 C. Y. Huang, W. C. Shen, Y. H. Tseng, Y. C. King, and C. J. Lin, IEEE Electron Device Letters 33, 1108 (2012). 16 S. Balatti, S. Ambrogio, Z. Wang, and D. Ielmini, IEEE Journal on Emerging and Selected Topics in Circuits and Systems 5, 214 (2015). 17 Y. Wang, W. Wen, H. Li, and M. Hu, in Proceedings of the 25th Edition on Great Lakes Symposium on VLSI, GLSVLSI '15 (ACM, New York, NY, USA, 2015) pp. 271–276. 18 M. Hu, Y. Wang, W. Wen, Y. Wang, and H. Li, IEEE Journal on Emerging and Selected Topics in Circuits and Systems 6, 235 (2016). 19 E. Piccinini, R. Brunetti, and M. Rudan, IEEE Transactions on Electron Devices PP, 1 (2017). ference (2014) pp. 103–104. 21 A. Fukushima, T. Seki, K. Yakushiji, H. Kubota, H. Imamura, and K. Ando, Applied Physics Express 7, 083001 S. Yuasa, (2014). 22 W. H. Choi, Y. Lv, J. Kim, A. Deshpande, G. Kang, J.-P. Wang, and C. H. Kim, in 2014 IEEE International Electron Devices Meeting (2014) pp. 12.5.1–12.5.4. 23 S. Oosawa, T. Konishi, N. Onizawa, and T. Hanyu, in 2015 IEEE 13th International New Circuits and Systems Conference (NEW- 20 X. Fong, M. C. Chen, and K. Roy, in 72nd Device Research Con- 43 A. Neiman, L. Schimansky-Geier, F. Moss, B. Shulgin, and J. J.
1910.11521	1	1910	2019-10-25T04:35:42	High electron density ${\beta}-(Al_{0.18}Ga_{0.82})_2O_3/Ga_2O_3$ modulation doping using ultra-thin (1 nm) spacer layer	[ "physics.app-ph", "cond-mat.mtrl-sci", "cond-mat.other" ]	We report on the design and demonstration of ${\beta}-(Al_{0.18}Ga_{0.82})_2O_3/Ga_2O_3$ modulation doped heterostructures to achieve high sheet charge density. The use of a thin spacer layer between the Si delta-doping and heterojunction interface was investigated in ${\beta}-(Al_{0.18}Ga_{0.82})_2O_3/Ga_2O_3$ modulation doped structures. We find that that this strategy enables higher 2DEG sheet charge density up to 6.1x10^12 cm^2 with mobility of 147 cm^2/Vs. The presence of a degenerate 2DEG channel was confirmed by the measurement of low temperature effective mobility of 378 cm^2/V-s and a lack of carrier freeze out from low temperature capacitance voltage measurements. The electron density of 6.1x10^12 cm^2 is the highest reported sheet charge density obtained without parallel conduction channels in an $(Al_{0.18}Ga_{0.82})_2O_3/Ga_2O_3$ heterostructure system.	physics.app-ph	physics	High electron density 𝜷𝜷-(Al0.18Ga0.82)2O3/Ga2O3 modulation doping using ultra-thin (1 nm) spacer layer Nidhin Kurian Kalarickal1, Zhanbo Xia1, Joe Mcglone1, Yumo Liu2, Wyatt Moore1, Aaron Arehart1, Steve Ringel1, 2, Siddharth Rajan1, 2 1Department of Electrical and Computer Engineering, The Ohio State University, Columbus, OH 43210, USA 2 Department of Materials Science and Engineering, The Ohio State University, Columbus, OH 43210, USA sheet charge density. The use of a thin spacer layer between the Si delta-doping and heterojunction interface was investigated We report on the design and demonstration of 𝛽𝛽-(AlGa)2O3/Ga2O3 modulation doped heterostructures to achieve high in 𝛽𝛽 -(AlGa)2O3/Ga2O3 modulation doped structures. We find that that this strategy enables higher 2DEG sheet charge density up to 6.1x1012 cm-2 with mobility of 147 cm2/Vs. The presence of a degenerate 2DEG channel was confirmed by the measurement of low temperature effective mobility of 378 cm2/V-s and a lack of carrier freeze out from low temperature capacitance voltage measurements. The electron density of 6.1x1012 cm-2 is the highest reported sheet charge density obtained without parallel conduction channels in an (AlGa)2O3/ Ga2O3 heterostructure system. frequency3,4 and power switching applications5,6. The high break down field enables shrinking the overall device footprint With a high theoretical breakdown field strength of 8 MV/cm 1,2, 𝛽𝛽-Ga2O3 has the potential to be useful in several high power amplifiers. Besides the superior breakdown field strength, the availability of native 𝛽𝛽-Ga2O3 substrates 7 -- 9 enables high which results in improved frequency performance for power switching devices and increased output power density for RF quality epitaxial growth using techniques such as molecular beam epitaxy 10,11, metal organic chemical vapor deposition 12,13, halide vapor phase epitaxy 14,15 and pulsed laser deposition 16,17. For lateral power devices it is essential for the channel to be placed close to the gate. Firstly, for enhancement mode devices (which are preferred for power electronics), a lower gate-to-channel spacing leads to higher gate-to-channel capacitance, and therefore enables higher sheet charge density for the same gate voltage swing. In addition, a scaled channel allows for better control of gate-drain electric field and lateral scaling of gate length. The former is important in achieving high average breakdown field strength while the latter is crucial in improving the frequency of operation in RF power amplifiers and reducing on resistance in power switching devices. Lateral devices like 𝛽𝛽-Ga2O3 MESFETs for high frequency application have been demonstrated with high on/off ratio and breakdown voltage but the performance of these devices is mainly limited by the low mobility (50-90 cm2/V-s) 3,4. Introducing 1 2-dimensional electron gas (2DEG) channels are the optimal solution since they provide the highest sheet charge density for a given gate voltage swing, while also providing superior transport properties. 𝛽𝛽-(AlGa)2O3/Ga2O3 modulation doped field effect transistors (MODFETs) are attractive in this respect since they can enable a 2DEG with excellent transport properties. 𝛽𝛽- mainly limited by a low conduction band offset (0.3- 0.4 eV) resulting in low sheet charge density (~2x1012 cm-2). Since 𝛽𝛽- (AlGa)2O3/Ga2O3 MODFETs with mobility as high as 180 cm2/V-s have been demonstrated recently 18, but these devices are (AlGa)2O3/Ga2O3 heterostructure is predicated to have a type I band alignment 19, the conduction band offset is determined by the Al mole fraction. Density functional theory based calculations show that Al composition in excess of 50% is required to 𝑛𝑛𝑠𝑠𝑚𝑚𝑚𝑚𝑚𝑚= 𝜖𝜖Δ𝐸𝐸𝑐𝑐 𝑞𝑞(𝑑𝑑𝑠𝑠+Δ𝑑𝑑), formation of a parallel channel is given by achieve a band offset of 1 eV 16,19. Achieving such high Al composition is limited by phase separation of (AlGa)2O3 above 30% of Al composition due to the dissimilar parent crystal structures 20. Currently, high quality 𝛽𝛽-(AlGa)2O3 films are limited to below 30% of Al incorporation. This fixes the achievable conduction band offset in 𝛽𝛽-(AlGa)2O3/Ga2O3 system to around 0.4 eV. In a delta doped modulation doped heterostructure the maximum permissible 2DEG charge density (𝑛𝑛𝑠𝑠𝑚𝑚𝑚𝑚𝑚𝑚) without the where Δ𝐸𝐸𝑐𝑐 is the conduction band offset between 𝛽𝛽-Ga2O3 and 𝛽𝛽-(AlGa)2O3, 𝑑𝑑𝑠𝑠 is the spacer thickness (Figure.1 (a)), Δ𝑑𝑑 is the effective quantum capacitance distance �𝜖𝜖𝜖𝜖ħ2𝑚𝑚∗𝑞𝑞2� and 𝜖𝜖 is the dielectric permittivity. For a fixed value of Δ𝐸𝐸𝑐𝑐 the only way to increase the charge density is by reducing the spacer thickness 𝑑𝑑𝑠𝑠. Increasing the charge density is crucial in reducing the on band diagram of a 𝛿𝛿 doped (9.5×1012 cm-2) 𝛽𝛽-(AlGa)2O3/Ga2O3 MODFET with a thin spacer layer (1 nm) showing the formation of a 2DEG (6.1×1012 cm-2) without the presence of parallel conducting channels. improved screening of the strong polar optical phonon scattering increasing the mobility 21. Figure.1 (b) shows the simulated resistance and increasing drain current in MODFET devices. Additionally, an increase in charge density would result in Figure.1 (a) Epitaxial diagram of 𝛽𝛽-(AlGa)2O3/Ga2O3 thin spacer modfet, (b) band diagram along vertical cutline through the gate. 2 In this report we demonstrate a method to increase sheet charge density in 𝛽𝛽-(AlGa)2O3/Ga2O3 modfets by aggressively scaling the spacer layer thickness to 1 nm. In the case of 𝛽𝛽-Ga2O3 since the polar optical phonon scattering limited mobility is already low (200 cm2/V-s 22), the increased remote ionized impurity scattering could be expected to have less impact on the total mobility. Therefore using a thin spacer could enable higher 2DEG sheet charge density without significant impact on 2DEG mobility. These ideas are confirmed by scattering rate calculations and experimental results discussed later in this letter (Figure.4 (a)). system equipped with a Veeco oxygen plasma source. The growth utilized Ga limited conditions with a Ga beam equivalent Epitaxial growth: The 𝛽𝛽-(AlGa)2O3/Ga2O3 modulation doped heterostructure was epitaxially grown in a Riber M7 MBE pressure (BEP) of 8×10−8 Torr, O2 pressure of 1.5×10−5 Torr, RF plasma power of 300 W and a substrate temperature of 610 ℃ (pyrometer calibrated to Si emissivity). We grew the epitaxial structure shown in Figure.1 (a) on an Fe-doped 𝛽𝛽-Ga2O3 950 ℃ 24. Al composition of 17% was estimated based on peak separation from HRXRD (Figure.2 (a)) and an RMS roughness of 0.6 nm was observed from AFM (not shown). The growth temperature used here, 610 ℃, was chosen to be lower than our (010) substrate (Novel crystal Technology 23). Fe delta-doping at the growth interface was used to eliminate interface impurity related parasitic conduction paths. Si delta doping was carried out by opening the Si shutter for 2.5 s at a cell temperature of typical optimal growth temperatures to reduce the spread of the Si delta sheet by diffusion through the thin spacer layer. Figure.2 (a) High resolution XRD showing 18% (AlGa)2O3, (b) SIMS profile showing the formation of Si impurity channel at the growth interface, (c) SIMS profile showing compenstation of Si impurity channel by introducing intentional Fe 𝛿𝛿 doping. We added an intentional Fe 𝛿𝛿 doping layer to compensate the parasitic channel formed at the growth interface due to Si without the Fe 𝛿𝛿 doped layer showing the formation of a Si impurity channel. This secondary channel prevents pinch off in contamination. Figure. 2 (b) shows the secondary ion mass spectrometry (SIMS, Evans Analytical Group 25) profile of a sample 3 transistor devices and leads to significant buffer leakage. Introducing acceptor Fe atoms at the growth interface leads to compensation and removal of the parasitic channel. Fe 𝛿𝛿 doping was achieved by opening the Fe shutter for 8s (1140 ℃) at the start of the growth and immediately cooling down the cell (Figure.2 (c)). The epitaxial surface riding tendency of Fe in 𝛽𝛽- Ga2O3 leads to a significant Fe tail of around 500 nm 26. Therefore, a minimum buffer thickness of 500 nm is required so that the active channel is spatially separated from the Fe tail. A 5 nm (AlGa)2O3 back barrier was added in addition to the buffer layer to further improve this spatial separation (Figure.1 (a)). regions and growing heavily doped n-type Ga2O3. Ti/Au (30 nm/ 100 nm) ohmic metal contact was later deposited using e- Charge and transport: We fabricated ohmic contacts on the 𝛽𝛽-(AlGa)2O3/Ga2O3 heterostructure by etching away the contact beam evaporation and annealed at 470 ℃ in N2 ambient. We carried out device isolation using a mesa structure etched using BCl3/Ar based dry etch (BCl3/Ar flow of 20/5 sccm, ICP/RIE power of 200/30 W and process pressure of 15 mTorr). Gate contacts were patterned using stepper lithography followed by e-beam evaporation of Ni/Au (30 nm/ 100 nm) schottky metal stack. measurements. These match fairly well with hall measurements carried out on van der pauw structures, from which we We estimated an ohmic contact resistance of 3.2 Ω.mm and sheet resistance of 8.3 𝐾𝐾Ω/∎ using transfer length extracted a Hall mobility of 147 cm2/V-s and a sheet charge density of 4.4 ×1012 cm-2. The Hall mobility for these structures is comparable to MODFETs with thick spacer layers (𝑑𝑑𝑠𝑠>4 nm) (Figure.3 (d)). Capacitance-voltage measurement (C-V) Based on the C-V measurements we estimate a total sheet charge density of 6.1 ×1012 cm-2 at room temperature, and 5.8 × 1012 cm- 2 at 77 K. Absence of significant carrier freeze out at cryogenic temperature suggests the presence of a degenerate was utilized to probe the 2DEG channel both at room temperature and liquid nitrogen temperature (77 K) (Figure.3 (a, b)). 2DEG without parallel conduction in the (AlGa)2O3 layer. Previous reports showed that when parallel conducting channels were present, the spatial overlap of donors and carriers resulted in carrier freeze out at low temperature 27. 4 Figure.3 (a) Capacitance-voltage characteristics (100 KHz) measured at 300 K and 77 K, (b) extracted charge profile as a function of depth, (c) effective mobility as a function of depth, (d) benchmark plot comparing this result with previous reports of modulation doped 𝛽𝛽-(AlGa)2O3/Ga2O3 heterostructures 18,27,28. We used large gate length structures (100 𝜇𝜇m gate length) to measure the effective mobility as a function of sheet charge. The large gate length reduces the effect of source and drain access regions (2 𝜇𝜇m each), so we can use the ideal gradual channel approximation to estimate the effective drift mobility as 𝜇𝜇𝑒𝑒𝑒𝑒𝑒𝑒=𝑔𝑔𝑑𝑑�𝑉𝑉𝑔𝑔�×𝐿𝐿𝑔𝑔 𝑞𝑞𝑛𝑛𝑠𝑠�𝑉𝑉𝑔𝑔�×𝑤𝑤 where 𝑔𝑔𝑑𝑑(𝑉𝑉𝑔𝑔) is the drain conductance, 𝑛𝑛𝑠𝑠�𝑉𝑉𝑔𝑔� is the channel sheet charge density, 𝐿𝐿𝑔𝑔 is the gate length and 𝑤𝑤 is the gate width. Figure.3 (c) compares the measured effective mobility as a function of depth �𝜖𝜖𝐶𝐶𝑔𝑔𝑔𝑔� at room temperature and 77 K. At room temperature, the effective mobility shows a peak value of 132 cm2/V-s (4.2 ×1012 cm-2 sheet density) and 120 cm2/V-s at zero gate bias (6.1 ×1012 cm-2 sheet density). As the temperature is lowered to 77 K, the mobility increases, with a peak mobility When compared with the lowest sheet resistance obtained in 𝛽𝛽-(AlGa)2O3/Ga2O3 MODFETs before this we find that we have carrier freeze out confirms a degenerate channel with a mobility that is not limited by ionized impurity scattering (Figure.4 (a)). of 378 cm2/V-s and a zero-bias mobility of 360 cm2/V-s. The increased low temperature mobility, coupled with the absence of a significantly lower room temperature sheet resistance due to the increased 2DEG density (Figure.3 (d)). 5 Mobility modelling: To understand the scattering mechanism that limits transport in the thin spacer MODFETs, we compared the experimental effective mobility with theoretical estimates. Scattering mechanisms considered include remote ionized impurity scattering (RI), polar optical phonon scattering (POP), background impurity scattering (BI) and interface roughness scattering (IR), which are expected to be the main limiting mechanisms for mobility. RI, BI and IR scattering limited mobility (𝜇𝜇𝑅𝑅𝑅𝑅, 𝜇𝜇𝐵𝐵𝑅𝑅,𝜇𝜇𝑅𝑅𝑅𝑅) values were calculated based on the analysis in Ref 29 and 30. We used the polar analysis in 31 to evaluate the optical phonon limited mobility (𝜇𝜇𝑃𝑃𝑃𝑃𝑃𝑃) and used Matheissen's rule to approximate the total scattering rate as a sum of the different mechanisms, giving 1𝜇𝜇𝑇𝑇𝑃𝑃𝑇𝑇= 1𝜇𝜇𝑅𝑅𝑅𝑅+ 1𝜇𝜇𝑃𝑃𝑃𝑃𝑃𝑃+ 1𝜇𝜇𝐵𝐵𝑅𝑅+ 1𝜇𝜇𝑅𝑅𝑅𝑅 Table I gives a list of relevant material parameters used for mobility modelling. As shown in Figure.4 (a), the measured effective mobility values are in reasonable agreement with the calculated mobility over a range of sheet carrier density values. Coulombic scattering mechanisms like RI and BI which limits the low temperature mobility (Figure.4 (a)) accounts for only a fraction of the room temperature mobility. The average scattering time at room temperature is therefore dominated by polar optical phonon scattering even at a spacer thickness of 1 nm. TABLE I Material parameters of 𝛽𝛽-Ga2O3 Parameter symbol Value used 0.3 mo Effective mass High frequency dielectric constant DC dielectric constant Optical phonon energy 𝑚𝑚∗ 𝜖𝜖 𝐸𝐸𝑜𝑜𝑜𝑜 𝜖𝜖∞ 𝑛𝑛𝑖𝑖𝑚𝑚𝑜𝑜2𝐷𝐷 Background impurity concentration 𝑁𝑁𝑏𝑏𝑚𝑚𝑐𝑐𝑏𝑏3𝐷𝐷 Δ 𝐿𝐿 Root mean square roughness Donor sheet charge density Correlation length 50 meV 8.85×10−13 F/cm 1013 cm-2 3.5×1018 cm-3 3.6 1.2 nm 4 nm 6 Three terminal devices: Figure. 4 (b), (c) shows the output and transfer characteristics of a three terminal device with a gate Figure.4 (a) Comparison of experimental effective mobility values with scattering theory at 77 K and 300 K, (b) Output characteristics of 𝛽𝛽- (AlGa)2O3/Ga2O3 thin spacer modfet, (c) Transfer characteristics of 𝛽𝛽-(AlGa)2O3/Ga2O3 thin spacer modfet measured at VD=5 V. length (𝐿𝐿𝑔𝑔) of 0.7 𝜇𝜇𝑚𝑚, gate to source spacing (𝐿𝐿𝑔𝑔𝑠𝑠) of 0.7 𝜇𝜇m and gate to drain spacing (𝐿𝐿𝑔𝑔𝑑𝑑) of 1.3 𝜇𝜇𝑚𝑚. The output characteristics show a peak drain current of 120 mA/mm which is the highest reported for a single channel 𝛽𝛽-(AlGa)2O3/Ga2O3 device also shows a peak transconductance (𝑔𝑔𝑚𝑚) of 38 mS/mm and a pinch off voltage of -8 V (Figure.4 (c)). The presence of MODFET. Improvement in output characteristics is a direct result of the improvement in 2DEG sheet charge density. The Si impurity channel at the growth interface is visible as a second peak in the transfer characteristics at -4 V (Figure.4 (c)). We believe this may be because intentional Fe delta doping at the growth interface has not completely compensated interfacial donor impurities. Therefore even though the 2DEG channel is depleted at a gate bias of -3.5 V, the parasitic channel leads to a more negative pinch-off voltage (-8 V). thickness the 2DEG mobility is limited by polar optical phonons and not by remote ionized impurity scattering. Three terminal In conclusion we have showed that usage of a thin (1 nm) spacer layer is an effective strategy in 𝛽𝛽-(AlGa)2O3/Ga2O3 modulation doped heterstuctures. Hall mobility of 147 cm2/V-s and 2DEG sheet charge density of 6.1×1012 cm-2 were demonstrated for Si 𝛿𝛿 doped thin spacer 𝛽𝛽-(AlGa)2O3/Ga2O3 MODFETs. Mobility modelling shows that even at 1 nm spacer devices on thin spacer 𝛽𝛽-(AlGa)2O3/Ga2O3 MODFETs show a peak drain current of 120 mA/mm and a peak transconductance dimensional electron gases could be attractive for power and high frequency transistors based on 𝛽𝛽-Ga2O3. of 38 mS/mm. Future epitaxial structures using such thin spacer layers, or multiple channels formed by stacking a few such 2- 7 We acknowledge funding from the AFOSR under Grant no. FA9550-18-1-0479 (GAME MURI, program manager Dr. Ali Sayir) and from Northrop Grumman NG NEXT Basic Research (Dr.Vincent Gambin). References 1 X. Yan, I.S. Esqueda, J. Ma, J. Tice, and H. Wang, Appl. Phys. Lett. 112, 032101 (2018). 2 M. Higashiwaki, K. Sasaki, A. Kuramata, T. Masui, and S. Yamakoshi, Appl. Phys. Lett. 100, 013504 (2012). 3 Z. Xia, H. Xue, C. Joishi, J. McGlone, N. Kalarickal, S. Sohel, M. Brenner, A. Arehart, S. Ringel, S. Lodha, W. Lu, and S. Rajan, IEEE Electron Device Lett. PP, 1 (2019). 4 A.J. Green, K.D. Chabak, M. Baldini, N. Moser, R. Gilbert, R.C. Fitch, G. Wagner, Z. Galazka, J. Mccandless, A. Crespo, K. Leedy, and G.H. Jessen, IEEE Electron Device Lett. 38, 790 (2017). 5 Gao, Y., Li, A., Feng, Q., Hu, Z., Feng, Z., Zhang, K., Lu, X., Zhang, C., Zhou, H., Mu, W. and Jia, Z., Nanoscale research letters, 14(1), p.8 (2019). 6 W. Li, Z. Hu, K. Nomoto, Z. Zhang, J.-Y. Hsu, Q.T. Thieu, K. Sasaki, A. Kuramata, D. Jena, and H.G. Xing, Appl. Phys. Lett. 113, 202101 (2018). 7 Z. Galazka, R. Uecker, D. Klimm, K. Irmscher, M. Naumann, M. Pietsch, A. Kwasniewski, R. Bertram, S. Ganschow, and M. Bickermann, ECS J. Solid State Sci. Technol. 6, Q3007 (2017). 8 K. Hoshikawa, E. Ohba, T. Kobayashi, J. Yanagisawa, C. Miyagawa, and Y. Nakamura, J. Cryst. Growth 447, 36 (2016). 9 A. Kuramata, K. Koshi, S. Watanabe, Y. Yamaoka, T. Masui, and S. Yamakoshi, Jpn. J. Appl. Phys. 55, 1202A2 (2016). 10 K. Sasaki, A. Kuramata, T. Masui, E.G. Víllora, K. Shimamura, and S. Yamakoshi, Appl. Phys. Express 5, 035502 (2012). 11 H. Okumura, M. Kita, K. Sasaki, A. Kuramata, M. Higashiwaki, and J.S. Speck, Appl. Phys. Express 7, 095501 (2014). 12 Y. Zhang, F. Alema, A. Mauze, O.S. Koksaldi, R. Miller, A. Osinsky, and J.S. Speck, APL Mater. 7, 022506 (2018). 13 R. Fornari, in Gallium Oxide, edited by S. Pearton, F. Ren, and M. Mastro (Elsevier, 2019), pp. 3 -- 30. 14 Y. Yao, S. Okur, L.A.M. Lyle, G.S. Tompa, T. Salagaj, N. Sbrockey, R.F. Davis, and L.M. Porter, Mater. Res. Lett. 6, 268 (2018). 15 J.H. Leach, K. Udwary, J. Rumsey, G. Dodson, H. Splawn, and K.R. Evans, APL Mater. 7, 022504 (2018). 16 F. Zhang, K. Saito, T. Tanaka, M. Nishio, M. Arita, and Q. Guo, Appl. Phys. Lett. 105, 162107 (2014). 17 Yu, F.P., Ou, S.L. and Wuu, D.S., Optical Materials Express, 5(5), pp.1240-1249 (2015). 18 Y. Zhang, A. Neal, Z. Xia, C. Joishi, J.M. Johnson, Y. Zheng, S. Bajaj, M. Brenner, D. Dorsey, K. Chabak, G. Jessen, J. Hwang, S. Mou, J.P. Heremans, and S. Rajan, Appl. Phys. Lett. 112, 173502 (2018). 19 H. Peelaers, J.B. Varley, J.S. Speck, and C.G. Van de Walle, Appl. Phys. Lett. 112, 242101 (2018). 20 S.W. Kaun, F. Wu, and J.S. Speck, J. Vac. Sci. Technol. A 33, 041508 (2015). 21 Ghosh, K. and Singisetti, U., Journal of Materials Research, 32(22), pp.4142-4152 (2017). 22 N. Ma, N. Tanen, A. Verma, Z. Guo, T. Luo, H. (Grace) Xing, and D. Jena, Appl. Phys. Lett. 109, 212101 (2016). 23 See https://www.novelcrystal.co.jp/eng/ for more information on substrates (last accessed October 23, 2019) 24 Kalarickal, N.K., Xia, Z., McGlone, J., Krishnamoorthy, S., Moore, W., Brenner, M., Arehart, A.R., Ringel, S.A. and Rajan, S., Applied Physics Letters, 115(15), p.152106 (2019). 25 See https://www.eag.com/techniques/mass-spec/secondary-ion-mass-spectrometry-sims/ for more information on SIMS scan (last accessed October 23, 2019) 26 A. Mauze, Y. Zhang, T. Mates, F. Wu, and J.S. Speck, Appl. Phys. Lett. 115, 052102 (2019). 27 Y. Zhang, C. Joishi, Z. Xia, M. Brenner, S. Lodha, and S. Rajan, Appl. Phys. Lett. 112, 233503 (2018). 28 S. Krishnamoorthy, Z. Xia, C. Joishi, Y. Zhang, J. McGlone, J. Johnson, M. Brenner, A.R. Arehart, J. Hwang, S. Lodha, and S. Rajan, Appl. Phys. Lett. GAO2018, 023502 (2017). 29 J.H. Davies, The Physics of Low-Dimensional Semiconductors: An Introduction (Cambridge University Press, Cambridge, U.K. ; New York, NY, USA, 1998). 30 D.K. Ferry, S. Goodnick, and J. Bird, Transport in Nanostructures (Cambridge University Press, 2009). 31 B.L. Gelmont, M. Shur, and M. Stroscio, J. Appl. Phys. 77, 657 (1995). 8
1803.07378	1	1803	2018-03-20T11:34:43	Research of migration behavior of space charge packet in polyethylene by electron beam irradiation method under the applied electric field	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	For accurately obtaining the relationship between the carrier mobility and the applied electric field, a new multi-layer sample has been designed. Polyvinyl fluoride (PVF) films were hot pressed on both sides of linear low density polyethylene( LLDPE) to block the charge injection from the electrode, so as to better observe the migration of irradiated electrons. The new multi-layer sample was firstly charged to form a charge packet in the electron beam (e-beam) irradiation setup. And then it was transferred to the Laser Induced Pressure Propagation (LIPP) setup to have the space charge evolution monitored under DC voltages on the order 10-70 kV/mm. The migration of the charge packet has been successfully obtained in this new multi-layer sample. By using the packet front as the reference point, the range of the average mobility of packets at a range from 0.0610-14 to 1.0210-14 m2/(V*s) under calibration local field. The experimental results coincide well with the curve relating charge mobility and the electric field predicted from the Gunn effect-like model.	physics.app-ph	physics	Research of migration behavior of space charge packet in polyethylene by electron beam irradiation method under the applied electric field Hui Zhao2, Yewen Zhang1, 2*, Jia Meng2, Feihu Zheng1 and Zhenlian An1 1 Department of Electrical Engineering, Tongji University, Shanghai 201804, China 2 Shanghai Key Laboratory of Special Artificial Microstructure Materials and Technology, Department of Physics, Tongji University, Shanghai 200092, China (Email: [email protected]) Abstract: For accurately obtaining the relationship between the carrier mobility and the applied electric field, a new multi-layer sample has been designed. Polyvinyl fluoride (PVF) films were hot pressed on both sides of linear low density polyethylene( LLDPE) to block the charge injection from the electrode, so as to better observe the migration of irradiated electrons. The new multi-layer sample was firstly charged to form a charge packet in the electron beam (e-beam) irradiation setup. And then it was transferred to the Laser Induced Pressure Propagation (LIPP) setup to have the space charge evolution monitored under DC voltages on the order 10-70 kV/mm. The migration of the charge packet has been successfully obtained in this new multi-layer sample. By using the packet front as the reference point, the range of the average mobility of packets at a range from 0.06×10-14 to 1.02×10-14 m2/(V·s) under calibration local field. The experimental results coincide well with the curve relating charge mobility and the electric field predicted from the Gunn effect-like model. Keywords: Linear low density polyethylene, polyvinyl fluoride, electron beam irradiation, space charge packet, negative differential mobility 1. INTRODUCTION Charge carrier mobility is a very important parameter for dielectric materials[1, 2] especially for the widely adopted polymeric insulations[3]. Although many researchers presented various methods to measuring carrier mobility in dielectric matetials, the currently accepted feasible measurements are as follows: time-of-flight method[4-6], transient current method[7, 8], surface potential decay method[9, 10] and irridiation eletron method[11, 12]. Recently, various models have been suggested to give a reasonable explanation of the packet-like space charge behavior which has been found in polyethylene material. A Gunn effect-like model, being proposed by many researchers[13-15], postulates that there exists a negative differential relationship for the charge velocity against the electric field when the electric field exceeds a threshold value. Based on this model, the simulation was demonstrated to be able to fit the experimental results very well for different cases[16]. However, it is impossible to verify the existence of the Gunn effect directly from the band structure point of view because of the complicated macrostructure of polymeric materials (e.g. crystalline regions, amorphous regions and the interfaces among them). In this paper, an electron beam irradiation system was constructed for injecting electrons into the multi-layered structure of polyethylene samples. By applying an accelerating voltage on the upper electrode of the electron beam irradiation instrument, amounts of electrons are initially injected into the sample at a certain depth from the surface to form a charge carrier packet. It has to be pointed out that the obtained carrier mobility is valid only when the external electric field is much larger than the electric field resulting from the charge packet itself. Then an external electric field is applied on the sample and the charge carrier packet starts to move. The apparent charge carrier mobility can thus be obtained from the carrier packet velocity under the electric field.The relationship between the carrier packet velocity and the applied electric field was obtained by LIPP method to test the validity of the Gunn effect-like model. 2. SAMPLES PREPARATION AND EXPERIMENT The raw material of polymer samples used in this study is the granular product LL1004 from Exxon Mobile Chemical C. Its characteristics are as follows: density of 0.918 g/cm3 and melt index of 2.8g/10min. The neat LLDPE particles were pressed by flat vulcanizing machine into the sheet samples at 130°C under the pressure of 15MPa for 12 min with a size of 0.5 mm in thickness and of 170 mm in diameter. PVF film with a size of 25μm in thickness and of 60mm in diameter was hot-pressed onto both sides of the sheet samples as blocking layers, and then both sides of the samples were evaporated by aluminum (Al) electrodes of 50 mm in diameter which can guarantee the reliable grounding when the samples are irradiated by the e-beam. The average thickness of Al electrodes is approximate 20nm. The structure of sample with blocking layers and Al electrode is shown in Fig. 1. Fig. 1 Irradiated LLDPE sample hot-pressed with PVF films The schematic setup for electron beam irradiation is as shown in Fig. 2. A filament used for the scanning electron microscope(SEM) acts as the electron beam source. The emitted electrons are directly accelerated toward the sample under a certain dc high voltage which can be adjusted between 0 kV and 80 kV. Without adjustment of e-beam, the quantity of injected charges in sample with higher accelerated energy is relatively larger than that by lower accelerated energy for the same irradiation duration. The sample with a multi-layered structure was placed at a grounded Al substrate with a diameter of 200 mm to ensure the bottom electrode reliable grounding. For the same purpose, a grounded Al plate with a concentric hole 40mm in diameter to allow the samples exposed to the electron beam. In this study, the accelerated voltage is selected for 70kV, the irradiation lasted for 180 s with current density 1.8μA/cm2 in a vacuum of 7.5×10-4 Pa. After the irradiation, the Al electrodes on both sides of the sample are covered with the poly(ethylene-co-vinyl acetate)(EVA) electrodes filled with carbon black particles . The semiconducting electrode was prepared by hot pressing the EVA granules at 90°C under the pressure of 15MPa for 6min with a thickness of about 0.4 mm and of 50 mm in diameter. In order to help propagation of pressure waves signal at interfaces between electrodes and the sample, A semi-conductive layer (silicone oil) should be placed between the sample and the Al electrode to improve the acoustic impedance matching. The irradiated multi-layered sample is transferred into the LIPP system for space charge measurement. This system is composed of the laser (Continuum SureliteII-10, wavelength of 1064 nm, laser pulse width of 7ns, and power of 500mJ), digital oscilloscope (Tektronix TSD5052), and a wide band amplifier (Miteq, 37dB). The principle and details of the LIPP system can be found in previous numerous publications[17, 18]. Fig. 2. Schematic setup for electron beam irradiation. 3. Experimental Results and discussion Figure 3 is a typical experimental result of the multi-layered sample polarized at 30 kV/mm DC voltage at 25°C. As is shown in the figure 3(a), a broad negative charge packet with a peak was detected in the bulk when measurement started, it's noticeable that the injected electron exists in the whole irradiated region. The maximum range of the injected electron located near the irradiated electrode with the distance of 80μm. Concerning about the penetration depth of the injected electron, the electron penetration depth can be calculated by an empirical formula suggested by Weber [19]: (1) Where, =0.55 g/cm2/MeV, E0 is the electron energy in MeV, is the dielectric density, =0.9841,  =3 MeV-1. According to Eq. (1), the electron beam energy of 70keV correspond to the electron penetration depth are about 80μm, which is in good accordance with the experimental results. These injected electrons formed a layer of negative charges that can migrate to the opposite electrode under a certain driving voltage. Corresponding to the experimental results, the space charge packet can be observed by the LIPP method. A typical space charge packet was obtained. After the measurement commenced, the negative charge packet moved from the negative electrode side to the positive one. As it moved on, the height of the peak of the packet gradually decreased. At the meantime, both positive and negative homo charge injections were invisible, denoting the well charge blocking effect of the PVF films[20]. According to the Gunn effect-like model, the velocity of charge packet will get slow during the transportation. This phenomenon is also verified by the experimental results as shown in the figure 3(a). Fig. 3. Typical results of charge distribution and calculated electric field profiles in a multi-layered sample under 30 kV/mm Figure 3(b) show the electric field distribution calculated from the charge profiles shown in figure 3(a). Actually, it must be stressed that the field distribution is no longer exactly the capacitive field distribution as the existence of space charge in the process of charge packet migration. The calculated results indicate that the captured charge, together with the charge packet, induce a significant distortion of local field. The field near the cathode keeps the same value no matter where the packet charge locates in, which means the PVF films own the excellent blocking effect as above mentioned. The local field gradually increases up to the rear edge of the packet and is sharply enhanced in the position of charge packet observed. The field at the front of the packet increases during the process of charge packet migration. 00E(1cm1ER20040060080010001200-15-10-505101520Irradiation directionPacket migration directionCharge density（C/cm3）Thickness(m) 0s 88s 147s 214s 297s 357s 417s 479s（a）020040060080010001200-100102030405060Irradiation directionPacket migration directionElectric field(kV/mm)Thickness(m) 0s 88s 147s 214s 297s 357s 417s 479s（b） Fig. 4. Dependence of calibration local field on migration distance Figure 4 shows the dependence of calibration local field on migration distance under each applied DC field from 10kV/mm to 70kV/mm. The colour scale and numbers to the right represent the migration velocity of charge packets in m/s (same for all plots). Two phenomena are worth mentioning here. On the one hand, for the X axis, all the migration velocity of packet possesses the gradually decreasing trend for each sample under different applied field, which has confirmed by the above experimental results. It's also consistent with the inference of the Gunn effect-like model. On the other hand, for the Y axis, the maximum value of the average migration velocity approximately appeared at the 25 kV/mm, it's about 1.02×10-14m2/(V·s). Correspondingly, the minimum value existed little beyond the 40 kV/mm, it's about 0.06×10-14m2/(V·s). This phenomenon is also verified by the following analytical results as shown in the figure 5. As shown in the Figure 5, the red curve represents the average velocity of charge packets, this figure also shows the dependence of packets velocity on calibration local field for each stressing sample from 10kV/mm to 70kV/mm. The colour scale and numbers to the right represent the migration distance of charge packets in m (same for all plots). According to this figure, we analysed from two aspects. Firstly, for each applied field, the migration velocity of charge packet decreases apparently along its propagation through the insulation which is consistent with the earlier analytical result. Secondly, according to the red curve, negative differential resistance can be observed and the onset of the negative differential resistance was found to be approximately as low as 25kV/mm and the second inflection point appeared little beyond 40kV/mm. The shape of this curve is strikingly similar to the Gunn effect occurring in semiconductors. Based on the analyses above, the experimental results coincide well with the curve relating charge mobility and the electric field predicted from the Gunn effect-like model. 0102030405060708090100200300400500Migration velocity (m/s)Electric field (kV/mm)Migration distance (m)0.0000.10000.20000.30000.40000.50000.60000.70000.80000.90001.0001.1001.200In this experiment, a range of electric fields from 10 kV/mm to 70 kV/mm were applied to irradiated samples and the average velocity of charge packets were obtained by the LIPP method. When the applied field reaches to 70 kV/mm, the curve of the average velocity was able to fully demonstrate the whole extent of the problem. Therefore, the present results are limited to 70 kV/mm. According to the propagation process, repetitive charge packets were not observed, this is because the source of charge packets is the irradiated electron, and the existence of the PVF film effectively blocked the injection of space charge from the sample electrode, but charge fronts propagation as relatively wave-packet could be identified obviously. Fig. 5. Dependence of packets velocity on calibration local field 4. Conclusion The dynamics of charge packets in the new multi-layer polyethylene samples have been investigated through LIPP method under various DC electric fields. In this paper, the source of charge packets is the irradiated electron which is ejected by the electron beam irradiation setup. According to the calculation results, the range of the average mobility of packets at a range from 0.06×10-14 to 1.02×10-14 m2/(V·s) under calibration local field and the results coincide well with the curve relating charge mobility and the electric field predicted from the Gunn effect-like model. 5. Acknowledgment Financial support from the National Natural Science Foundation of China (NSFC Nos. 51277133, 51477118) and Ph.D. Programs Foundation of Ministry of Education of China (No. 20130072110046) is gratefully acknowledged. References [1] Y. Inuishi, "High field conduction and breakdown in solid dielectrics," IEEE Transactions on 01020304050607080900.00.20.40.60.81.01.21.41.61.82.02.201020304050607080900.00.20.40.60.81.01.21.41.61.82.02.2100.0150.0200.0250.0300.0350.0400.0450.0Migration distance (m)Migration velocity (m/s)Electric field (kV/mm) average velocity of charge packetsElectrical Insulation, no. 3, pp. 139-151, 1980. [2] T. Mizutani, "Behavior of charge carriers in organic insulating materials," in Electrical Insulation and Dielectric Phenomena, 2006 IEEE Conference on, 2006, pp. 1-10: IEEE. [3] K. Yoshino, "Carrier mobility measurement," IEEE transactions on electrical insulation, no. 6, pp. 999-1006, 1986. [4] W. E. Spear, "Drift mobility techniques for the study of electrical transport properties in insulating solids," Journal of Non-Crystalline Solids, vol. 1, no. 3, pp. 197-214, 1969. [5] E. Lebedev, T. Dittrich, V. Petrova-Koch, S. Karg, and W. Brütting, "Charge carrier mobility in poly (p-phenylenevinylene) studied by the time-of-flight technique," Applied physics letters, vol. 71, no. 18, pp. 2686-2688, 1997. [6] J. P. Markham et al., "Nondispersive hole transport in a spin-coated dendrimer film measured by the charge-generation-layer time-of-flight method," Applied physics letters, vol. 81, no. 17, pp. 3266-3268, 2002. [7] Y. Zhang, J. Lewiner, C. Alquié, and N. Hampton, "Space charges of different mobilities: a novel data processing method," Journal of applied physics, vol. 77, no. 10, pp. 5195-5200, 1995. [8] N. Hozumi, Y. Muramoto, M. Nagao, and Y. Zhang, "Carrier mobility in ethylene-vinylacetate copolymer estimated by transient space charge," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 8, no. 5, pp. 849-853, 2001. [9] D. K. Davies, "Carrier transport in polythene," Journal of Physics D: Applied Physics, vol. 5, no. 1, p. 162, 1972. [10] T. Yasuda, Y. Yamaguchi, D.-C. Zou, and T. Tsutsui, "Carrier mobilities in organic electron transport materials determined from space charge limited current," Japanese journal of applied physics, vol. 41, no. 9R, p. 5626, 2002. E. Martin and J. Hirsch, "Electron‐Induced Conduction in Plastics. I. Determination of Carrier [11] Mobility," Journal of Applied Physics, vol. 43, no. 3, pp. 1001-1007, 1972. [12] H. Miyake et al., "Space charge accumulation in polymeric materials for spacecraft irradiated electron and proton," in Electrical Insulation and Dielectric Phenomena, 2007. CEIDP 2007. Annual Report-Conference on, 2007, pp. 763-766: IEEE. [13] J. Xia, Y. Zhang, F. Zheng, and C. Xiao, "A physical model about the space charge packet phenomenon in low-density polyethylene," in Properties and applications of Dielectric Materials, 2006. 8th International Conference on, 2006, pp. 147-149: IEEE. [14] K. Matsui, Y. Tanaka, T. Takada, and T. Maeno, "Numerical analysis of packet-like charge behavior in low-density polyethylene under DC high electric field," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 15, no. 3, 2008. [15] J. Jones, J. Llewellyn, and T. Lewis, "The contribution of field-induced morphological change to the electrical aging and breakdown of polyethylene," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 5, pp. 951-966, 2005. [16] J. Xia, Y. Zhang, F. Zheng, Z. An, and Q. Lei, "Numerical analysis of packetlike charge behavior in low-density polyethylene by a Gunn effectlike model," Journal of Applied Physics, vol. 109, no. 3, p. 034101, 2011. [17] P. Laurenceau, G. Dreyfus, and J. Lewiner, "New principle for the determination of potential distributions in dielectrics," Physical Review Letters, vol. 38, no. 1, p. 46, 1977. [18] J. Lewiner, "Evolution of experimental techniques for the study of the electrical properties of insulating materials," IEEE transactions on electrical insulation, no. 3, pp. 351-360, 1986. [19] K. Weber, "Eine einfache reichweite-energie-beziehung für elektronen im energiebereich von 3 keV bis 3 MeV," Nuclear Instruments and Methods, vol. 25, pp. 261-264, 1963. [20] J. Xia, Y. Zhang, Z. An, and F. Zheng, "Blocking effect of PVF on space charge injection into low density polyethylene," in Solid Dielectrics (ICSD), 2010 10th IEEE International Conference on, 2010, pp. 1-4: IEEE.
1902.08443	1	1902	2019-02-22T11:34:20	Lens-Based Beamformer for Low-Complexity Millimeter-Wave Cellular Systems	[ "physics.app-ph" ]	Data rate requirements for cellular communications are expected to increase 1000-fold by 2020, compared to 2010. This is mainly because of the rapid increase in the number of wireless devices and data hungry applications per-device. This creates a formidable bandwidth crisis. Milimeter-wave (mmWave) systems with massive multiple-input multiple-output (MIMO) operation are two complementary concepts poised to meet this exploding demand, as verified by the wealth of investigations by both industry and academia. Nevertheless, existing MIMO processing techniques, requiring a dedicated radio-frequency (RF) up/down-conversion chain per antenna, results in prohibitively high complexity and cost of the mmWave prototypes. The primary objective of this paper is to present a complete feasibility study on alternative beamforming techniques targeted to reduce the system complexity without drastically compromising its performance. More precisely, at 28 GHz, we investigate the end-to-end performance of two different lens-based beamforming topologies. Our study is an amalgam of theoretical modelling, electromagnetic design, and prototype manufacturing, yielding a comprehensive mmWave, massive MIMO performance evaluation.	physics.app-ph	physics	Lens -- Based Beamformer for Low -- Complexity Millimeter -- Wave Cellular Systems M. Ali Babar Abbasi, Vincent F. Fusco, Harsh Tataria, and Michail Matthaiou Institute of Electronics, Communications and Information Technology (ECIT), Queen's University Belfast, UK Email: {m.abbasi, h.tataria, v.fusco, m.matthaiou}@qub.ac.uk INTRODUCTION Data rate requirements for cellular communications are expected to increase 1000 -- fold by 2020, compared to 2010. This is mainly because of the rapid increase in the number of wireless devices and data hungry applications per -- device. This creates a formidable bandwidth crisis. Milimeter -- wave (mmWave) systems with massive multiple -- input multiple -- output (MIMO) operation are two complementary concepts poised to meet this exploding demand, as verified by the wealth of investigations by both industry and academia. Nevertheless, existing MIMO processing techniques, requiring a dedicated radio -- frequency (RF) up/down -- conversion chain per antenna, results in prohibitively high complexity and cost of the mmWave prototypes. The primary objective of this paper is to present a complete feasibility study on alternative beamforming techniques targeted to reduce the system complexity without drastically compromising its performance. More precisely, at 28 GHz, we investigate the end -- to -- end performance of two different lens -- based beamforming topologies. Our study is an amalgam of theoretical modelling, electromagnetic design, and prototype manufacturing, yielding a comprehensive mmWave, massive MIMO performance evaluation. 2 -- D ROTMAN LENS BASED BEAMFORMER: The first proposed beamformer topology is presented in Fig. 1(a). The beamformer comprises of a uniform rectangular array (URA) with two stages of stacked Rotman lenses. URA consists of a coax fed microstrip patch antenna [1] operating at 28 GHz. The antenna has a S11 < -- 10 dB bandwidth of 3650 MHz and is replicated in 3 row 5 columns to form a URA. Coax feed of each antenna element is connected to high frequency SMP mini connectors [2]. URA is followed by horizontally placed stage -- 1 lenses and vertically placed stage -- 2 lenses. All the lenses are developed using tri -- focal lens synthesis approach and optimization guidelines presented in [3], [4]. A Detailed description of design parameters is given in [5]. The Stage -- 1 lenses have 5 array ports and 3 beam ports. The array ports have a one -- to -- one connection to the 5 antenna elements in each row of the URA. Stage -- 2 lenses have 3 array ports and 3 beam ports. In summary, the beamformer can support a total number of 9 RF chains, corresponding to beam ports of 3 × stage -- 2 lenses. The beamformer is capable of steering in both azimuthal and elevation planes depending upon the excitation of a particular RF chain, as show in Fig. 1(b). Vertically placed lenses are responsible of creating a phase ramp in elevation plane while the horizontally placed lenses do the same in azimuth plane. We designed and simulated the entire geometry, presented in Fig. 1(a), in three sequential steps using three numerical simulation software packages namely: Keysight advanced design system (ADS) [6], computer simulation technology microwave studio (CST MS) and CST design studio (CST DS) [7]. Step -- 1: The URA was designed and simulated in CST MS and the performance of all the radiating elements were recorded. The URA unit cell was than optimized for high efficiency operation (88%) at 28 GHz in the presence of other antenna elements and surface mount SMP connectors. Similarly, Rotman lens for both stages was designed in separate CST MS files. Step -- 2: The Lens structures were than separately imported in ADS and phased aligned 50 Ohm transmissions were designed to connect the Lens's body to the edge of the PCB, such that a precise connector placement is ensured. All the transmission lines were designed keeping in mind the physical dimensions required to realize the stacked configuration of Fig. 1(a). Next both Rotman lenses with the transmission lines were imported as separate files of CST MS. Simulations were carried out using finite -- difference -- time -- domain (FDTD) method, while including the edge mount SMP -- M connectors with the Rotman lenses. Step -- 3: Finally, three separate CST MS studio files, i.e. for URA, stage -- 1 and stage -- 2 Rotman lenses were combined as an assembly in CST DS. Using co -- simulation setup, we evaluated the far -- field beamforming capabilities of the proposed beamformer topology. The final results are shown in Fig. 1(b) and summarized in Table 1. Note that the Rotman lens have a capability of replicating the phase -- shifter network, however, due to inherent defects in the lens body, phase -- shifting is not always accurate. These defects are not covered in this study and further details can be found in our concurrent work [8], [9]. To ensure equal power beams in azimuth and elevation zones, we carried out a power equalization in RF chain excitations. It can be noticed from Table 1 that the half power beam width (a) (b) (c) Fig. 1. Two -- stage stacked Rotman lens and URA based beamformer for azimuth and elevation zone coverage. (b) Far -- field realized gain in linear scale. (c) Ergodic sum spectral efficiency estimation comparison of the proposed topology with single -- staged Rotman lens based beamformer. Table 1. Simulated results of two -- stage stacked Rotman lens beamformer when RF chains are excited RF chain Pin (Watt) HPBW along azimuth HPBW along elevation 1 1.89 30.6° 42.4° 2 1.00 28.6° 47.0° 3 1.91 29.7° 44.7° 4 2.63 24.2° 29.0° 5 1.60 24.6° 35.4° 6 2.67 24.5° 31.2° 7 1.97 30.3° 42.5° 8 1.05 28.1° 47.1° 9 1.99 29.8° 44.3° (HPBW) along azimuth is lower than the elevation because of unequal antenna elements in URA rows and columns. In azimuth plane, as the beam tilts from the broad side direction, the HPBW increases by a factor of ~5°. The opposite is true for the elevation plane where HPBW decrease of ~4° is seen, as we move away from the broadside direction. We noticed that this imbalance between the HPBW is a result of power loses that are different for every individual lens port and transmission line. We also noticed that the overall performance of stage -- 2 lens is better compared to stage -- 1 lens. Once reason for that is the physical separation between beam ports in stage -- 2 lens is wider, providing a better port isolation. Note that on average, a 0 dB signal at the input beam ports faces an overall loss of ~8.52 dB before being able to radiate from the URA. Using the beamformer and far -- field pattern measurements, we estimated the end -- to -- end spectral efficiency for an uplink massive MU -- MIMO system. A Detailed description on the system model and propagation channel simulation setup is provided in [5]. The results presented in Fig. 1(c) demonstrates the superiority of a two -- stage stacked Rotman lens topology compared to single stage Rotman lens based beamformer used at MU-MIMO cellular base station. Contrary to single -- stage Rotman lens based beamformer [10] -- [14], the proposed topology is capable of steering in both azimuth of elevation zones, resulting in better separate -- ability of dominant multi -- path components. The difference in ergodic sum spectral efficiency is low for lower SNR values, but it increases with an increase in SNR, due to the second dimension by the proposed topology. HIGH DIRECTIVITY DIELECTRIC LENS BEAMFORMER: The second type of beamformer proposed in this study is the constantεr lens beamformer. The lens structure is based on a homogenous material with a constant permittivity, permeability, conductivity, and mass density. The working principles of such lens type had thoroughly been investigated in the past [15] -- [17]. In the paper, we present an extension to this technique considering the mmWave MIMO array requirements. Assume an EM plane wave exciting an antenna aperture, similar to a MIMO URA receiver. Spherical dialectics have a capacity to focus this energy to a single point, however, this energy convergence cannot be identical for an entire plane wave. Consider, for example, an EM energy path represented by rays a, b and c hitting the lens at multiple points, as shown in Fig. 2(a). The total path -- length for ray a is more than that for ray b, and ray c. In ideal conditions, focusing of these rays to a single point at the same time instance is not possible. Consider now a careful decrease in the EM wave's phase velocity for the ray c, and timing it to match with the total phase velocity of ray a. This way, a plane wave can be converted to a focal point using a spherical lens with carefully selected refractive index. Conversely, placement of a radiating element at this focal point will result in a plane EM wave leaving the lens. For different angle-of-arrivals (AOA), the EM wave focusing is not possible on a plane, as in URA, so Petzval curvature needs to be considered as a focusing contour [18]. Constantεr lens provides a unique flexibility of radiating element placement outside the lens surface, making it a low-complexity, low- cost and a scalable solution. It is important to mention that almost all the hardware anomalies associated to an optical 100 80 60 40 20 0 24 Radiation Efficiency Total Efficiency 26 28 Frequency (GHz) 30 32 (d) (e) (a) (b) (c) Fig. 2. Ray tracing representation of s projected by a plane wave, (b) geodesic placement of radiating elements, (c) Normalized E -- field plot when a single radiating element (a patch antenna) is excited. (d) Constantεr lens antenna efficiency when element Rx1y1 is excited. (e) Far -- field gain corresponding to elements excitation along the xz -- plane. lens hold valid with this lens type (such as astigmatism and coma -- aberration etc.). However, due to the technological advancement of modern microwave engineering, a number of anomalies are intelligible and resolvable. We designed a lens using dielectric material with εr = 2.53. First, we evaluated the optimum Petzval curvature in CST MS [7] using Multilevel Fast Multipole Method (MLFMM) solver. We designed and placed a patch antenna as a radiator at the focal point. The assembly is capable of beamforming in single direction, which is not enough for MIMO operation. We than evaluated the electrical field minima on the Petzval curvature, and placed multiple antennas to realize multi-beams using the same lens aperture. Electric field minima were considered to ensure the least possible mutual coupling between neighbouring radiating elements. We simulated the entre assembly in CST MS using FDTD method, the results are present in Fig. 2(c) -- (e). It can be noticed from Fig. 3(c) that when an individual patch antenna radiator is excited, EM energy enters the lens surface following the condition λlens < λair. In addition, there are surface waves outside lens that creep along the lens surface, which can create an RF cross-talk with other radiating elements. To avoid this, highly directive and isolated radiators can be used, such as wave guides and horn antennas. However, we observed that when such radiators are used to excite the lens, physical location sensitivity increases to an extent that a ~100 μm displacement of the radiator along z -- axis can cost a loss of ~1.5 dB in the peak far-field gain. Thus, a high precision hardware manufacturing is required. We tailored the beamformer in Fig. 2(b) for 28 GHz operation. Highest radiation efficiency of the assembly was achieved when the lens radius is 66 mm, focal radius is 70.5 mm and inter element geodesic spacing along xz -- and yz -- planes is ~13 mm. With an equal power excitation, the HPBW of the beamformer when Rx1y1, Rx2y1 and Rx3y1 are excited is 4.0°, 4.2° and 4.3° (Fig. 2(e)). The HPBW corresponding to any of the elements shown in Fig. 2(b) stayed within the range 4.0° to 4.5°, with a gain value deviation of ±0.4 dB. Very low HPBW and a very high gain makes the proposed architecture a very lucrative choice for high directivity demands of mmWave MIMO RF front-end. CONCLUSION AND FUTURE PROSPECTIVES: Two beamformer topologies for the RF front-end of mmWave MIMO hybrid architecture are presented. Both topologies are studied and optimized for a successful operation at 28 GHz. The first topology comprises of a URA with a two -- stages of stacked Rotman lenses. The second topology is based on a constantεr lens excited by patch antenna radiators placed at the Petzval curvature. In the next step, we are planning to estimate an accurate theoretical upper limit of both beamforming topologies for MU-MIMO scenarios. Practical demonstration of both systems in combination with baseband SP is one of the future goals. REFERENCES [1] [2] K.-L. Wong, Compact and broadband microstrip antennas, vol. 168. John Wiley & Sons, 2004. Pasternack Enterprises, "Full detent mini SMP male connector datasheet," 2018. [Online]. Available: https://www.pasternack.com/images/ProductPDF/PE44489.pdf. R. C. Hansen, "Design trades for Rotman lenses," IEEE Trans. Antennas Propag., vol. 39, no. 4, pp. 464 -- 472, 1991. E. O. Rausch and A. F. Peterson, "Rotman lens design issues," in Antennas and Propagation Society International Symposium, 2005 IEEE, 2005, vol. 2, pp. 35 -- 38. M. A. B. Abbasi, H. Tataria, V. F. Fusco, and M. Matthaiou, "Performance of a 28 GHz two -- stage Rotman lens beamformer for millimeter wave cellular systems," in Proc. IEEE European Conference on Antennas and Propagation (EuCAP 2019), 2019. Keysight, "Advanced Design System (ADS)," 2018. [Online]. Available: https://www.keysight.com/en/pc- [3] [4] [5] [6] [7] [8] [9] 1297113/advanced-design-system-ads?cc=US&lc=eng. "Computer Simulation Technology (CST)," 2018. [Online]. Available: www.cst.com. M. A. B. Abbasi, H. Tataria, V. F. Fusco, and M. Matthaiou, "On the impact of spillover losses in 28 GHz Rotman lens arrays for 5G applications," in Proc. IEEE MTT-S International Microwave Workshop Series on 5G Hardware and System Technologies (IMWS-5G), 2018, pp. 1 -- 3. M. A. B. Abbasi, H. Tataria, V. F. Fusco, M. Matthaiou, and G. C. Alexandropoulos, "Aggressive RF circuit reduction techniques in millimeter wave cellular systems," in Proc. IEEE ICC 2019, 2019. [10] Y. Gao, M. Khaliel, F. Zheng, and T. Kaiser, "Rotman lens based hybrid analog -- digital beamforming in massive MIMO systems: Array architectures, beam selection algorithms and experiments," IEEE Trans. Veh. Technol., vol. 66, no. 10, pp. 9134 -- 9148, Oct. 2017. [11] M. A. Hassanien, M. Jenning, and D. Plettemeier, "Beam steering system using rotman lens for 5G applications at 28 GHz," in Proc. IEEE International Symposium on Antennas and Propagation & USNC/URSI National Radio Science Meeting, , 2017, pp. 2091 -- 2092. [12] A. Darvazehban, O. Manoochehri, M. A. Salari, P. Dehkhoda, and A. Tavakoli, "Ultra-wideband scanning antenna array with Rotman lens," IEEE Trans. Microw. Theory Tech., vol. 65, no. 9, pp. 3435 -- 3442, Sept. 2017. S. Christie et al., "Rotman lens-based retrodirective array," IEEE Trans. Antennas Propag., vol. 60, no. 3, pp. 1343 -- 1351, 2012. [13] [15] [14] Y. J. Cheng, W. Hong, and K. Wu, "Design of a substrate integrated waveguide modified R-KR lens for millimetre-wave application," IET microwaves, Antennas Propag., vol. 4, no. 4, pp. 484 -- 491, 2010. L. Gunderson, "An electromagnetic analysis of a cylindrical homogeneous lens," IEEE Trans. Antennas Propag., vol. 20, no. 4, pp. 476 -- 479, 1972. T. C. Cheston and E. J. Luoma, "Constant-K lenses," Emerson Cuming, APL Tech. Dig. Mars-Avri, 1963. [16] [17] W. Free, F. Cain, C. Ryan, C. Burns, and E. Turner, "High-power constant-index lens antennas," IEEE Trans. Antennas Propag., vol. 22, no. 4, pp. 582 -- 584, 1974. [18] M. Katz, Introduction to geometrical optics. World Scientific Publishing Company, 2002.
1802.02063	1	1802	2018-01-06T02:34:09	Apically Dominant Mechanism for Improving Catalytic Activities of N-Doped Carbon Nanotube Arrays in Rechargeable Zinc-Air Battery	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The oxygen reduction (ORR) and oxygen evolution reactions (OER) in Zn-air batteries (ZABs) require highly efficient, cost-effective and stable electrocatalysts as replacements to traditionally high cost, inconsistently stable and low poison resistant Platinum group metals (PGM) catalysts. Although, nitrogen-doped carbon nanotube (NCNT) arrays have been developed over recent decades through various advanced technologies are now capable of catalyzing ORR efficiently, their underdeveloped bifunctional property, hydrophobic surface, and detrimental preparation strategy are found to limit practical large-scale commercialization for effective rechargeable ZABs. Here, we have demonstrated fabrication of a three-dimensional (3D) nickel foam supported NCNT arrays with CoNi nanoparticles (NPs) encapsulated within the apical domain (denoted as CoNi@NCNT/NF) that exhibits excellent bifunctional catalytic performance toward both ORR (onset potential of 0.97 V vs. RHE) and OER (overpotential of 1.54 V vs. RHE at 10 mA cm$^{-2}$). We further examined the practicability of this CoNi@NCNT/NF material being used as an air electrode for rechargeable ZAB coin cell and pouch cell systems. The ZAB coin cell showed a peak power density of 108 mW cm$^{-2}$ with an energy density of 845 Wh kg$_{Zn}^{-1}$ and robust rechargeability over 28h under ambient conditions, which exceeds the performance of PGM catalysts and leading non-PGM electrocatalysts. In addition, density functional theory (DFT) calculations revealed that the ORR and OER catalytic performance of the CoNi@NCNT/NF electrode are mainly derived from the d-orbitals from the CoNi NPs encapsulated within the apical dominant end of the NCNTs.	physics.app-ph	physics	1 Apically Dominant Mechanism for Improving Catalytic Activities of N-Doped Carbon Nanotube Arrays in Rechargeable Zinc-Air Battery Wenhan Niu1,#, Srimanta Pakhira2,3,4,5,#, Kyle Marcus6, Zhao Li6, Jose L. Mendoza-Cortes 2,3,4,5,* and Yang Yang1,6,* 1NanoScience Technology Center, University of Central Florida, Orlando, FL 32826, United States 2 Department of Chemical & Biomedical Engineering, Florida A&M University - Florida State University, Joint College of Engineering, Tallahassee, Florida, 32310, USA 3 Condensed Matter Theory, National High Magnetic Field Laboratory (NHMFL), Florida State University (FSU), Tallahassee, Florida, 32310, USA 4 Materials Science and Engineering, High Performance Materials Institute (HPMI), Florida State University, Tallahassee, Florida, 32310, USA 5 Department of Scientific Computing, 400 Dirac Science Library, Florida State University, Tallahassee, FL 32306-4120, USA 6Department of Materials Science and Engineering, University of Central Florida, Orlando, FL 32826, United States * Corresponding author: [email protected]; [email protected] # These authors contribute equally to this work. 2 ABSTRACT The oxygen reduction (ORR) and oxygen evolution reactions (OER) in Zn-air batteries (ZABs) require highly efficient, cost-effective and stable electrocatalysts as replacements to traditionally high cost, inconsistently stable and low poison resistant Platinum group metals (PGM) catalysts. Although, nitrogen-doped carbon nanotube (NCNT) arrays have been developed over recent decades through various advanced technologies are now capable of catalyzing ORR efficiently, their underdeveloped bifunctional property, hydrophobic surface, and detrimental preparation strategy are found to limit practical large-scale commercialization for effective rechargeable ZABs. Here, we have demonstrated fabrication of a three-dimensional (3D) nickel foam supported NCNT arrays with CoNi nanoparticles (NPs) encapsulated within the apical domain (denoted as CoNi@NCNT/NF) that exhibits excellent bifunctional catalytic performance toward both ORR (onset potential of 0.97 V vs. RHE) and OER (overpotential of 1.54 V vs. RHE at 10 mA cm-2). We further examined the practicability of this CoNi@NCNT/NF material being used as an air electrode for rechargeable ZAB coin cell and pouch cell systems. The ZAB coin cell showed a peak power density of 108 mW cm-2 with an energy density of 845 Wh kgZn-1 and robust rechargeability over 28h under ambient conditions, which exceeds the performance of PGM catalysts and leading non-PGM electrocatalysts. In addition, density functional theory (DFT) calculations revealed that the ORR and OER catalytic performance of the CoNi@NCNT/NF electrode are mainly derived from the d-orbitals from the CoNi NPs encapsulated within the apical dominant end of the NCNTs. KEYWORDS: apically dominant mechanism; N-doped carbon nanotubes; electrocatalyst; PGM-free; Zn-air battery 3 INTRODUCTION Benefiting from high-energy density, rechargeability and environmental benignancy, Zinc-air battery (ZABs) technology has been widely regarded as one of the most promising energy conversion and storage technologies to meet the growing energy demands of large-scale application for electric vehicles and other electricity-related devices.1-5 The two prominent reactions involved in ZABs are oxygen reduction reaction (ORR) and oxygen evolution reaction (OER), both of which determine the practical performance of ZABs. Thus, the use of highly efficient and stable electrocatalysts as air electrodes is of paramount significance for facilitating the sluggish ORR and OER, thereby attaining high power and long operating lifetime performance for ZABs. Typically, platinum group metals (PGM) are most favorable as ORR and OER catalysts displaying absolute superiority of catalytic activity. However, the high cost, poor stability/durability and low poison resistance of PGM have been the primary barriers that hamper widespread commercialization of ZABs.2, 6-9 With decades of extensive research in developing cost-effective catalysts that possess remarkable catalytic performance, a new generation of PGM-free electrocatalysts based on carbon materials such as heteroatom doped carbon nanotubes, graphene, and porous carbons have been studied. Most of these carbon catalysts exhibit comparable activity to PGM catalysts, which can be attributed to the heteroatom-introduced electron density redistribution of surface carbon atoms and an increase in pentagonal graphene edge defects. Among these, nitrogen-doped carbon nanotube (NCNT) arrays have previously shown impressive performance in fuel cells due to their favorable properties such as the in-situ growth of NCNT electrocatalyst on electrode substrate.5, 10-11 However, many well-developed NCNT arrays present with a hydrophobic surface characteristic that does not allow for O2, H2O, OH- and other ORR or OER intermediates to 4 sufficiently permeate through the NCNT arrays, creating an isolation of interior active sites. Additionally, the fabrication of carbon nanotube arrays are generally achieved through bottom-up processes by using transitional metal nanoparticles (NPs), mainly distributed at the base of NCNT arrays, as seeds for nanotube growth. This nano-architecture displaying hydrophobic properties at the carbon surface is unfavorable for fully accessible active sites being derived from carbon-encapsulated transition metals at the NCNT array base. Besides, neither the commonly used Si or Cu substrates are suitable for serving as an air electrode in ZABs, due to their air-impenetrability and easy corrosion in alkaline electrolyte.12-13 Thus, to further improve the performance of carbon nanotube arrays electrocatalysts in fuel cells will be a great challenge using the conventional synthesis strategy. Inspired by the mechanism of apical dominance in plant growth, where the shoot apex upward growth is crucial for better access to a light source, thus promoting more efficient photosynthesis when compared to lateral bud growth.14-15 Hence, according to plant physiology, it is conceivable that vertically grown NCNT arrays that present apically dominant active sites are better equipped to absorb O2, H2O, OH- and other associated reactants for a more immediate ORR or OER response, than compared with the base-grown NCNT arrays. For this reason, it is of great interest to rationally design a NCNT array electrode with enhanced catalytic activity at the NCNT apex. Herein, we have developed a highly efficient PGM-free catalyst comprised of three-dimensional (3D) nickel foam (NF) supported NCNT arrays in which CoNi alloy NPs were rationally implanted into the NCNT apex (the final product denotes CoNi@NCNT/NF). As a result, the as-prepared CoNi@NCNT/NF exhibited excellent catalytic performance toward both ORR and OER. Moreover, the rechargeable ZAB coin cells were assembled using CoNi@NCNT/NF that showed impressive performance in terms of output power density, and 5 long-term cyclability, outperforming equivalent PGM catalysts. RESULTS AND DISCUSSION The CoNi@NCNT/NF catalyst was synthesized through the following steps (Fig. S1): Firstly, a g-C3N4 layer acted as the carbon source for NCNTs growth, was polymerized and deposited on nickel foam by one step thermal treatment of melamine-covered nickel foam in air (referred to g-C3N4@NF, Fig. S1b); Then, Co3O4 nano-flakes were deposited on the g-C3N4@NF surface by repeatedly soaking g-C3N4@NF in cobalt nitrate solution and subsequent heat treatment in a muffle furnace under air atmosphere (denoted as Co3O4@g-C3N4@NF, Fig. S1d); Finally, after pyrolysis of Co3O4@g-C3N4@NF under nitrogen flow, NCNTs arrays were grown onto the nickel foam by following a "tip-growth" mechanism. More precisely, during heat treatment, the Ni atoms from nickel foam underwent diffusion and subsequent alloying with Co metal to form CoNi alloy NPs that served as catalytic seeds for NCNTs tip-growth. As a result, the NCNTs with CoNi NPs encapsulated within their apical domains were synthesized (denoted as CoNi@NCNT/NF, Fig. S1e). Scanning electron microscopic (SEM) images in Fig. 1a reveal a 3D structure of the nickel foam with open pathways for electrolyte and gas diffusion. A closer SEM observation in Fig. 1b shows vertical NCNT arrays are homogeneously grown on the surface of nickel foam. The corresponding SEM elemental mapping in Fig. S2 shows that C, N, O, Co and Ni elements uniformly exist within the skeleton of CoNi@NCNT/NF. Interestingly, as depicted in Fig. 1 c, the metal NPs (indicated by yellow circles) are only observed within the apex of each carbon nanotube. To further investigate the growth mechanism of CoNi@NCNT/NF, synthesis of the NCNT/NF without encapsulating CoNi NPs was completed for comparison. Elemental mapping of NCNT/NF indicate that the C, N, O and Ni elements are evenly distributed within the nickel 6 foam surface (Fig. S3), where there is an apparent lack of metal NPs in the carbon nanotube arrays apex, as shown in Fig. S4 a and b. This implies that the growth of the NCNT/NF follows a "base-growth" mechanism, further indicating the incorporation of Co compounds into the fabrication precursor plays a vital role in the apically dominant growth mechanism for CoNi@NCNT/NF. Transmission electron microscopy (TEM) image (Fig. 1d) of the CoNi@NCNT catalyst further confirms that metal NPs are encapsulated within the carbon nanotube apex. High-resolution transmission electron microscopy (HRTEM) images in Fig. 1e show a well-defined graphite lattice spacing in the range of 0.42-0.54 nm (see Fig. 1f), which is larger than the typical lattice spacing of pristine graphite (0.34 nm). This can be attributed to the addition of nitrogen elemental dopants in the graphite layers, which increases the spacing of carbon layers.16 This is especially apparent for the carbon layers directly surrounding the metal NPs. The number of carbon layers surrounding the metal NP are also variant and highlighted in Fig. 1e and f as: 1-layer corresponding to the blue and red areas, 3-layer for the yellow area, and 6-layer for the purple area. This implies the metal NP has more effect on the most-outer carbon layer of tip section in NCNT than those of other places. Moreover, a lattice spacing of 0.2 nm is observed for the selected area in HRTEM image (Fig. 1e), which agrees well with the (111) crystal plane of CoNi alloy.17 TEM elemental mapping shows that N and C signals are uniformly distributed within carbon nanotubes, whereas Co and Ni signals are only detected at the black-contrast apical domains in the upper carbon nanotube sections. These findings suggest that nitrogen element is doped into the carbon nanotubes, and the metal NPs within the carbon nanotubes are comprised of Co and Ni elements. X-ray diffraction (XRD) patterns for both CoNi@NCNT/NF and NCNT/NF show diffraction peaks located at 26° (Fig. S5), which is corresponding to the 7 (002) crystal plane of CNTs. Notably, the (002) diffraction peak of CoNi@NCNT/NF exhibits higher intensity and a smaller peak width at half height (FWHM) than that of NCNT/NF, indicating the high graphitization degree of CoNi@NCNT/NF.18 Concurrently, diffraction peaks of the CoNi@NCNT/NF are present at an approximate 0.5º lower than those of nickel metal (JCPDS: 04-0850) and NCNT/NF, indicating the CoNi alloy NPs are encapsulated within the NCNT arrays. Raman spectra of CoNi@NCNT/NF (Fig. S7) shows two prominent peaks at 1359 and 1587 cm-1, corresponding to D and G bands of NCNT, respectively. A higher ratio of D and G band intensities (ID/IG) indicates a more disordered carbon structure. In our case, the value of ID/IG for CoNi@NCNT/NF (0.99) is larger than that of NCNT/NF (0.90), suggesting more defects exist in the carbon layers on the surface near CoNi NPs. X-ray photoelectron spectra (XPS) for both CoNi@NCNT/NF and Ni@NCNT/NF are presented in Fig. S8. As expected, C, N, O and Ni elements are found in both samples, where a Co elemental signal was only detected in the CoNi@NCNT/NF sample. The mole ratio of Co to Ni was estimated to be 0.5 in the CoNi@NCNT/NF. The nitrogen dopants in NCNT can be deconvoluted into pyridinic nitrogen (Pyr-N) and graphitic nitrogen (Gra-N), corresponding to 3.2 at.% for Pyr-N and 5.8 at.% for Gra-N in CoNi@NCNT/NF (Fig. S9). In contrast, there are 4.2 at.% Pyr-N and 5.9 at.% Gra-N (Fig. S10) in NCNT/NF, which are slightly higher than those in CoNi@NCNT/NF. Meanwhile, previous reports suggested that both Pyr-N and Gra-N species can act as active sites for ORR and OER.19-20 Therefore, it may be expected that a higher catalytic activity would be observed on the NCNT/NF as compared to the CoNi@NCNT/NF. On contrary, the following electrochemical measurements performed in this work have demonstrated that the catalytic activity of CoNi@NCNT/NF was indeed higher than NCNT/NF. We hypothesize that the alloyed CoNi NPs 8 within the CoNi@NCNT/NF may have synergistic effects that further improve nitrogen dopant catalytic activity.17 To better understand the effect of apical dominance for improving the catalytic activity of NCNT arrays, we proposed a mechanistic model to illustrate the electrochemical process on CoNi@NCNT/NF and NCNT/NF electrodes, respectively. A typical schematic diagram of the CoNi@NCNT/NF electrode is illustrated in Fig. 2a. We assume that the NCNTs are vertically grown from the NF surface, where much of the electrochemically relevant species have access to only the NCNT array apex, which can be attributed to the hydrophobic NCNT surface preventing molecular permeation. The implantation of CoNi NPs within the NCNT apical domains could promote catalytic efficiency of the surface active sites, thus accelerating ORR/OER processes on the CoNi@NCNT/NF electrode. For comparison, the schematic illustration in Fig. 2b shows a passive catalytic performance for NCNT/NF with scattered reactants (O2, OH- and H+ etc.) being absorbed on the NCNT surface. These reactants have limited access to the active sites for ORR and OER due to the hydrophobic characteristic of NCNT. To verify the previous assumptions, ORR and OER catalytic performance for both samples were first evaluated by steady-state linear sweep voltammograms (LSV) in 0.1 M KOH solution. Fig. 2c shows the CoNi@NCNT/NF electrode with an onset potential (Eonset) of 0.97 V vs. reversible hydrogen electrode (RHE) for ORR, which is more positive than those of Pt/C+RuO2/NF (Eonset = 0.91 V) and NiCNT/NF (Eonset = 0.81 V) electrodes. For OER catalytic performance, the overpotential at 10 mA cm-2 is measured to be E10 = 1.54 V vs. RHE for the CoNi@NCNT/NF electrode, outperforming Pt/C+RuO2/NF (E10 = 1.63 V), NCNT/NF (E10 = 1.59 V), NF (E10 = 1.71 V) electrodes and the other state-of-the-art catalysts (Supplementary Table 1).21-24 The NCNT arrays also provide the nickel foam with an increased, readily accessible active area, as 9 verified by the high electrochemical double layer capacitance for CoNi@NCNT/NF (Cdl = 18.4 mF cm-2) and NCNT/NF (Cdl = 18.3 mF cm-2), which are greater than that of nickel foam (Cdl = 2.4 mF cm-2, Fig. S11). This suggests that the higher active surface area of NCNT/NF and CoNi@NCNT/NF are a result from the modification of nickel foam with NCNT arrays, leading to effective promotion of electrochemically accessible surface. Stability is an important parameter for evaluating catalytic performance of electrocatalysts, the chronopotentiometry curves (Fig. 2d) show the stability of the CoNi@NCNT/NF electrode for ORR and OER, which is much higher than those of other electrodes in our study, with almost no decay of operating potential for either ORR or OER after 10h of operation at 10 mA cm-2. Interestingly, the OER overpotential for CoNi@NCNT/NF is shown to decrease after stability testing for 10h, which could be attributed to the formation of hydroxylated surface.25-26 In contrast, an increased overpotential is observed for both the NCNT/NF electrode and the Pt/C+RuO2/NF electrode after ORR/OER testing for 10h (Fig. 2e). Moreover, the potential difference between LSVs of ORR and OER at 10 mA cm-2 (∆E10) on the CoNi@NCNT/NF is decreased by 0.04 V after 10h operation, whereas increased ∆E10 values are observed with the NCNT/NF (∆E10 = 0.08 V) and Pt/C+RuO2/NF (∆E10 = 0.05 V) electrodes suggesting outstanding catalytic activity and stability for the CoNi@NCNT/NF electrode in ORR and OER. The same phenomenon can also be obtained through the amperometric (i-t) analysis (Fig. S12). To gain further insight into the effect of apical dominance on NCNTs, the catalytic contribution derived from the nickel foam should be ruled out. Therefore, the NCNTs from CoNi@NCNT/NF and NCNT/NF electrodes were individually stripped from nickel foam by sonication, and compared by cyclic voltammetry (CV) using rotating disk electrode (RDE) and rotating ring-disk electrode (RRDE) evaluations. As shown in Fig. S13, CV curves are found to exhibit oxygen 10 reduction peaks for all electrodes in O2-saturated KOH solution. The observed oxygen reduction peak for CoNi@NCNT (0.87 V vs. RHE) is more positive than those of NCNT (0.71 V vs. RHE) and commercial Pt/C (0.83 V vs. RHE), suggesting the remarkable electrocatalytic activity of CoNi@NCNT for ORR. LSV curves in Fig. 3a confirm once again the superior ORR catalytic performance of CoNi@NCNT electrode with a positive onset potential (Eonset) at 0.97 V vs. RHE and a half-wave potential (E1/2) at 0.87 V vs. RHE. In contrast, those values of NCNT electrode (Eonset = 0.88 V and E1/2 = 0.70 V vs. RHE) and Pt/C electrode (Eonset = 0.94 V and E1/2 = 0.82 V vs. RHE) are much smaller than those of CoNi@NCNT, highlighting the enhanced ORR activity of CoNi@NCNT is attributed to the NCNTs being filled with CoNi NPs. In general, the electrical conductivity of electrocatalyst also plays a vital role in ORR performance. The electrochemical impedance spectroscopy (EIS, Fig. S14) of the CoNi@NCNT electrode has a smaller semicircle than that of NCNT electrode, which implies a facilitated ORR process with the CoNi@NCNT electrode arising from the encapsulation of CoNi NPs into NCNTs. Interestingly, the CoNi@NCNT electrode also shows more stable catalytic activity with less ∆E1/2 attenuation (3 mV) than that of Pt/C electrode (32 mV) after 8000 cycles in 0.1 M KOH electrolyte. From the RDE voltammograms in Fig. 3b, it can be observed that the limiting current of CoNi@NCNT electrode increases with an increase in rotation speed (from 225 to 2025 rpm). The corresponding Koutecky-Levich (K-L) plots within the potential range from 0.55 to 0.75 V vs. RHE show a good linear relationship and similar slope to that of Pt/C. This suggests first order ORR reaction kinetics, which are related to oxygen concentration in the electrolyte and lead to a four-electron ORR pathway for the CoNi@NCNT electrode. The kinetic current for CoNi@NCNT (Jk = 26.3 mA cm-2) is extracted from the fitted K-L plot intercept at 0.65 V vs. RHE and is larger than that of Pt/C (Jk = 16.2 mA cm-2) and NCNT electrodes (Jk = 9.4 mA cm-2) 11 as can be seen in Fig. 3d. This describes a facilitated kinetics process for CoNi@NCNT showing an increase in Gibbs free energy (∆G) for ORR. Although, nitrogen content in CoNi@NCNT is lower than that of NCNT (Fig. S10), the encapsulation of CoNi NPs into the apical domain of the NCNTs could indeed optimize the accessibility of reactants to the active sites and therefore enhance catalytic activities of surface carbons as reflected by the above experimental observations. The ORR catalytic performance for CoNi@NCNT were also evaluated by RRDE measurements. The RRDE voltammograms in Fig. 3e show a larger disk current and more positive overpotential of CoNi@NCNT than those of Pt/C. Moreover, the ring current of CoNi@NCNT is as low as that of Pt/C. Concurrently, the electron transfer number (n) and HO2- yield for electrochemical reduction of oxygen can be calculated from ring and disk current (Fig. S15). The calculated values of n = 3.94 and HO2- yield = 7.8% for CoNi@NCNT are close to n = 3.96 and HO2- yield = 2% for the Pt/C electrode, implying a four-electron pathway with high selectivity toward ORR. Despite the CoNi@NCNT electrode performed well in the electro-reduction of oxygen, the actual active sites are still ambiguous since various active species that could be formed on the carbon surface during the pyrolysis, such as N-C, Metal-N/C and metal compounds such as carbides and oxides.6, 27 To ascertain the identity of the active sites, we studied the electrochemical performance of CoNi@NCNT under different detection experiments. As shown in Fig. 3f, compared with the LSV curve for a fresh CoNi@NCNT electrode (black line), no visible diminishment is detected on the CoNi@NCNT electrode after acid etching (red dashed line). This phenomenon suggests the carbides, oxides and other metal compounds are absent in the active sites of the CoNi@NCNT electrode. Furthermore, for detecting the insoluble Metal-N/C species, 10 mM and 20 mM KSCN were separately added into the electrolyte before LSV 12 scanning. One can see from Fig. 3f, the LSVs for CoNi@NCNT after adding 10 mM and 20 mM KSCN into 0.1 M KOH maintain the same feature to that of the original one, clarifying no Metal-N/C species exist in the active sites of CoNi@NCNT. Moreover, the CoNi@NCNT electrode also shows a strong tolerance to methanol crossover, as demonstrated by unnoticeable changes of current density in LSV curves after the addition of 1 M and 3 M methanol to the electrolyte (Fig. S16a). Whereas an obvious peak at 0.7 V vs. RHE appears in the Pt/C electrode LSV in 0.1 M KOH containing 1 M methanol solution, corresponding to the oxidation of methanol. Also, it is not a surprise that the CoNi@NCNT electrode has a superior OER activity than that of NCNT electrode, as evidenced by the LSV curves for OER in Fig. S17. These before mentioned results highlight good activity and stability of CoNi@NCNT ascribe to an optimal balance of surface active site density, nitrogen content and electron conductivity. In addition, we have computationally investigated the catalytic activity of the CoNi catalyst which was supported or encapsulated by NCNT i.e. CoNi@NCNT catalyst by calculating electronic properties of the CNT, NCNT and CoNi@NCNT systems; see Fig. 4. Our present calculations showed that there is a very small electron density appearing around the Fermi Energy (EF) level of the CNT although two bands overlapped and showed Dirac-point features like graphene,28 as shown in Fig. 4a. The Dirac point feature was lost in the band structure of the two wall CNT and two bands are just crossed the Fermi level resulting in less electron density around the EF as depicted in Fig. 4b. This calculation suggests that there is a repulsion between two layers in the two wall CNT, which is a feature that is also found in bilayer graphene.28 The interesting results were observed in the N-doped single wall CNT (i.e. NCNT), where doping caused a large change in the electronic properties of the NCNT. The present computation found that the bands overlapped below the Fermi level and an electron density appeared around the EF 13 of the single wall N-doped CNT as shown in Fig. 4c. The structure of the CoNi catalyst with NCNT i.e. CoNi@NCNT catalyst is shown in Figure Xd along with the band structures and density of states (DOSs). The bands are overlapped on each other resulting a large electron density appeared around the EF, which suggests a metallic behavior. To investigate the large electron density which appeared around the EF, we studied contribution of the d-subshell electron density of the CoNi catalyst which was encapsulated by NCNT. The DOSs of the d-subshell electrons of the individuals Co and Ni atoms were computed and plotted in the inset in Fig. 4d. The diagram suggest that the d-subshell electrons of the Ni atoms provide the major electron density around the Fermi level in the total density of states. Therefore, this is consistent with the d-subshell electron density of the Ni atoms plays an important role in improving the catalytic activity of the CoNi@NCNT and thus this catalyst enhances the catalytic performance for overall oxygen reduction reaction (ORR). To shed light on the ORR catalytic behaviors of, we further investigated the chemical reaction mechanism for the reaction on the surfaces of the CoNi@NCNT. To study the detail ORR mechanism, we computationally designed a cluster model of CoNi cluster and determined the lowest energy state and stability of the CoNi cluster. We found that the CoNi cluster with the spin=3 and multiplicity=4 is the stable conformer and it has the lowest energy state. The CoNi cluster was encapsulated by the N-doped CNT in our model system. To balance the C-valances, extra H-atoms were added in the CoNi@NCNT model system as shown in Fig. 5. We find that ORR starts with O-O chemisorbing onto the surface of CoNi@NCNT. To describe the reaction network for the ORR mechanism using our model system on the surfaces of the CoNi@NCNT, we followed the total four reaction steps below: (1) First, an O2 molecule is adsorbed on the surface of the CoNi@NCNT catalyst, 14 O2 + H2O + e− + [M] [M]-OOH + OH− (1) The DFT calculations show that O2 prefers to adsorb at the bridge sites on the CoNi@NCNT surface with presence of water. In the above Eqn. 2, M represents the active adsorption site of the CoNi@NCNT. After absorbing the O2 on the surface of the catalyst, it forms [M]-OOH and OH-, and therefore the 4e− mechanisms were included for the sequential reduction reactions of O2, and then reduced by H2O together with one electron transfer to form [M]-OOH at top sites and a solvated OH− anion. (2) Second, the surface of [M] bound [M]-OOH formed in this step gives OH− as result of a direct one electron reduction and formation of [M]-O [M]-OOH + e− [M]-O + OH− (2) (3) In the third step, [M]-O was reacted by one electron reduction with H2O and formed [M]-OH as an intermediate. [M]-O + H2O + e− [M]-OH + OH− (3) (4) In the last step, [M]-OH was dissociated into [M] and OH− by absorbing one electron. [M]-OH + e− [M] + OH− (4) The oxygen evolution reaction pathways are shown in Fig. 5, and the reactions pathways were plotted Gibbs' free energy (∆G) versus the reaction coordinate. According to our present DFT-D calculations, the effect of water on the binding energies of OH and OOH is more significant than that of O2 and O in agreement with our experiment. The present study showed how variations in binding energies, which is nothing but ∆G, due to water solvation impact the potential energy diagram of the ORR; see Fig. 5. Our results show that water solvation clearly affects the energetics of elementary steps involved in the ORR. 15 Due to the remarkable catalytic properties toward both ORR and OER have been demonstrated by electrochemical analysis for CoNi@NCNT/NF, it is possible to allow its use in a rechargeable ZAB as an air electrode. For this purpose, a rechargeable ZAB was successfully assembled into a CR2032 coin cell. As illustrated in Fig. 6a, the schematic configuration of the rechargeable ZAB coin cell is comprised of CoNi@NCNT/NF, glass fiber and Zn plate as cathode, separator and anode, respectively. As depicted in Fig. S18, an open-circuit potential (OCP) of 1.40 V was observed at ambient atmosphere by applying a multimeter to the as-constructed ZAB coin cell, suggesting that CoNi@NCNT/NF has excellent catalytic performance even in an open-to-air coin cell configuration. Polarization and power density curves were conducted to evaluate the highest output power of ZAB. As depicted in Fig. 6b, a current density of 146 mA cm-2 and a peak power density of 108 mW cm-2 are observed for the battery using CoNi@NCNT/NF as cathode, which are higher than 80 mA cm-2 and 54 mW cm-2 of ZAB coin cell using commercial Pt/C and those of other ZAB tested in coin cell configuration with ambient conditions (Table S2).2, 19 These results highlight the outstanding catalytic activity of CoNi@NCNT/NF in a ZAB system and are ascribed to the 3D porous structure of nickel foam combined with highly active NCNTs arrays which facilitate O2 diffusion and the optimized surface nature of NCNT arrays by incorporation of CoNi NPs into NCNTs. The specific capacity for our ZAB coin cell was calculated to be 655 mAh gZn-1 at 5 mA cm-2 discharging, corresponded to an energy density of 845 Wh kgZn-1. When the discharge current was increased to 20 mA cm-2, the specific capacity of our battery reached as high as 631 mAh gZn-1 (energy density of 738 Wh kgZn-1). Note that the energy densities are higher than the state-of-the-art ZAB using N, P co-doped carbon foam catalysts.2 Meanwhile, the discharging potentials for ZAB using CoNi@NCNT/NF at 5 mA cm-2 and 20 mA cm-2 were estimated to be over 1.29 V and 1.17 V, respectively. In a contrast, the 16 operating potential of ZAB using commercial Pt/C was only 0.8 V under 20 mA cm-2 discharging. These results confirm again a favorable catalytic activity for CoNi@NCNT/NF serving as air electrode in ZAB. Given that the CoNi@NCNT/NF exhibited good bifunctional activity toward both ORR and OER, the discharge and charge polarization measurements for ZAB should also be considered for assessing the rechargeability using CoNi@NCNT/NF electrode. The polarization curves in Fig. 6d reveal that both the discharging and charging curves of the CoNi@NCNT/NF electrode in ZAB show lower overpotentials at different current densities than those of the ZAB comprised of the Pt/C+RuO2/NF electrode. Similarly, the sufficient rechargeability of ZAB coin cell with bifunctional CoNi@NCNT/NF electrode was also identified by the stable galvanostatic charge-discharge profiles in which no obvious potential drop was detected after a long-term cycling of over 28h at 5 mA cm-2. It is worthy of note that the difference between discharge and charge potential (∆EDC) of the ZAB assembled with the CoNi@NCNT/NF electrode (∆EDC = 0.52 V) is much smaller than that of the battery using the Pt/C+RuO2/NF electrode (∆EDC = 0.8 V) and the leading results reported elsewhere.2, 5 The rechargeable performance sufficiently proves the highly efficient and stable catalytic activity of the CoNi@NCNT/NF electrode toward both ORR and OER in such a harsh environment of ZAB coin cell. As shown in Fig. S19, it is found that the CoNi@NCNT/NF electrode still maintains a high-density of NCNT arrays on nickel foam surface after 28h cycling test, suggesting robust characteristics with a long-term stability in ZAB. In a certain sense, a ZAB with high capacity and controllable potentials is desirable for application beyond small electronic devices such hearing aids. For this consideration, we constructed a rechargeable ZAB pouch cell to show that our bifunctional CoNi@NCNT/NF electrode can be applied in a large scale rechargeable ZAB. As shown in Fig. S20a, a 17 rechargeable ZAB pouch cell contains the same configuration to that of the ZAB coin cell. However, the ZAB can be designed with a flexible configuration such as a serial or parallel circuit using the pouch cell. Unsurprisingly, the rechargeable ZAB pouch cell assembled with the CoNi@NCNT/NF electrode can be used to start-up a timer (1.5 V, Fig. S20b). Meanwhile, we also examined the durability of the CoNi@NCNT/NF in the rechargeable ZAB pouch cell by long-term discharge-charge cycling test. As a result, the pouch cell using CoNi@NCNT/NF electrode shows superior cycling performance with almost no attenuation after 810 cycles (18h) to that of the pouch cell using Pt/C+RuO2/NF (756 cycles with 16.8h, Fig. S21). For the application of ZAB with a higher output potential, we also designed a pouch cell with two tandem ZABs, as illustrated in Fig. S22. The as-prepared pouch cell with two tandem ZABs was discharged under 10 mA cm-2 over 25h, which exhibited a stable discharge potential at 2.31 V without attenuation (Fig. S23). At the same time, the discharge potentials of 1.06 V and 3.46 V were obtained for the single and the three tandem circuits of ZAB pouch cells, respectively, which suggests an effective strategy to extend the application of our ZAB by controlling of the inner configuration and circuit. CONCLUSIONS In summary, we have developed a strategy for the fabrication of 3D nickel foam supported NCNT arrays where CoNi NPs were individually encapsulated within the NCNT apex. The as-prepared CoNi@NCNT/NF showed efficient catalytic activity toward both ORR and OER, that can be attributed to synergetic contributions from the carbon-encapsulated CoNi NPs and nitrogen doped carbon sites. As a proof of concept, ZAB coin and pouch cells were assembled using CoNi@NCNT/NF as the air electrode. The developed ZAB exhibited an open-circuit potential of 1.40 V, an energy density of 845 Wh kgZn-1 and a peak power density of 108 mW cm-18 2, as well as excellent durability (over 28h of discharge-charge operation) in an air condition, which are better than those of PGM catalysts and other leading electrocatalysts reported recently. The DFT calculations revealed that the carbon-encapsulated CoNi NPs within the NCNT apical domain, due to the d-electrons, further improved catalytic activity of the outermost nitrogen doped carbon and provided additional catalytically active sites on the electrode surface. Incorporation of the apical dominance theory for NCNT array fabrication will undoubtedly pave a way for rationally designed advanced bifunctional electrodes for renewable energy systems. ASSOCIATED CONTENT Supporting Information. This material is available free of charge via the Internet at http://pubs.acs.org." Detailed method, material characterizations and electrochemical analysis for morphology, composition, catalytic performances of all samples. AUTHOR INFORMATION Corresponding Author * [email protected] * [email protected] Author Contributions Y.Y. and W.N. designed the experiments. W.N. synthesized and characterized the material. S.P. and J.L.M.C. performed and designed the computational and theoretical studies. Z.L. performed the XRD and Raman analysis. Y.Y., W.N. and K.M. analyzed the experimental data, S.P. and J.L.M.C. analyzed the computational simulations. All authors discussed the results and commented on the manuscript. 19 # These authors contribute equally to this work. Notes The authors declare no competing financial interests. ACKNOWLEDGMENT This work was financially supported by the University of Central Florida through a startup grant (No. 20080741). S.P. and J.L.M-C. were supported by Florida State University (FSU). J.L.M-C. gratefully acknowledges the support from the Energy and Materials Initiative at FSU. The authors thank the High Performance Computer cluster at the Research Computing Center (RCC) in FSU for providing computational resources and support. REFERENCES (1) Li, Y.; Gong, M.; Liang, Y.; Feng, J.; Kim, J. E.; Wang, H.; Hong, G.; Zhang, B.; Dai, H. Nat. Commun. 2013, 4, 1805. (2) Zhang, J.; Zhao, Z.; Xia, Z.; Dai, L. Nat. Nanotechnol. 2015, 10, 444. (3) Niu, W. H.; Li, Z.; Marcus, K.; Zhou, L.; Li, Y. L.; Ye, R. Q.; Liang, K.; Yang, Y. Adv. Energy Mater. 2017, DOI: 10.1002/aenm.201701642. (4) Fu, J.; Hassan, F. M.; Zhong, C.; Lu, J.; Liu, H.; Yu, A. P.; Chen, Z. W. Adv. Mater. 2017, 29, 1702526. (5) Meng, F. L.; Zhong, H. X.; Bao, D.; Yan, J. M.; Zhang, X. B. J. Am. Chem. Soc. 2016, 138, 10226. (6) Niu, W. H.; Li, L. G.; Liu, X. J.; Wang, N.; Liu, J.; Zhou, W. J.; Tang, Z. H.; Chen, S. W. J. Am. Chem. Soc. 2015, 137, 5555. (7) Yang, Y.; Fei, H. L.; Ruan, G. D.; Li, L.; Wang, G.; Kim, N. D.; Tour, J. M. ACS Appl. Mater. Inter. 2015, 7, 20607. 20 (8) Yang, Y.; Fei, H. L.; Ruan, G. D.; Tour, J. M. Adv. Mater. 2015, 27, 3175. (9) Gorlin, M.; Chernev, P.; de Araujo, J. F.; Reier, T.; Dresp, S.; Paul, B.; Krahnert, R.; Dau, H.; Strasser, P. J. Am. Chem. Soc. 2016, 138, 5603. (10) Yu, D. S.; Xue, Y. H.; Dai, L. M. J. Phys. Chem. Lett. 2012, 3, 2863. (11) Lu, Z. Y.; Xu, W. W.; Ma, J.; Li, Y. J.; Sun, X. M.; Jiang, L. Adv. Mater. 2016, 28, 7155. (12) Shui, J.; Wang, M.; Du, F.; Dai, L. M. Sci. Adv. 2015, 1, 1. (13) Zhang, Q.; Huang, J. Q.; Qian, W. Z.; Zhang, Y. Y.; Wei, F. Small 2013, 9, 1237. (14) Cline, M. G. Physiol. Plantarum. 1994, 90, 230. (15) Mikityuk, O. D.; Makeev, A. M.; Makoveichuk, A. Y.; Chkanikov, D. I. Sov. Plant. Physiol. 1982, 29, 910. (16) Ding, W.; Wei, Z. D.; Chen, S. G.; Qi, X. Q.; Yang, T.; Hu, J. S.; Wang, D.; Wan, L. J.; Alvi, S. F.; Li, L. Angew. Chem. Int. Ed. 2013, 125, 11971. (17) Deng, J.; Ren, P. J.; Deng, D. H.; Bao, X. H. Angew. Chem. Int. Ed. 2015, 54, 2100. (18) Andrews, R.; Jacques, D.; Qian, D.; Dickey, E. C. Carbon 2001, 39, 1681. (19) Yang, H. B.; Miao, J. W.; Hung, S. F.; Chen, J. Z.; Tao, H. B.; Wang, X. Z.; Zhang, L. P.; Chen, R.; Gao, J. J.; Chen, H. M.; Dai, L. M.; Liu, B. Sci. Adv. 2016, 2, 4. (20) Guo, D. H.; Shibuya, R.; Akiba, C.; Saji, S.; Kondo, T.; Nakamura, J. Science 2016, 351, 361. (21) Fu, J.; Hassan, F. M.; Li, J. D.; Lee, D. U.; Ghannoum, A. R.; Lui, G.; Hoque, M. A.; Chen, Z. W. Adv. Mater. 2016, 28, 6421. (22) Zeng, M.; Liu, Y.; Zhao, F.; Nie, K.; Han, N.; Wang, X.; Huang, W.; Song, X.; Zhong, J.; Li, Y. Adv. Func. Mater. 2016, 26, 4397. (23) Liu, Z. Q.; Cheng, H.; Li, N.; Ma, T. Y.; Su, Y. Z. Adv. Mater. 2016, 28, 3777. 21 (24) Gadipelli, S.; Zhao, T. T.; Shevlin, S. A.; Guo, Z. X. Energy Environ. Sci. 2016, 9, 1661. (25) Wang, G. L.; Cao, D. X.; Yin, C. L.; Gao, Y. Y.; Yin, J. L.; Cheng, L. Chem. Mater. 2009, 21, 5112. (26) Wang, X. G.; Li, W.; Xiong, D. H.; Liu, L. F. J. Mater. Chem. A 2016, 4, 5639. (27) Jiang, W. J.; Gu, L.; Li, L.; Zhang, Y.; Zhang, X.; Zhang, L. J.; Wang, J. Q.; Hu, J. S.; Wei, Z. D.; Wan, L. J. J. Am. Chem. Soc. 2016, 138, 3570. (28) Srimanta Pakhira; Kevin P. Lucht; Mendoza-Cortes, J. L. arXiv.org 2017, ArXiv1704.0880v2. 22 Figure 1. (a-c) SEM images, (d) TEM, (e) HRTEM, (f) lattice spacings of graphitic layers corresponding to the marked areas in (e), and (g) elemental mapping of CoNi@NCNT/NF. The yellow cycles marked in (c) indicate the CoNi NPs encapsulated within the top of NCNTs. (e) shows four-colored markers in the graphitic layers, suggesting the blue, red, yellow, purple areas possess 1-, 1-, 3- and 6-layer graphene on the surface of metal NP, respectively. The scale bars in (a-e) and (g) are 20 μm, 5 μm, 200 nm, 50 nm, 5 nm, and 50 nm, respectively. The scale bar in the inset of (e) denotes 1 nm. 23 Figure 2. Schematic illustrations of (a) CoNi@NCNT/NF and (b) NCNT/NF. (c) LSV curves, (d) chronopotentiometry (v-t) curves and (e) potential differences (∆E10) between LSV curves at 10 mA cm-2 for CoNi@NCNT/NF, NCNT/NF, NF and Pt/C+RuO2/NF electrodes toward ORR and OER in 0.1 M KOH before and after 10h's testing. 24 Figure 3. (a) Linear sweep voltammogram (LSV) curves for all samples in O2-saturated 0.1 M KOH at 1600 rpm; (b) LSV curves at various rotating speeds and (c) Koutecky-Levich (K-L) plots for the rotating disk electrode (RDE) modified with CoNi@NCNT catalyst; (d) Kinetic current for all samples recorded at 0.7 V vs. RHE; (e) Rotating ring-disk electrode (RRDE) voltammograms for CoNi@NCNT and Pt/C in O2-saturated 0.1 M KOH at 1600 rpm; (f) LSV curves of CoNi@NCNT electrode before and after acid etching, and in 0.1 M KOH with 10 mM KSCN and 20 mM KSCN. 0.20.40.60.81.0‐6‐4‐20ECoNi@NCNT= 3 mV j (mA cm‐2)E (V vs. RHE) Pt/C After 8000 cycles CoNi@NCNT After 8000 cycles NCNTEPt/C= 32 mV0.20.40.60.81.0‐6‐4‐20 j (mA cm‐2)E ( V vs. RHE) CoNi@NCNT After acid etching After adding 10mM KSCN After adding 20mM KSCN0.20.40.60.81.0‐6‐4‐20j (mA cm‐2)E (V vs. RHE) 225 400 625 900 1225 1600 2025 r. p. m.0.20.40.60.81.0‐6‐4‐20Ring current j (mA cm‐2)E (V vs. RHE) CoNi@NCNT Pt/CDisc current0.060.090.120.150.20.30.4‐1/2 (s1/2 rad‐1/2)j‐1 (mA‐1 cm2) 0.75 V 0.7 V 0.65 V 0.6 V 0.55 V Pt/C (0.75 V)0102030Pt/CNCNT jk (mA cm‐2)CoNi@NCNTabcdef25 Figure 4. Equilibrium structures and electronics properties: band structures and density of states (DOSs) were calculated for (a) single wall CNT; (b) two walls CNT; (c) single wall N-doped CNT (i.e. NCNT); and (d) CoNi catalyst which was encapsulated by NCNT (i.e. CoNi@NCNT). The contribution of the d-subshell electron density of the Ni and Co atoms was shown at the bottom of right hand side of (d). 26 Figure 5. Oxygen reduction reaction (ORR) pathways. The relative Gibb's free energy (∆G) vs. reaction coordinate is shown. 27 Figure 6. (a) Schematic illustration of the typical rechargeable ZAB coin cell configuration. (b) Polarization curves and power density curves of the ZAB coin cells using CoNi@NCNT/NF, NCNT/NF and Pt/C/NF as cathodes. (c) Discharging curves using CoNi@NCNT/NF as cathode at two different current densities compared with the ZAB using Pt/C/NF. (d) Discharging and charging polarization curves, (e) galvanostatic charge-discharge cycling (5 mA cm-2, 1 hour for each cycle) of the ZAB coin cells using different air electrodes. (f) Photograph of a timer (1.5 V) driven by the prototype rechargeable ZAB coin cell using CoNi@NCNT/NF cathode. 28 TOC 29 Apically Dominant Mechanism for Improving Catalytic Activities of N-Doped Carbon Nanotube Arrays in Rechargeable Zinc-Air Battery Wenhan Niu1#, Srimanta Pakhira2,3,4,5#, Kyle Marcus6, Zhao Li6, Jose L. Mendoza-Cortes2,3,4,5* and Yang Yang1,6* 1 NanoScience Technology Center, University of Central Florida, Orlando, FL 32826, USA 2 Department of Chemical & Biomedical Engineering, Florida A&M University - Florida State University, Joint College of Engineering, Tallahassee, Florida, 32310, USA 3 Condensed Matter Theory, National High Magnetic Field Laboratory (NHMFL), Florida State University (FSU), Tallahassee, Florida, 32310, USA 4 Materials Science and Engineering, High Performance Materials Institute (HPMI), Florida State University, Tallahassee, Florida, 32310, USA 5 Department of Scientific Computing, 400 Dirac Science Library, Florida State University, Tallahassee, FL 32306-4120, USA 6 Department of Materials Science and Engineering, University of Central Florida, Orlando, FL 32826, USA Corresponding authors: [email protected] (J.L.M-C.) and [email protected] (Y.Y.). # These authors (W.N., S.P.) contribute equally to this work. 30 List of Contents 1. Experimental section 2. Characterization 3. Supplementary Tables: Table S1. Comparison of electrocatalytic activities of CoNi@NCNT with leading catalysts reported recently. Table S2. The performance of reported solid or quasi-solid rechargeable Zn-air battery and our Zn-air coin battery. 4. Supplementary Figures: Fig. S1. Schematic illustration of the synthesis process for CoNi@NCNT/NF and NCNT/NF. Fig. S2. SEM image and elemental mapping of CoNi@NCNT/NF. Fig. S3. SEM image and elemental mapping of NCNT/NF Fig. S4. SEM images of (a, b) NCNT/NF and (c, d) CoNi@NCNT/NF. Fig. S5. XRD patterns of CoNi@NCNT/NF and NCNT/NF. Fig. S6. (a) BET surface area and (b) pore size distribution of CoNi@NCNT/NF Fig. S7. Raman spectrums of CoNi@NCNT/NF, Ni@NCNT/NF and CoNi@NCNT/NF after 28h testing in Zn-air battery. Fig. S8. XPS survey spectra of CoNi@NCNT/NF and NCNT/NF. Fig. S9. High resolution XPS spectra of N 1s for CoNi@NCNT/NF. Fig. S10. High resolution XPS spectra of N 1s for NCNT/NF. Fig. S11. CVs of (a) CoNi@NCNT/NF, (b) NCNT/NF and (c) NF at various scan rates and the corresponding linear fitting of the capacitive current densities against the scan rates. Fig. S12. Amperometric (i-t) curves for CoNi@NCNT/NF, NCNT/NF, NF and RuO2/NF electrodes at 1.6 V vs. RHE. Fig. S13. CVs, LSV curves and K-L plots for CoNi@NCNT, NCNT and Pt/C samples in N2 and O2 saturated 0.1 M KOH. Fig. S14. Electrochemical impedance spectra of CoNi@NCNT and NCNT recorded in O2-saturated 0.1 M KOH. Fig. S15. Electron transfer number (n) and HO2- yield ratio of all samples. Fig. S16. (a) LSVs of CoNi@NCNT electrode before and after adding 1 M or 3 M methanol into O2-saturated 0.1 M KOH. (b) LSVs of Pt/C before and after adding 1 M methanol into O2-31 saturated 0.1 M KOH. Fig. S17. LSVs of CoNi@NCNT and NCNT electrodes in O2-saturated 0.1 M KOH. Fig. S18. Optical image of an open circuit voltage with 1.4 V for the CoNi@NCNT/NF assembled in a rechargeable Zn-air button battery. Fig. S19. SEM images of the CoNi@NCNT/NF electrode after the long-term galvanostatic charge-discharge cycling over 28h. Fig. S20. (a) Schematic illustration of a rechargeable Zn-air pouch cell using CoNi@NCNT/NF electrode. (b) Optical images of a timer (1.5 V) being driving by the rechargeable pouch cell Zn-air battery using CoNi@NCNT/NF as cathode. Fig. S21. Galvanostatic charge-discharge cycling of the rechargeable Zn-air battery pouch cell using CoNi@NCNT/NF cathode compared with the battery using Pt/C+RuO2/NF cathode. Fig. S22. Schematic illustration of two tandem Zn-air circuits in pouch cell using CoNi@NCNT/NF electrode. Fig. S23. Discharge curves of the Zn-air pouch cell with one single, two tandem and three tandem Zn-air circuits using CoNi@NCNT/NF cathode. 32 Experimental section Synthesis of CoNi@NCNT/NF The synthesis of CoNi@NCNT/NF contains the following three steps: i) Firstly, one piece of cleaned nickel foam (denotes as NF, the size is 3.5 cm length and 2 cm width) was fully coated with 5 g melamine powder in a ceramic boat by repeated compression, and followed by heat treating of the melamine-coated nickel foam in a furnace with air atmosphere within 1h to form g-C3N4-coated nickel foam (as depicted in Fig. S1b, the product denotes C3N4/NF); ii) Then, the C3N4/NF was immersed in 0.1 M cobalt nitrate solution, dried and subsequently heat treated in furnace within 10-20s to allow the C3O4 deposition on C3N4/NF surface. This procedure was repeated three times to form Co3O4/C3N4/NF (Fig. S1d); iii) Finally, the CoNi@NCNT/NF was obtained by heat treating of Co3O4/C3N4/NF in CVD at 700 oC with N2 flow followed by washing and drying. Synthesis of NCNT/NF The NCNT/NF was synthesized following the same procedure to the CoNi@NCNT/NF, except that the second step of Co3O4 deposition was removed. Characterization The samples were characterized by a field-emission scanning electron microscope (FE-SEM, ZEISS ultra 55) and a high-resolution JEOL JEM-2100F TEM at 200 kV. Raman spectra were obtained using a Renishaw Raman RE01 scope (Renishaw, Inc.) with a 532 nm excitation argon laser. The chemical state of elements were detected by X-ray photoelectron spectra (XPS) measurements using a PHI Quantera SXM scanning X-ray microscope. The X-ray diffraction (XRD) patterns were obtained by using a powder XRD system (Rigaku D/Max Ultima II, Cu Kα radiation). Nitrogen adsorption-desorption isotherms were measured at 77 K by a Quantachrome Autosorb-3b Brunauer-Emmett-Teller (BET) Surface Analyzer. Before measurements, all samples were degassed at 383 K under vacuum for 2h. Electrochemical analysis All electrochemical measurements were conducted with a CHI 760E electrochemical workstation in a conventional three-electrode cell, with platinum foil as the counter electrode, Ag/AgCl as the reference electrode, and a catalyst-modified glassy carbon electrode (GCE) as the working electrode. The catalyst inks were fabricated by dispersing 2 mg of catalyst into a 33 solution containing deionized water, isopropanol and Nafion (5%) at a volume ratio of 4:1:0.025 to form a homogeneous suspension with the catalyst concentration being 2 mg mL-1. A calculated amount of suspension was then evenly casted on a cleaned GCE surface via syringe and dried in air. Catalyst loading was calculated to be 408 g cm-2 for as-prepared catalysts, 204 μg cm-2 for Pt/C and RuO2, respectively. For steady-state linear sweep voltammetry (LSV) measurements, the working electrode was prepared by tailoring CoNi@NCNT/NF, NCNT/NF into 1 x 1 cm2 area. The Pt/C+RuO2/NF electrode was prepared by drop casting the catalyst ink with 2 mg Pt/C and 1 mg RuO2 into nickel foam, and subsequently drying in air. LSVs for ORR were acquired in an O2-saturated 0.1 M KOH aqueous solution at various rotating speeds (225-2025 rpm) with a scan rate of 5 mV s-1. CVs of all samples were tested in a potential range of 0 to -1 V at a scan rate of 50 mV s-1. Electrochemical impedance spectroscopy (EIS) measurements were performed by applying an AC voltage with 5 mV amplitude in a frequency range from 100000 to 1 Hz and recorded at 0.85 V (E vs. RHE). The electrical double layer capacitance (Cdl) of the all catalysts were measured from double-layer charging curves using cyclic voltammograms (CVs) in a potential range of 1.08-1.18 V. Before collecting the CV data, the working electrodes were scanned for several cycles until their voltammograms were stabilized. The plot of the current density against scan rate was collected at 1.16 V, which shows a linear relationship, with its slope as the double layer capacitance (Cdl). The number of electron transfer (n, eq. 1) and the HO2- yield (eq. 2) in oxygen reduction were estimated by the following equations: (cid:1866)(cid:3404)(cid:2872)(cid:3010)(cid:3253)(cid:3284)(cid:3294)(cid:3286)(cid:3010)(cid:3267)(cid:3284)(cid:3289)(cid:3282)(cid:3015)⁄(cid:2878)(cid:3010)(cid:3253)(cid:3284)(cid:3294)(cid:3286) (1) (cid:1834)(cid:1841)(cid:3398)2(cid:3404)(cid:2870)(cid:2868)(cid:2868)(cid:3010)(cid:3267)(cid:3284)(cid:3289)(cid:3282)(cid:3015)⁄(cid:3010)(cid:3267)(cid:3284)(cid:3289)(cid:3282)(cid:3015)⁄(cid:2878)(cid:3010)(cid:3253)(cid:3284)(cid:3294)(cid:3286) (2) where N is the collection efficiency (37%), IDisk and IRing are the voltammetric currents at the disk and ring electrodes, respectively. The rechargeable Zn-air coin batteries (CR2032 coin cell) tests were performed in a two electrode cell with CoNi@NCNT/NF or Ni@NCNT as the cathode (For precious metal electrocatalysts, the Pt/C or RuO2 ink within ethanol/Nafion solution was drop-casted on nickel foam electrode as air electrode), a polished Zn plate as anode and a glass fiber containing 6 M KOH and 0.2 M Zn(OAc)2ꞏ6H2O as the separator. The battery testing was performed in ambient environment on a LAND CT2001A instrument. 34 Computational methods Periodic Calculations. We have studied the equilibrium structures and electronic properties of the CNT, NCNT and CoNi@NCNT using periodic hybrid dispersion-corrected density functional theory; B3LYP-D3,1-4 or DFT-D for short, as implemented in CRYSTAL17.5 In the present computations, we have incorporated van der Waals (vdW) "-D3" dispersion corrections proposed by Grimme, which includes an additional R-8 term in the dispersion series.2-4 Triple-zeta valence with polarization function quality (TZVP) basis sets were used for the C, N, Co and Ni atoms in the periodic DFT-D calculations.6 The threshold used for evaluating the convergence of the energy, forces, and electron density was 10-7 (a.u.) for each parameter. Integrations inside of the first Brillouin zone were sampled on 4 x 4 x 8 Monkhorst-Pack7 k-mesh grids for all the CNT, NCNT and CoNi@NCNT systems during geometry optimization and 20 x 20 x 20 Monkhorst-Pack k-mesh grids for the calculations of band structure and density of states. Non-periodic Calculations. We have also performed a hybrid dispersion-corrected DFT-D (i.e. B3LYP-D3) calculation to investigate the catalytic activity of the CoNi@NCNT considering a molecular cluster model system.1-4 DFT-D calculations of all the molecular systems studied here were performed using the general purpose electronic structure quantum chemistry program Gaussian 09 package suite with the default convergence criteria.8 The 6-31+G* Gaussian type basis sets were used for the H, C, N and O atoms9 as well as LANL2DZ with effective core potentials (ECPs) were also used for both the Co and Ni atoms.10-11 The LANL2DZ basis set with ECPs, combines the efficiency of a core-potential-containing basis set with the accuracy of all-electron basis sets such as 6-31+G**, and is well balanced with such basis sets. The harmonic vibrational analysis was performed at the optimized geometry using the same level of theory to obtain the zero-point vibrational energy (ZPE). The vibration calculations did not have any imaginary frequencies, that confirms the stability of the present compounds. The reaction enthalpy changes at 298 K was calculated as the sum of the changes in the electronic energy and the calculated enthalpy corrections. The DFT-D method was used for geometry optimization because densities and energies obtained with the method are less affected by spin contamination than other approaches.12-14 The corresponding relative Gibb's free energy of the ORR was calculated as 35 (cid:0) G = E + TSvib + ZPE + Hvib (3) where E is the electronic energy and TΔSvib represented the entropic contributions. Zero-point energy (ZPE) corrections were also included based on the calculated vibrational frequencies. We computed entropy contributions to the energies at ambient temperature via frequency analysis, and used the enthalpy (Hvib) and entropy (Svib) outputs to calculate free energies of formation for all relevant intermediates using the standard established protocol. 36 Supplementary Table 1. Comparison of electrocatalytic activities of CoNi@NCNT/NF with leading catalysts reported recently. Supplementary Table 2. The performance of reported solid or quasi-solid rechargeable Zn-air battery and our Zn-air coin battery. 37 Fig. S1. Schematic illustration of the synthesis process for CoNi@NCNT/NF and NCNT/NF. 38 Fig. S2. SEM image and elemental mapping of CoNi@NCNT/NF Fig. S3. SEM image and elemental mapping of NCNT/NF 39 Fig. S4. SEM images of (a, b) NCNT/NF and (c, d) CoNi@NCNT/NF. The yellow circle areas denote the carbon-encapsulated CoNi NPs within the top of NCNT arrays. Fig. S5. XRD patterns of CoNi@NCNT/NF and NCNT/NF. The inset shows the (002) crystal plane of NCNT at 26°, indicating carbon nanotube arrays were successfully grown on the nickel foam skeleton. 40 Fig. S6. (a) BET surface area and (b) pore size distribution of CoNi@NCNT/NF. Brunauer-Emmett-Teller (BET) analyses were conducted to investigate the specific surface area and pore size distribution of CoNi@NCNT/NF. Nitrogen adsorption-desorption isotherms of CoNi@NCNT/NF (Fig. S6a) show an obvious nitrogen adsorption at the low relative pressure and a hysteresis loops with a wide relative pressure range (0.4-0.9), indicating the co-existence of mesopores and macropores in NCNT arrays. Barrett-Joyner-Halenda (BJH) pore size distribution curves (Fig. S6b) derived from the N2 desorption branches confirm the pore size distribution from 10 to 90 nm. Fig. S7. Raman spectrums of CoNi@NCNT/NF, Ni@NCNT/NF and CoNi@NCNT/NF after 28h testing in Zn-air battery. 41 Fig. S8. XPS survey spectra of CoNi@NCNT/NF and NCNT/NF. Fig. S9. High resolution XPS spectra of N 1s for CoNi@NCNT/NF. 42 Fig. S10. High resolution XPS spectra of N 1s for NCNT/NF. Supplementary Fig.11. CVs of (a) CoNi@NCNT/NF, (b) NCNT/NF and (c) NF were recorded in the potential range of 1.08-1.18 V vs. RHE at various scan rates and the corresponding linear fitting of the capacitive current densities against the scan rates. The results suggest both CoNi@NCNT/NF and NCNT/NF possess a higher electrochemical active area than that of NF, owing to the NCNT arrays grown on nickel foam. Fig. S12. Amperometric (i-t) curves for CoNi@NCNT/NF, NCNT/NF, NF and RuO2/NF electrodes at 1.6 V vs. RHE. As can be seen, the current densities are increased with the time increasing for CoNi@NCNT/NF, NCNT/NF and RuO2/NF electrodes, which suggests the nickel foam substrate modified with CoNi@NCNT arrays, NCNT arrays and RuO2 could effective alleviate the loss of active sites, even facilitate the formation of NiOOH with high OER activity. Whereas the NF electrode is directly contacted with electrolyte and O2, which will lead to the slow dissolution of surface active sites on NF. 43 Fig. S13. CVs, LSV curves and K-L plots for CoNi@NCNT, NCNT and Pt/C samples in N2 and O2 saturated 0.1 M KOH. The scan rate for CV is 50 mV s-1 and for LSV is 5 mV s-1. Supplementary Fig.14. Electrochemical impedance spectra of CoNi@NCNT and NCNT recorded in O2-saturated 0.1 M KOH at 0.85 V vs. RHE with AC amplitude 5 mV and frequency range 10 kHz to 0.01 Hz. The electrode rotation speed is 1600 rpm. 44 Fig. S15. Electron transfer number (n) and HO2- yield ratio of all samples. Fig. S16. (a) LSVs of CoNi@NCNT electrode with 1600 r. p. m. before and after adding 1 M or 3 M methanol into O2-saturated 0.1 M KOH. (b) LSVs of Pt/C with 1600 r. p. m. before and after adding 1 M methanol into O2-saturated 0.1 M KOH. 45 Fig. S17. LSVs of CoNi@NCNT and NCNT electrodes in O2-saturated 0.1 M KOH at 1600 rpm. Fig. S18. Optical image of an open circuit voltage with 1.4 V for the CoNi@NCNT/NF assembled in a rechargeable Zn-air button battery. 46 Fig. S19. SEM images of the CoNi@NCNT/NF electrode after the long-term galvanostatic charge-discharge cycling over 28h. Fig. S20. (a) Schematic illustration of a rechargeable Zn-air pouch cell using CoNi@NCNT/NF electrode. (b) Optical images of a timer (1.5 V) being driving by the rechargeable Zn-air pouch cell using CoNi@CNT/NF as cathode. Fig. S21. Galvanostatic charge-discharge cycling (2 mA cm-2, 80s for each cycle) of the rechargeable Zn-air pouch cell using CoNi@CNT/NF cathode compared with the battery using Pt/C+RuO2/NF air electrode. 47 Fig. S22. Schematic illustration of two tandem Zn-air components in pouch cell using CoNi@NCNT/NF electrode. Fig. S23. Discharge curves of the pouch cell Zn-air battery with one single, two tandem and three tandem Zn-air components using CoNi@NCNT/NF cathode (discharge current: 10 mA cm-2). 48 References 1. Becke, A. D. Density-functional Thermochemistry. III. The Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648−5652. 2. Grimme, S.; Antony, J.; Ehrlich, E.; Krieg, H. A Consistent and Accurate ab initio Parametrization of Density Functional Dispersion Correction (DFT-D) for the 94 Elements H-Pu. J. Chem. Phys. 2010, 132, 154104. 3. Pakhira, S.; Sen, K.; Sahu, C.; Das, A. K. Performance of Double Hybrid Dispersion-Corrected Density Functional Theory: A Computational Study of OCS-Hydrocarbon van der Waals Complexes. J. Chem. Phys. 2013, 138, 164319. 4. Pakhira, S.; Takayanagi, M.; Nagaoka, M. Diverse Rotational Flexibility of Substituted Dicarboxylate Ligands in Functional Porous Coordination Polymers. J. Phys. Chem. C. 2015, 119, 28789-28799. 5. Erba, A.; Baima, J.; Bush, I.; Orlando, R.; Doves, R. Large-Scale Condensed Matter DFT Simulations: Performance and Capabilities of the CRYSTAL Code. J. Chem. Theory Comput., 2017, 13, 5019-5027. 6. Peintinger, M. F.; Oliveira, D. V.; Bredow, T. Consistent Gaussian Basis Sets of Triple-Zeta Valence with Polarization Quality for SolidState Calculations. J. Comput. Chem. 2013, 34, 451−459. 7. Monkhorst, H. J.; Pack, J. D. Special Points for Brillouin-Zone Integrations. Phys. Rev. B 1976, 13, 5188−5192. 8. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Mennucci, B.; Petersson, G. A.; Nakatsuji, H.; Caricato, M.; Li, X.; Hratchian, H. P.; Izmaylov, A. F.; Bloino, J.; Zheng, G.; Sonnenberg, J. L.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Vreven, T.; Montgomery, J. A., Jr.; Peralta, J. E.; Ogliaro, F.; Bearpark, M.; Heyd, J. J.; Brothers, E.; Kudin, K. N.; Staroverov, V. N.; Kobayashi, R.; Normand, J.; Raghavachari, K.; Rendell, A.; Burant, J. C.; Iyengar, S. S.; Tomasi, J.; Cossi, M.; Rega, N.; Millam, J. M.; Klene, M.; Knox, J. E.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Martin, R. L.; Morokuma, K.; Zakrzewski, V. G.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Dapprich, S.; Daniels, A. D.; Farkas, Ö.; Foresman, J. B.; Ortiz, J. V.; Cioslowski, J.; Fox, D. J. Gaussian 09, revision E.01; 49 Gaussian: Wallingford, CT, USA, 2009. 9. Hehre, W. J.; Ditchfield, R.; Pople, J. A. Self-Consistent Molecular Orbital Methods. XII. Further Extensions of Gaussian-Type Basis Sets for Use in Molecular Orbital Studies of Organic Molecules, J. Chem. Phys. 56, 2257 (1972). 10. Hay, P. J.; Wadt, W. R. Ab initio effective core potentials for molecular calculations. Potentials for the transition metal atoms Sc to Hg, J. Chem. Phys. 82, 270-283 (1985). 11. Hay, P. J.; Wadt, W. R. Ab initio effective core potentials for molecular calculations. Potentials for the transition metal atoms Na to Bi, J. Chem. Phys. 82, 283-298 (1985). 12. Baker, J.; Scheiner, A.; Andzelm, J. Spin Contamination in Density Functional Theory. Chem. Phys. Lett. 1993, 216, 380−388. 13. Montoya, A.; Truong, T. N.; Sarofim, A. F. Spin Contamination in Hartree-Fock and Density Functional Theory Wavefunctions in Modeling of Adsorption on Graphite. J. Phys. Chem. A 2000, 104, 6108−6110. 14. Pakhira, S.; Lucht, K. P.; Mendoza-Cortes, J. L. Dirac Cone in two dimensional bilayer graphene by intercalation with V, Nb, and Ta transition metalsArXiv1704.0880v2, Condensed Matter, Materials Science. 2017.
1906.10919	2	1906	2019-08-15T11:58:59	Thermal conductance of interleaving fins	[ "physics.app-ph" ]	Interleaving fins can significantly increase the heat transfer by increasing the effective area per unit base area. The fins are separated uniformly by a gap, which is filled with a flow medium to control the heat flux. The heat flux between the plates depends strongly on the thermal conductivity of the fin material and the medium between them as well as the dimensions. In earlier studies empirical a fitting method is used to determine the total effectiveness of the fins. However, it required complete characterization of the fins for each new set of operating conditions. In this paper, a simplified analytical model, but still preserving the main physical traits of the problem is developed. This model reveals the dimensionless parameter group containing both material properties and the fin geometry that govern the heat transfer. Rigorously testing of the model using a numerical finite element model shows an accuracy within 2% over a large parameter space, varying both dimensions and material properties. Lastly, this model is put to test with previously measured experimental data and a good agreement is obtained.	physics.app-ph	physics	Thermal conductance of interleaving fins Michiel A. J. van Limbeeka,b 1 and Srinivas Vanapallib aMax Planck Institute for Dynamics and Self-Organization, 37077 Gottingen, Germany bUniversity of Twente, Postbus 217, 7500 AE Enschede, The Netherlands Abstract Interleaving fins can significantly increase the heat transfer by increasing the effective area per unit base area. The fins are separated uniformly by a gap, which is filled with a flow medium to control the heat flux. The heat flux between the plates depends strongly on the thermal conductivity of the fin material and the medium between them as well as the dimensions. In earlier studies empirical a fitting method is used to determine the total effectiveness of the fins. However, it required complete characterization of the fins for each new set of operating conditions. In this paper, a simplified analytical model, but still preserving the main physical traits of the problem is developed. This model reveals the dimensionless parameter group containing both material properties and the fin geometry that govern the heat transfer. Rigorously testing of the model using a numerical finite element model shows an accuracy within 2 % over a large parameter space, varying both dimensions and material properties. Lastly, this model is put to test with previously measured experimental data and a good agreement is obtained. 1 Introduction Temperature control is essential in many situations: from regulating the body tempera- ture of biological species to keeping your drinks cold in your fridge and preventing your computer from overheating. Whereas rapid heating can be easily achieved by resistive electrical heating, rapid cooling is more complicated as heat needs to be transported towards a heat sink. Many cases use conduction for transport for which the thermal conductance is constant. For some cases however, this is not desirable and one wants to regulate the conductance ie change it in time. As an example in hyperpolarized MRI molecular imaging systems, the sample sleeves that accommodate variable temperature inserts are integrated in the cryostat structure, requiring the entire system to be warmed up for servicing. In a recent de- velopment a heat switch was used to thermally disconnect the sample sleeve from the cold plate for quick servicing [1]. Another common example is a thermal battery, where some (phase change) material [2, 3, 4, 5] is used to maintain a desired operating temperature window for the instrument. The battery should be thermally coupled to a cold sink to charge the system [6]. In situ cryofixation of cells is of growing interest in phasecontrast microscopy for studying dynamic cellular processes associated with 1Corresponding author: [email protected] 1 physiological and pathophysiological conditions [7, 8]. Moreover, many space instru- ments require that the cryocooler system possess a very high level of reliability. This need for high reliability requires that some form of redundancy be incorporated. One common implementation is standby redundancy. Until the primary unit breaks down the thermal connection to the redundant cooler should be low [9]. However, when the standby unit is needed in case of the break down of the primary unit, the thermal con- ductance should be high. A similar thermal behavior is essential for the connection between equipment and space radiators [10, 11, 12]. Unifying these examples is the use of a thermal switch, whose conductance can be changed rapidly. In the off-state, the heat transfer is minimized, whereas in the on-state it is maximized in order to exchange heat rapidly with the heat sink/source. Such a switch can be constructed by having two plates with a small gap between them [13]. The conductivity across the gap increases dramatically when the content is changed from a vacuum by a gas [14] or by replacing it with a liquid. The gas pressure can be controlled actively or passively by an evaporator [15] or sorption cells [13, 16]. To increase the heat flux density per base area, interleaving fins can be used. Rows of fins are attached on the two aforementioned plates in a staggered configuration. Previous research on interleaved fins explored the effective thermal resistance across the base plates as a function of the fin dimensions assuming uniform temperature along the length of the fin. This assumption require high thermal conductivity of the fin compared to the gas, and a further requirement is to have a low fin aspect ratio. Studies were done at both cryogenic [17, 18] and room temperature [19, 20]. Although fins are an efficient way to increase heat exchange to the surroundings, they can loose its efficiency when the fin cools down in the longitudinal direction. This effect can be characterized by the Biot number of the fin: Bi = hL/k where h is the heat transfer coefficient, L the length of the fin and k the thermal conductivity. For Bi (cid:28) 1 the conduction in the fin can easily supply the heat towards the cooling edges and no gradients occur, whereas for Bi (cid:29) 1 the opposite holds and the fin has a non- uniform temperature profile. Although here the heat transfer coefficient is associated with convective transport towards the far-field temperature T∞, we expect that in the present case of two opposing staggered fin arrays a similar behaviour can be expected. Indeed, when replacing h by purely conduction over a gas filled gap we obtain Bi∗ = kgL/(ksD), where D denotes the (uniform) gap thickness and the subscripts g and s refer to the gas and solid domains. What differs however from the classical fin is that here T∞ is now the local temperature of the opposing fin, in this way coupling the two sides of the thermal link. In this study we develop an analytical one-dimensional model, which is validated by numerical solutions of the problem. We identify a more accurate Biot number and find that indeed the cooling starts to be relevant when this number is of the order unity. Several ways are then explored to test the applicability of the model and for which parameters it suffers accuracy. We also provide an expression for the heat flux across the thermal link, which is more relevant for the design. Finally we test the gap exper- imentally in a range where the fins suffer from cooling, finding good agreement with our analytical prediction. 2 Figure 1: Sketch of the geometry of a single pair of fins 2 Problem description and Approach The thermal link is made up by two opposing plates with staggered fins. In between them is a gap of width D, which filled with a gas of conductivity kg, where the hat is used to indicate dimensional quantities. The hot plate and cold plate of the stack is separated by a distance L and the problem is quasi-2D. To increase the effective area, fins of width W are attached on both plates in the space between them. Although in general the spacing and length are free to choose, we here restrict ourselves to the case of a constant separation D. The fins start from a position δ from the top or bottom of the stack. The region where heat is exchanged is then L = L − 2δ. Figure 1 shows this geometry. Many fins are positioned next to each other in the y-dimension with spacing 2 W . The hot plate is at T = Th and the cold plate at T = Tc. We are interested on the heat transfer between the plate, which have a conductivity ks. We identified thus four geometrical parameters, namely L, δ, W , D and three parameters related to the heat transfer: the conductivities of the gas kg and solid ks and the temperature differ- ence between the hot and cold plate. Let us now outline the procedure to describe the heat transfer analytically, without the need of a full quasi- two dimensional numerical simulation. Isothermal fins Similar to the classical problem of a single fin in an infinite fluid (see [21] 2.7), we expect in the limit of ks → ∞ no gradients to emerge into the fin. The heat exchanged Q is then simply −kg A∆ T / D, with A the fin area. Consider a array of interleaving fins on a plate of size Z times Y . One finds for the total area A = 2 W 2( W + L − 2δ− D) Z and obtains for the heat flux q(cid:48)(cid:48) = Q/( Z Y ): Y q(cid:48)(cid:48) = −( Th − Tc)kg W + L − 2δ− D D W . (1) This solution will be used as a reference to compare the case in which the fin is subject to non-isothermal effects, which will be discussed now. Non isothermal fins Let us now sketch the solution for non-isothermal fins. The slenderness of the fins prompts one to use a one-dimensional model, similar to the 3 classical fin problem. We study a single pair of fins, for which we find a profile T (x) for the temperature in the fin at the cold plate. Similarly for the fin at the hot plate we have the profile T ( x). We will then use the temperature difference Θ( x) between them, as the heat exchanged at any position in the fin is proportional to Θ. The problem is described by two coupled heat equations, which require four boundary conditions to solve. The problem will be solved in the following section. 3 Modelling cooling in the fins Depending on the length scales and material properties, we explore the most simple solution to the problem, without introducing too many errors compared the full 2-D (numerical) calculation. Let us first investigate the conditions where a one dimensional approach is applicable. Since we here study a steady-state situation without any heat sources, the heat equation in the domain reads ∇·(cid:126)j = 0, (2) where (cid:126)j = k∇ T ; the heat flux which is modelled using Fourier's law. In the current study we assume k to be independent of the temperature. Considering a single fin now, we can rescale the x and y coordinate with the length L and width W of the fin as x = L ¯x and y = W ¯y ,= obtaining: ks(cid:0)∂ x x T + ∂ y y T(cid:1) = ks(cid:18)∂ ¯x ¯x T + L2 ( W − D)2 ∂ ¯y ¯y T(cid:19) = 0, (3) (4) (5) L2 which acts on the domain x,y = [0,1]. ( W− D)2 the gradients in the ¯y direction become small compared those in the ¯x direction. This prompt us to assume T ( x, y) ≈ T ( x), which we will validate a posteriori. Equation 2 is now integrated over a small slice of the fin. The slice extends over the full width of the fin W − D and has a thickness ∆ x so we obtain using Gauss-theorem in 2 dimensions: Inspection shows that for large (cid:73) (cid:90) ∇·(cid:126)j dV = (cid:126)j·(cid:126)n d(cid:96) = 0 ks( W − D)(cid:18)d T ( x + ∆ x) d x (cid:19) + 2∆ x qgap = 0. d T ( x) − Evaluating the limit of ∆ x → 0 yields, ks( W − D) d2 T ( x) d x2 + 2 qgap = 0, d x v where a Taylor expansion was used. Here qgap is the heat exchanged with the opposing kg D ( T ( x)− T ( x)). Using a similar procedure for the opposing fin, which we model as − fin gives us the following coupled equations for the temperature profiles T ( x) and T ( x): 2kg D ks( W − D) ks( W − D) d2 T d x2 − d2 T d x2 − ( T ( x)− T ) = 0 2kg ( T − T ) = 0 D and . (6) 4 d T dx(cid:12)(cid:12)(cid:12)−L/2 dx(cid:12)(cid:12)(cid:12)L/2 The coupled system of equations is subject to the following boundary conditions: The temperature profiles should satisfy the Dirichlet boundary condition T (−L/2) = Tc or T (L/2) = Th respectively. The problem is closed by noting that heat is conserved in the system, from which we find: ks = ks d T . It is interesting to investigate similarities with section 2.7.4 in reference Bejan[21] to validate the one dimensional model. The present model is valid as long as the Biot number for the width of the fin Bi† = h( W − D)/ks = kg( W − D)/(ks D) (cid:28) 1. Here we used h = kg/ D for the heat transfer coefficient over the conducting gas gap. As long as Bi† = hW /k = kgW /(kD) < 1, no gradients occur in the cross sectional plane. We now non-dimensionalize the spacial dimension and lengths by the thickness L of the stack: x = x L and the temperatures using the boundary temperatures (e.g. T = ( Th − Tc)T + Tc). For thermal conductivities we rescale by the conductivity of the solid: k = kg/ks. Let Θ = T − T , and C2 = we then obtain 4kg L2 ks D( W− D) Θ(cid:48)(cid:48) −C2Θ = 0, 1 2 C2Θ = 0, T (cid:48)(cid:48) + T(cid:48)(cid:48) − 1 2 C2Θ = 0, (7) (8) and where we used primes to represent the derivatives. Let the centre of the stack be at x = 0. We then find for the boundary conditions: T (−1/2) = 0, T (1/2) = 1 and T(cid:48)(−1/2) = T (cid:48)(1/2). The solution for Θ is then αexp (Cx) + βexp (−Cx). Theorem 1. The symmetry of the geometry results in Θ to be an even function. Proof. Let Θ and Θ be solutions to Eq. 3 and Θ(x) = Θ(−x). We then find Θ(−x)(cid:48)(cid:48) − C2Θ(−x) = 0 → Θ(x)(cid:48)(cid:48) −C2 Θ(−x) = 0. At x = 0 the two solutions coincide: Θ(0) = Θ(0) and hence from uniqueness we deduce Θ(x) = Θ(x) = Θ(−x) to be an even function. In order for Θ to be even we find α = β ≡ 2Θ0, which yields: Θ = Θ0 cosh (Cx) , Substitution of Eq. 7 into Eq. 8 yields for T after integrating twice: T = 1 2 Θ0 cosh (Cx) + a + b x, (9) (10) where we find three integration constants: a,b and Θ0. First, we use continuity of flux in a single fin to find per pair of opposing fins, i.e. per 2W : − (W − D)T(cid:48)− 1 2 +δ = 2 k D Θ dx− (W − D) T(cid:48) 1 2−δ. (11) The left-hand side denotes the flux at the base of the fin, the integral the flux crossing the gap sideways and the last term the heat exchange at the tip of the fin. Using Eq. 10 one obtains (cid:90) 1 2−δ − 1 2 +δ b = Θ0 C 2 sinh (C )− 2 W − D k D Θ0 2 C sinh (C )− k D Θ0 cosh(C ), 5 where C = C( 1 2 + δ). This way we can express a as: 2W(cid:0)T (− 1 and directly flowing from the base into the tip of the opposing fin: −(W −D)T(cid:48)− 1 (W − D) k 2 − δ). Next we approximate the flux through the support of the fins as 2 + δ)(cid:1) /δ. This flux is equal to the sum of the heat entering the fin 2 )− T (− 1 2 +δ + DΘ(− 1 Θ0 cosh (C )− b(cid:18)− 1 a =− 2 2W (cid:18)Θ0 W − D − δ 2 + δ(cid:19) . . . sinh(C )− b(cid:19) + δ W − D 2W Θ0 cosh(C ). (12) k D C 2 1 Finally, we can close the problem by evaluating the solution at x = −1/2, where we find T = 0, to obtain the value of Θ0. Note that Θ0 is the temperature difference between the fins at x = 0. The symmetry of the two opposing fins allows us to write Θ0 = 2(T (x = 0)− 1/2), where 1/2 represents (Th + T c)/2. We then find a = 1/2, from which one can obtain Θ0 more easily using Eq. 12. Θ0 is therefore a measure for the amount of cooling in the system: in the isothermal limit we find Θ0 = −1. We now arrive at the following equation for the temperature field: T = −(x + 1 Eq. 10 1− (x + 1 2 ) W−D C 2W (cid:0)Θ0 2W (cid:0)Θ0 2 sinh (C )− b(cid:1) 2 sinh (C )− b(cid:1) C 2 ) W−D 2 : − 1 : − 1 : − 1 (cid:54) x < − 1 2 + δ 2 + δ (cid:54) x < 1 2 − δ (cid:54) x < 1 2 2 − δ (13) whose dimensional form can be obtained easily. A similar procedure can be fol- lowed to obtain T . Heat flux From an application perspective, the quantity of interest is the (dimen- sional) heat exchanged per frontal unit area. We find for the width of a single pair of fins: (14) (15) 2W q(cid:48)(cid:48) = or in dimensional units 2 W q(cid:48)(cid:48) = ∆ T k D kg L D C sinh (C ) + 2(W − D)cosh (C )(cid:19) , Θ0(cid:18) 4 cosh (C )(cid:19) , Θ0(cid:18) 4 L(cid:17) + 2 W− D 2 W DΘ0(cid:16)4·(cid:16) 1 W − D L sinh (C ) + 2 kg L C from which q(cid:48)(cid:48) can be obtained. It is interesting to expand Eq. 15 in the limit of C → 0 L (cid:17) +O(C2) + . . .. as it reduces correctly to Eq. 1: q(cid:48)(cid:48) = ∆ T 2 − δ 4 Results The analytical model is first validated using numerical solutions of the problem using the commercial package Comsol [22]. The steady state heat equation was solved by a finite element method on a triangular grid. We consider one half of a pair of oppos- ing fins inside a large stack, avoiding any edge effects. The hot and cold faces have 6 Figure 2: Numerical solution for D = 0.0625, δ = 0.22 and W = 0.14. Results for a low conducting gas of k = kg/ks = 10−5 is displayed in panel (a), whereas panel (b) was solved with k = 10−1, resulting in C = 0.9 and 9.1 resp. The coloured lines indicate isotherms between T = 0 (blue) and 1 (red). Dirichlet boundary conditions where the longitudinal boundaries satisfy the no-flux condition. The solid-gas boundaries satisfy continuation of heat flux and a no-jump condition in temperature. Two cases are presented in Fig. 2 for C = 0.9 and 9.1, where the gas conductivity is varied. It can clearly be seen that the isotherms in Fig. 2(a), shown by the coloured lines, lay mostly within the gas gap. Strong gradients appear in the panel (b) however, as expected since C > 1. In the centre, the temperature difference between the two pro- files Θ0 is added. We find that for isothermal fins it is close to unity, while decreasing when significant gradients occur in the film since the conductivity of the gas increases. We test now how trustworthy C is in predicting the occurrence of non-isothermal fins. We vary the parameters W , L, D, ks and kg independently and study the behaviour of Θ0, which is presented in Fig. 3. Despite through which parameter C is changed, the behaviour of Θ0 is the same. We clearly find that for C < 0.1 the fins are isothermal, whereas for C = [0.1,1] it only decreases by a fifth at most. Increasing C further results quickly in a decrease in effectiveness where from C ≈ 10 no significant temperature difference between the fins is found. The behaviour of Θ0 was accurately fitted using a single fitting parameter b: Θ0 = (16) 1 2(cid:0)erf(cid:0)log10(cid:0)C2(cid:1) + b(cid:1)− 1(cid:1) , as presented in Fig. 3, where we found b = −0.7. So far we have focussed purely on numerical results, so let us now focus on com- paring the solutions to the model developed earlier in section 3. From the numerical solution for C = 2.9 and δ = 0.22 we obtain the profiles in the hot and cold fin, as shown in Fig. 4. Focussing on the the blue temperature profile T of the cold fin, we use Eq. 13, plotted in purple, yellow and green. Although the agreement looks good for all 7 Figure 3: Temperature difference Θ0 between the two fins. Rescaling the parameters using C reveals a single curve for which Eq. 16 was fitted (black line). Figure 4: Comparison between numerical solutions of T (blue) and T (red) and the analytical model for C = 2.9. Θ0 was evaluated to be −0.4. 8 pieces, we quantify this by comparing the following properties: first we compare the difference in C and Θ0 by fitting the temperature difference of the numerical solution using Eq. 9. We compare the fit then with the expression for C and Θ0. Secondly we compare the difference of the fin profiles between the numeric- and analytical model using the relative error e f in = 1 ∆x 2 ¯Tnum + ¯TEq.6 (17) ∑ i (cid:12)(cid:12)Tnum − TEq.6(cid:12)(cid:12) , where the sum is taken over all positions i in the fin, being spaced by ∆x and the bar representing the mean temperature of the fin. For an application perspective, the error in the heat flux between the plates is more relevant, which can be expressed as qnum/qEq.10. We now vary the conductivity ratio k ≡ kg/ks as well as the slenderness of the fins W /L, while keeping the ratio between W and D constant. Figure 5 shows the performance of the model varying k = [10−5,10−1] and L = [3.2,32] mm. The numerical data was evaluated in the centre of the hot and the cold fin after which the difference was fitted by Eq. 9 with C and Θ0 as fitting parameters. For large C, the data was fitted in log-space for numerical stability. The results are presented in Fig. 5, where the error in C is the ratio between C (as defined in section 3) and the fit. We obtain good agreement over the complete phase space. The error in Θ0 was obtained by dividing the numerical fitting parameter by the value from the analytical model developed in section 3. We find good agreement for the parameter space where Θ0 > −0.05. The analytical model however starts to underestimate Θ0 for C > 20. The deviation seems to be more strongly related to the cooling, i.e., high k then the slenderness L/(W − D) of the fins. The two panels on the right show the error in the heat flux, deviating only more than 5% for Θ0 > −0.05 as well. Evaluating Eq. 17 shows that the mean difference between the numerical temperature profiles and the analytical model is smaller than 3% for the complete phase space studied. The good agreement found between the analytical model and the numerical results demonstrate the robustness as well as the accuracy of our one dimensional model. 5 Discussion In the previous section we developed and validated an analytical model to calculate the temperature profile in the fins of a heat switch and gave predictions for the heat flux. The parametric study for D, L,W, kg and ks shows that non-isothermal effects can be understood by the use of C. The fit Eq. 16 was able to capture the behaviour of Θ0 up to a single fitting constant: b. We found that the mesh generated for numerical study became too coarse for L > 8, whose results were therefore neglected in the fitting procedure. The fit is of great value in a quick evaluation of the heat exchange across the heat switch: Equation 15 can now be evaluated by the system parameters, ∆T and Eq. 16. Let us briefly comment here on the fit parameter b. Additional simulations of the problem made clear that b is a function of the thickness of the material δ. By fixing the fin geometry and varying C through k, we found that b(0) = −0.4 for δ → 0 and more 9 Figure 5: The accuracy of the analytical model was tested for L = [3.2,32] mm, δ = 0.7 mm, W = 0.45 mm and D = 0.2 mm, from which C (top left panel) and Θ0 (top middle panel) are presented. Errors are calculated using the methods in the text. Values of −Θ0 < 0.05 are left of the green lines, whereas the gray area denotes errors exceeding 1.3. general b ≈ −0.42 1− 1.7δ , (18) valid for δ = [0,15]. This results in a shift of the function along the C- axis since C is based on the complete domain. Basing C on the fin length does however not eliminate the δ dependency of b completely as the thickness δ of solid material itself acts as an insulating layer. For practical applications therefore, one wants to minimize the thickness δ for this reason and use b ≈ −0.4(1 + 1.7δ) as an approximation of Eq. 18 for moderate δ. From the validation study presented in Fig. 5 of the analytical model we obtain good agreement in the regions of significant exchange between the fins. First of all, the fitted value for C and Θ0 agrees within a few percent with the analytical expression (see Eq. 7), as well as the deviations in the temperature profiles. We encounter only strong deviations for C > 50 or k > 0.05. In those situations however, the effectiveness of the fins is greatly reduced anyway: The length of the fins act in fact as an insulating layer, negating the effect of the additional surface area is in this way. Our model performs thus best in the parameter space, that is relevant to the problem. We speculate that deviations occur by two reasons: First, for constant C, less slender fins suffer from cooling in the perpendicular directions and the one-dimensional model becomes less accurate. Secondly, for increasing k, the isotherms in the gap become less parallel to the side walls of the fin, see Fig. 2b. As a result, the flux lines −k∇T are no longer normal to the wall, which is not included in the analytical model. Despite these two 10 −5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)CNumerical020406080100−5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)Θ0Numerical−1−0.8−0.6−0.4−0.20−5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)ErrorinC0.970.9811.02−5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)ErrorinΘ011.11.21.3−5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)Errorinheatux0.80.911.05−5−4−3−2−10.81.11.41.72log10(k)log10(cid:16)LW−D(cid:17)Errorinntemperature0.2123·10−2effects the performance of the analytical model is still good. Finally we compare our model with the experimental measurements performed in our lab at ambient [20] and cryogenic [23] temperatures. Let us briefly elaborate on the relevant details of the setup. The heat switch is constructed out of a sintered titanium alloy (Ti6Al4V grade 5), having 49 rows of interleaving fins. The switch operates between the 'On'-state, where a high heat transfer is required, and the 'Off'-state by removing the gas in the gas gap. Although the gas conductivity is at most weakly dependent on the ambient pressure for bulk values, it can drastically be reduced as a result of the confinement by the fins. For pressures below 100 Pa the gas must be treated as a Knudsen gas. Krielaart et al. [20] studied helium, nitrogen and hydrogen gas at room temperature (Tm = 294 K), whereas Vanapalli et al. [23] studied helium and nitrogen at cryogenic conditions. For the cryogenic measurements the setup was operating at an mean temperature of 117.5 K, resulting in a bulk lower conductivity of the gas, as well as that of the solid compared to the ambient case. As shown in both references, proper modelling of the full experimental setup re- quired the inclusion of a contact resistance at both sides of the heat switch, see for instance [24]. Graphite foil of 350 µm and thermal paste were added to provide good thermal contact with the segments controlling the hot and cold boundary conditions [23]. To obtain good agreement with the model, the total contact resistance was found to be around 1/(2× 103) K/W, which can be estimated using (cid:96)/kc. Here kc = 0.2 is the conductivity of the thermal paste (Apizon N) and (cid:96) ≈ 100 µm the typical layer thickness. This estimate is reasonable as it represents the roughness of the heat switch, which was made out of sintered titanium alloy grains having a diameter of 100 µm. Since graphite is a good thermal conductor it does not contribute to the contact resis- tance. The comparison of the experimental measurements with the analytical model is presented in Fig. 6. Both the helium and hydrogen measurements show good agreement with the model, whereas the largest discrepancy was found for nitrogen measurements, being approximately 10%. The thermal contact resistance resulted in the reduction of the total heat transfer of 40% for the highest gas conductivities measured: indeed, for those cases the thermal resistance of the heat switch becomes of the same order as the contact resistance, resulting in a significant temperature drop across the thermal paste layer. 6 Conclusion and Outlook place. We found that the non-dimensional parameter C =(cid:16) 4kg L2 We found solutions to the interleaving fins where the fins may develop thermal gra- dients. The performance of the fin strongly depend whether or not this effect takes characterizes this well, collapsing numerical solutions of the problem when varying C through the different parameters over six orders of magnitude. For C < 1 thermal gradients may be neglected and one should use Eq. 1. In the other cases, we developed and tested an analytical model which showed excellent agreement with both numerical solutions to the problem for a large parameter space as well as with experiments. Our work offers ks D( W− D)(cid:17)1/2 11 Figure 6: Comparison of the model (solid lines) with experiments. The conductivity ratio k is varied by reducing the pressure in the heat switch, for which the system is subject to Knudsen effects. analytical solution to find the heat flux of this type of heat exchanger and gives new in- sights, essential in designing and optimizing such systems once the thermal properties are known. While we studied the steady state behaviour, let us finally briefly discuss the ex- pected behaviour of the transient operation. When the system of interleaving fins is used in a heat switch for switching between 'on' and 'off' states, the corresponding cooling and heating of the fins themselves become relevant. Intuitively, one can expect the off state to be in the isothermal regime. Recovery from an previous 'on' state is thus determined by heat diffusing from the hot or cold side to the tip of a fin. The (dimensional) diffusive timescale is then τrec = L2 ρ cp/ks, where ρ and cp are the den- sity and specific heat of the solid. When the switch is turned 'on' again, heat is first exchanged locally, for which we use the thermal timescale for transient heat transfer: τtrans = ks ρ cp/(kg/ D)2 [21]. It is interesting to look at the ratio between the two for which we find (kg/ks · D/L)2. We find then that this gives a ratio of time-spans for cycling between on and off states, provided that the fluid-flow time scale is fast. This is not always ensured as viscous effect can start to play a role for small length scales. This trade off is important if one is interested in a transient operation of the fins. 12 References [1] W. Stautner, R. Chen, A. Comment, and E. Budesheim, "An efficient liquid he- lium / gas-gap switch allowing rapidly servicing low-temperature dynamic nu- clear polarization systems," IOP Conference Series: Materials Science and Engi- neering, vol. 502, p. 012162, apr 2019. [2] M. Medrano, M. Yilmaz, M. Nogu´es, I. Martorell, J. Roca, and L. F. Cabeza, "Experimental evaluation of commercial heat exchangers for use as pcm thermal storage systems," Applied energy, vol. 86, no. 10, pp. 2047 -- 2055, 2009. [3] A. Castell, M. Belusko, F. Bruno, and L. F. Cabeza, "Maximisation of heat trans- fer in a coil in tank pcm cold storage system," Applied energy, vol. 88, no. 11, pp. 4120 -- 4127, 2011. [4] E. M. Alawadhi and C. H. Amon, "Performance analysis of an enhanced pcm thermal control unit," in ITHERM 2000. The Seventh Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (Cat. No. 00CH37069), vol. 1, pp. 283 -- 289, IEEE, 2000. [5] M. M. Farid, A. M. Khudhair, S. A. K. Razack, and S. Al-Hallaj, "A review on phase change energy storage: materials and applications," Energy conversion and management, vol. 45, no. 9-10, pp. 1597 -- 1615, 2004. [6] G. Bonfait, I. Catarino, J. Afonso, D. Martins, M. Linder, and L. Duband, "20k energy storage unit," Cryogenics, vol. 49, no. 7, pp. 326 -- 333, 2009. [7] G. Schneider, B. Niemann, P. Guttmann, D. Rudolph, and G. Schmahl, "Cryo x-ray microscopy," 1995. [8] G. Schneider, "Cryo x-ray microscopy with high spatial resolution in amplitude and phase contrast," Ultramicroscopy, vol. 75, no. 2, pp. 85 -- 104, 1998. [9] R. Ross, "Cryocooler reliability and redundancy considerations for long-life space missions," in Cryocoolers 11, pp. 637 -- 648, Springer, 2002. [10] D. C. Bugby, J. T. Farmer, B. F. OConnor, M. J. Wirzburger, E. D. Abel, and C. J. Stouffer, "Two-phase thermal switching system for a small, extended dura- tion lunar surface science platform," in AIP Conference Proceedings, vol. 1208-1, pp. 76 -- 83, AIP, 2010. [11] D. S. Glaister, D. G. T. Curran, V. N. Mahajan, and M. Stoyanof, "Application of cryogenic thermal switch technology to dual focal plane concept for brilliant eyes sensor payload," in 1996 IEEE Aerospace Applications Conference. Proceedings, vol. 2, pp. 115 -- 127, IEEE, 1996. [12] S. Gross, "Thermal coupling of equipment by interleaving fins," Journal of Space- craft and Rockets, vol. 7, no. 4, pp. 489 -- 491, 1970. [13] I. Catarino, G. Bonfait, and L. Duband, "Neon gas-gap heat switch," Cryogenics, vol. 48, no. 1-2, pp. 17 -- 25, 2008. 13 [14] W. R. Hamburgen, J. S. Fitch, and R. A. Eustace, "Interleaved-fin thermal con- nector," Aug. 4 1998. US Patent 5,787,976. [15] F. Romera, D. Mishkinis, A. Kulakov, and A. Torres, "Control of lhp operation temperature by a pressure regulating valve," in 15th Int. Heat Pipe Conf., Clem- son, USA, pp. 25 -- 30, 2010. [16] J. Burger, H. Ter Brake, H. Holland, R. Meijer, T. Veenstra, G. Venhorst, D. Lozano-Castello, M. Coesel, and A. Sirbi, "Long-life vibration-free 4.5 k sorp- tion cooler for space applications," Review of scientific instruments, vol. 78, no. 6, p. 065102, 2007. [17] P. J. Shirron and M. J. Di Pirro, "Passive gas-gap heat switch for adiabatic de- magnetization refrigerator," Nov. 1 2005. US Patent 6,959,554. [18] M. J. Di Pirro and P. J. Shirron, "Heat switches for ADRs," Cryogenics, vol. 62, pp. 172 -- 176, 2014. [19] S. Vanapalli, B. Colijn, C. Vermeer, H. Holland, T. Tirolien, and H. J. M. ter Brake, "A passive, adaptive and autonomous gas gap heat switch," Physics pro- cedia, vol. 67, pp. 1206 -- 1211, 2015. [20] M. A. R. Krielaart, C. H. Vermeer, and S. Vanapalli, "Compact flat-panel gas-gap heat switch operating at 295 K," Review of scientific instruments, vol. 86, no. 11, p. 115116, 2015. [21] A. Bejan, Heat Transfer. New York: John Wiley @ Sons, Inc., 1st ed., 1993. [22] COMSOL AB, COMSOL Multiphysics v. 5.2. Stockholm, Sweden. [23] S. Vanapalli, R. Keijzer, P. Buitelaar, and H. J. M. ter Brake, "Cryogenic flat-panel gas-gap heat switch," Cryogenics, vol. 78, pp. 83 -- 88, 2016. [24] M. M. Yovanovich, "Effect of foils upon joint resistance: Evidence of optimum thickness," AIAA Progress in Astronautics and Aeronautics, Thermal Control and Radiation, vol. 3, p. l, 1973. 14
1906.04718	3	1906	2019-11-06T01:46:32	Skyrmion-electronics: Writing, deleting, reading and processing magnetic skyrmions toward spintronic applications	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The field of magnetic skyrmions has been actively investigated across a wide range of topics during the last decades. In this topical review, we mainly review and discuss key results and findings in skyrmion research since the first experimental observation of magnetic skyrmions in 2009. We particularly focus on the theoretical, computational and experimental findings and advances that are directly relevant to the spintronic applications based on magnetic skyrmions, i.e. their writing, deleting, reading and processing driven by magnetic field, electric current and thermal energy. We then review several potential applications including information storage, logic computing gates and non-conventional devices such as neuromorphic computing devices. Finally, we discuss possible future research directions on magnetic skyrmions, which also cover rich topics on other topological textures such as antiskyrmions and bimerons in antiferromagnets and frustrated magnets.	physics.app-ph	physics	arXiv:1906.04718 [physics.app-ph] Topical Review Skyrmion-Electronics: Writing, Deleting, Reading and Processing Magnetic Skyrmions Toward Spintronic Applications Xichao Zhang,1 Yan Zhou,1 Kyung Mee Song,2 Tae-Eon Park,2 Jing Xia,1 Motohiko Ezawa,3 Xiaoxi Liu,4 Weisheng Zhao,5 Guoping Zhao,6 and Seonghoon Woo7, † 1School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen, Guangdong 518172, China 2Center for Spintronics, Korea Institute of Science and Technology (KIST), Seoul 02792, Korea 3Department of Applied Physics, University of Tokyo, Hongo 7-3-1, Tokyo 113-8656, Japan 4Department of Electrical and Computer Engineering, Shinshu University, Wakasato 4-17-1, Nagano 380-8553, Japan 5Fert Beijing Institute, BDBC, and School of Microelectronics, Beihang University, Beijing 100191, China 6College of Physics and Electronic Engineering, Sichuan Normal University, Chengdu 610101, China 7IBM Thomas J. Watson Research Center, Yorktown Heights, New York 10598, USA †E-mail: [email protected] (Dated: November 5, 2019) DOI: 10.1088/1361-648X/ab5488 Page 1 of 80 arXiv:1906.04718 [physics.app-ph] ABSTRACT The field of magnetic skyrmions has been actively investigated across a wide range of topics during the last decades. In this topical review, we mainly review and discuss key results and findings in skyrmion research since the first experimental observation of magnetic skyrmions in 2009. We particularly focus on the theoretical, computational and experimental findings and advances that are directly relevant to the spintronic applications based on magnetic skyrmions, i.e. their writing, deleting, reading and processing driven by magnetic field, electric current and thermal energy. We then review several potential applications including information storage, logic computing gates and non-conventional devices such as neuromorphic computing devices. Finally, we discuss possible future research directions on magnetic skyrmions, which also cover rich topics on other topological textures such as antiskyrmions and bimerons in antiferromagnets and frustrated magnets. Keywords: spintronics, magnetism, magnetic skyrmion, magnetic antiskyrmion, racetrack memory, logic gates, bio-inspired computing (Some figures may appear in color only in the online journal) Page 2 of 80 arXiv:1906.04718 [physics.app-ph] CONTENTS 1. Background ............................................................................................................................ 4 1.1. Topological Spin Textures .......................................................................................... 4 1.2. Magnetic Skyrmions at Low Temperature ............................................................... 11 1.3. Discovery of Room-Temperature Skyrmions ........................................................... 13 1.4. Potential Applications ............................................................................................... 17 2. Writing and Deleting Skyrmions ......................................................................................... 18 2.1. Magnetic Field .......................................................................................................... 18 2.2. Spin-Polarized Electric Current ................................................................................ 20 2.3. Local Electric Field ................................................................................................... 23 2.4. Laser .......................................................................................................................... 25 2.5. Imprinting ................................................................................................................. 27 3. Reading Skyrmions .............................................................................................................. 30 3.1. Microscopy Imaging of Magnetic Skyrmions: X-ray, TEM, SPM, MOKE ............. 30 3.2. Electrical Reading 1: Topological Hall Resistivity Measurements .......................... 33 3.3. Electrical Reading 2: Magnetic Tunnel Junction (MTJ)........................................... 35 4. Processing Skyrmions .......................................................................................................... 36 4.1. Current-Driven Dynamics of Magnetic Skyrmions .................................................. 36 4.2. Information Storage: Racetrack Memory and Skyrmion-MTJ ................................. 40 4.3. Information Computing: Transistor-Like Devices and Skyrmionic Logic ............... 42 4.4. Bio-Inspired Computing: Skyrmions for Neuromorphic Devices ............................ 44 5. Summary and Outlook ......................................................................................................... 45 5.1. Potential Novel Materials: Antiferromagnet, Synthetic Antiferromagnet, Ferrimagnet, Frustrated Magnet and 2D van der Waals Magnet ..................................... 45 5.2. Other Topological Spin Textures: Antiskyrmion, Skyrmionium, Biskyrmion, Meron, Antimeron, and Bimeron ..................................................................................... 49 5.3. Summary ................................................................................................................... 51 6. Acknowledgements .............................................................................................................. 52 7. References ............................................................................................................................ 53 Page 3 of 80 arXiv:1906.04718 [physics.app-ph] 1. Background 1.1. Topological Spin Textures Topology is an important discipline of pure and applied mathematics, which also plays a significant role in the understanding of many real-world physical phenomena. In the field of magnetism and spintronics, the concept of topology is particularly important to the physics of some exotic magnetic spin textures (i.e., the so-called magnetic solitons) [1-7], including different types of magnetic skyrmions and vortex-like spin textures, where the behaviors of textures are determined or affected by their topological characteristics. These topologically non-trivial spin textures can be one-dimensional (1D), two-dimensional (2D) and three-dimensional (3D) in real space, and usually carry integer or half-integer topological charges determined by their spin textures in the topological space [1-22]. For most cases, the well-studied topological spin textures are those localized in 2D or quasi-2D magnetic thin films and multilayers [11, 19, 21, 23-36], because that magnetic thin films and multilayers are better understood and preferred for building nanoscale spintronic applications. Several recent works have also demonstrated promising properties of 3D topological spin textures [10, 37-50], leading to the prediction that 3D spin textures might be used in combination with 2D textures in the future. Indeed, the 1D magnetic solitons have also been investigated in the field [9, 12-17, 51, 52], which revealed many fundamental properties of topologically non- trivial objects. For example, the chiral helicoidal soliton lattice was experimentally observed in a chiral helimagnet in 2012 [52], which was theoretically envisioned by I. E. Dzyaloshinskii in 1964 [15]. In this review, however, we will focus on and limit our discussion to 2D topological spin textures in magnetic materials. Figure 1 shows various exemplary topological spin textures that can be found in such 2D or quasi-2D magnetic materials. Note that quasi-2D system means the variation of spin texture along the thickness direction can be ignored, such as in the thin film with a finite thickness. The concept of skyrmion model was first proposed in the field of nuclear physics by T. H. R. Skyrme [53] in 1962. In 1989, Bogdanov and Yablonskiǐ for the first time predicted that chiral magnetic skyrmions can also be found in magnetic materials [16]. Then in 1994, Bogdanov and Hubert theoretically studied skyrmion structures in easy-axis magnetic materials with the Dzyaloshinskii-Moriya (DM) interaction [23, 54], where the DM interaction was found to be an important energy term to stabilize skyrmions, which will be discussed below. In 2001, Bogdanov and Rössler developed a phenomenological theory of chiral symmetry breaking in magnetic thin films and multilayers, and also predicted the existence of skyrmions in magnetic thin films and multilayers stabilized by induced DM Page 4 of 80 arXiv:1906.04718 [physics.app-ph] interactions [18]. In 2006, Rössler and coworkers further theoretically demonstrated that skyrmion structures can be formed as spontaneous ground states in magnetic metals with DM interactions without the assistance of external fields or the proliferation of defects [20]. Due to the topological nature, spin textures including chiral magnetic skyrmions can be regarded as localized quasi-particles in magnetic materials and show topology-dependent static and dynamic properties [1-7, 55]. In the context of 2D and quasi-2D systems, the topological structure of a spin texture is generally characterized by the topological charge [55-57], which is also referred to as the Pontryagin number [2, 3]: 𝑄 = ∫ 𝑑2𝒓 𝜌(𝒓), (1) where the topological charge density 𝜌(𝒓) reads 𝜌(𝒓) = 1 4𝜋 𝒎(𝒓) ∙ (𝜕𝑥𝒎(𝒓) × 𝜕𝑦𝒎(𝒓)). (2) Thus, the topological charge can be calculated based on the exact spin texture configuration according to 𝑄 = 1 4𝜋 ∫ 𝑑2𝒓 ∙ 𝒎(𝒓) ∙ (𝜕𝑥𝒎(𝒓) × 𝜕𝑦𝒎(𝒓)). (3) The topological charge basically counts how many times the reduced local magnetization (i.e., the spin) 𝒎(𝒓) wraps the 2D surface of a 3D ball in 3D space (i.e., the 2-sphere) as the coordinate (𝑥, 𝑦) spans the whole planar space. For example, as shown in Fig. 1, the Néel- type and Bloch-type skyrmions have topological charges of Q = -1 (for the core magnetization of -mz), while the antiskyrmion of the same core magnetization has the opposite topological charge of Q = +1 (often referred as the antiparticle of skyrmions with Q = -1). It is also called as "the skyrmion number" [2, 55] for a series of skyrmion-like spin textures that will be discussed throughout this review. In fact, chiral magnetic skyrmions are not limited to the case of topological charges Q = 1, and could be of any topological charge [24, 58-60]. For example, the skyrmionium can be regarded as a topological combination of a skyrmion with Q = +1 and a skyrmion with Q = -1, which carries a net topological charge of Q = 0. The skyrmionium structure was first studied in a theoretical work by Bogdanov and Hubert in 1999 [24]. It is also referred to as the target skyrmion [61]. The topological charge difference between the skyrmionium with Q = 0 and skyrmion with Q = +1 originates from their out-of-plane spin textures. A serious of theoretical works have suggested that skyrmioniums in magnetic materials can be manipulated by different external stimuli [11, 58, 62-73]. As other examples, the biskyrmion has a topological charge of Q = -2, which can be formed in some materials such as chiral bulk or frustrated magnets when two skyrmions with the same topological number (Q = -1 in this Page 5 of 80 arXiv:1906.04718 [physics.app-ph] case) are approaching to each other. The meron and bimeron have topological charges of Q = -0.5 and Q = -1, respectively, where the bimeron consists of a meron with Q = -0.5 and an antimeron with Q = -0.5. The bimeron with Q = -1 can be regarded as a counterpart of skyrmions with Q = -1 in magnetic materials with easy-plane magnetic anisotropy, where magnetization prefer to lie in the plane [74]. We will discuss these skyrmion-like structures in detail in Sec. 5.2. Obviously, these multifarious topological spin textures cannot be fully distinguished only using the topological charge Q, because there might be some degenerate states for a given value of the topological charge. Taking the skyrmion with Q = -1 as an example, there are additional degrees of freedom in the in-plane spin configuration. Therefore, one could calculate the vorticity number Qv and the helicity number Qh, which completely characterize the topological spin texture accompanying the topological charge Q. First, a point of the x-y space is parameterized as By applying the mapping 𝑟 = 0 as 𝑧 → −1 , 𝑟 = 1 as 𝑧 = 0 , 𝑟 → ∞ as 𝑧 → 1 , and 𝑥 = 𝑟 cos 𝜑, 𝑦 = 𝑟 sin 𝜑. (4) lim 𝑟→∞ 𝒎(𝑥, 𝑦) = lim 𝑧→1 parameterized by 𝒎(𝑧, 𝜙) [56], we map the x-y space onto the 2D surface of a 3D ball 𝑥 = √1 − cos2 𝜃 cos 𝜙, 𝑦 = √1 − cos2 𝜃 sin 𝜙, 𝑧 = 𝑧 = cos 𝜃, (5) with z being defined as 𝑟 = (1 + 𝑧) (1 − 𝑧) = ⁄ re-write the local magnetization direction as (1 + cos 𝜃) (1 − cos 𝜃) ⁄ . Therefore, we can 𝒎(𝒓) = 𝒎(𝜃, 𝜙) = (sin 𝜃 cos 𝜙, sin 𝜃 sin 𝜙, cos 𝜃). (6) By substituting Eq. (6) into Eq. (3), we obtain the topological charge determined by θ and ϕ, given as 𝑄 = 1 4𝜋 0 ∫ sin 𝜃 𝑑𝜃 ∫ 𝑑𝜙 𝜋 2𝜋 0 = − 1 4𝜋 [cos 𝜃]𝜋 2𝜋 0 [𝜙]0 . (7) From Eq. (7), it can be seen that the topological charge is actually determined by both the out-of-plane (θ) and in-plane (ϕ) spin textures. It should be noted that here we only consider the skyrmion solution where θ rotates π when r goes from zero to infinity. For the skyrmionium solution, θ rotates 2π when r goes from zero to infinity [24] Hence, in order to describe the in-plane spin texture, we define the vorticity number as 𝑄v = 1 2𝜋 ∮ 𝑑𝜙𝐶 = 1 2𝜋 𝜑=2𝜋 [𝜙]𝜑=0 . (8) One also needs to introduce the helicity number, which is the phase appearing in 𝜙 = 𝑄v𝜑 + 𝑄ℎ. (9) Page 6 of 80 arXiv:1906.04718 [physics.app-ph] Therefore, Eq. (6) can be as re-write as 𝒎(𝜃, 𝜑) = [sin 𝜃 cos(𝑄v𝜑 + 𝑄ℎ), sin 𝜃 sin(𝑄v𝜑 + 𝑄ℎ), cos 𝜃]. (10) Therefore, topological spin textures can be fully characterized by the three distinct numbers (Q, Qv, Qh). Here it is worth mentioning that the spin structures of chiral spin textures can also be described and observed in reciprocal or momentum space [8, 75-77], which allows diffraction-based techniques using e.g. neutrons to be effectively used to characterize the crystal of magnetic textures [75, 78, 79]. For example, the hexagonal chiral magnetic skyrmion lattice in real space can be described in reciprocal space as a superposition of three helical modulations (meaning three basis vectors of the crystalline structure) [77], which can be observed as six-fold scattering patterns in reciprocal space [75, 78, 80]. Figure 2 depicts some examples of degenerate skyrmion textures with varying Qv and Qh, where the basic skyrmion textures with the topological charge Q = 1 with different in- plane magnetization rotational senses are shown. For example, for the Néel-type skyrmion with a spin-up (i.e. pointing along the +z direction) and spin-down edge (i.e. pointing along the -- z direction), its in-plane spins can point toward or away from the skyrmion core. For the Bloch-type skyrmion, its in-plane spins can form a circle in either a clockwise or counterclockwise fashion. However, for the antiskyrmion with Q = 1, although its helicity number can vary between 0 and 2π, the spin texture actually has a two-fold rotational symmetry with respect to the skyrmion core. So, the antiskyrmion structure is effectively determined by its vorticity number [81]. It is noteworthy that skyrmions and antiskyrmions carry positive and negative Qv, respectively. The topological spin textures with different topological structures build a large family and lead to the emerging field of topological magnetism, which promises new opportunities for magnetic and spintronic applications. As the topological spin textures are non-collinear spin textures, their existence in magnetic materials is usually a result of delicate interplay among different energy terms. From the viewpoint of micromagnetism at zero temperature, the energy terms for common magnetic materials include the Heisenberg exchange interaction, dipolar interaction energy terms as well as magnetic anisotropy and Zeeman energy terms. The dominated Heisenberg exchange interaction favors the parallel or antiparallel alignment of the adjacent magnetic spins in most cases, so that it does not stabilize any non-collinear spin texture. However, the strong emergence of various other energy terms that may naturally exist in or be introduced to the magnetic materials can lead to the stabilization of non-collinear topological spin textures. In particular, the most important energy term, which has been studied for many years, is the asymmetric exchange interaction, i.e. the above mentioned DM interaction [82, Page 7 of 80 arXiv:1906.04718 [physics.app-ph] 83]. The DM interaction favors a right angle between the adjacent magnetic spins, thus a delicate competition between the Heisenberg exchange and DM interactions could result in the formation of non-collinear domain wall structures and topological spin textures with a fixed rotation fashion (called "chirality"), such as the chiral magnetic skyrmion and magnetic helical state. It is noteworthy that the spin-helix length in magnetic materials with DM interactions is determined by 4𝜋𝐴/𝐷, where A is the Heisenberg ferromagnetic exchange constant. The homogeneous DM interaction can exist in bulk materials lacking inversion symmetry and is expressed as 𝐸bDM = 𝑑bDM ∑ 〈𝑖,𝑗〉 𝑢𝑖𝑗 ∙ (𝒎𝑖 × 𝒎𝑗) , (11) where <i, j> denotes the nearest-neighbor sites, mi and mj are the reduced magnetic spin vectors at sites i and j, respectively. dbDM is the bulk DM interaction coupling energy, uij is the unit vector between mi and mj. In the micromagnetic model [11], the bulk DM interaction reads 𝐸bDM = 𝑏 ∬ 𝐷bDM[𝒎 ∙ (∇ × 𝒎)]𝑑2𝒓, (12) with DbDM being the continuous effective bulk DM interaction constant, and b being the magnetic film thickness. The bulk DM interaction could result in the stabilization of Bloch- type skyrmions with (Q = +1, Qv = +1, Qh = π) and (Q = -1, Qv = +1, Qh = 0). On the other hand, the DM interaction can also be induced at the interface between an ultrathin magnetic film and a non-magnetic film with a large spin-orbit coupling (SOC), given as 𝐸iDM = 𝑑iDM ∑ (𝑢𝑖𝑗 × 𝑧) ∙ (𝒎𝑖 × 𝒎𝑗) 〈𝑖,𝑗〉 , (13) where diDM is the interface-induced DM interaction coupling energy, and 𝑧 is the normal to the interface determined by Moriya's rule [83], oriented from the large SOC material to the magnetic film. In the micromagnetic model [11], the interface-induced DM interaction reads 𝐸iDM = 𝑏 ∬ 𝐷iDM[𝑚𝑧(𝒎 ∙ ∇) − (∇ ∙ 𝒎)𝑚𝑧]𝑑2𝒓, (14) with DiDM being the continuous effective interface-induced DM interaction constant. The interface-induced DM interaction could result in the stabilization of Néel-type skyrmions with (Q = +1, Qv = +1, Qh = π) and (Q = -1, Qv = +1, Qh = 0). Note that the structures of skyrmions stabilized by DM interactions are not only limited to Bloch- and Néel-types, as recent reports have reported the presence of intermediate type of skyrmions [84-87] stabilized by other DM interactions. Also, the Dresselhaus SOC in Page 8 of 80 arXiv:1906.04718 [physics.app-ph] materials with bulk inversion asymmetry can generate a type of DM interactions, which stabilizes antiskyrmions instead of skyrmions [88]. Indeed, the topological spin textures can also be stabilized by other mechanisms such as the frustrated exchange interactions [89], four-spin exchange interactions [8] as well as the long-range dipolar interactions [90-93]. However, most theoretical and experimental studies in the field focused on the systems with DM interactions during the last decade, as the DM interaction can be induced and adjusted by using modern interface engineering and multilayer fabrication techniques [2, 3, 5-7, 94-101]. As such, for the rest of this review, we will focus on the magnetic skyrmion textures stabilized by DM interactions, while other mechanisms will also be briefly reviewed in the last section. It is worth mentioning that, although conventional magnetic bubbles can be stabilized by the competition between ferromagnetic Heisenberg exchange and long-range dipolar interactions. The in-plane spin texture (i.e., helicity and chirality) of these common bubbles are not fixed, so that they cannot be treated as a topological object in the strict context of topological magnetism. For the last decade, these topological spin textures have been significantly highlighted, mainly inspired by their potential to handle information in devices at low energy consumption and/or high processing speed. In particular, the study of magnetic skyrmions for the purpose of designing novel spintronic applications has led to an emerging research field called skyrmion-electronics [1-3], which is also referred to as skyrmionics in some contexts [5]. It is an important task to study and understand how to stabilize and manipulate these topological spin textures, to eventually realize spin texture-based spintronic applications, such as memories, logic computing elements and transistor-like functional devices. In the following, we will review the representative topological spin texture -- the magnetic skyrmion, from the points of views of writing, deleting, reading and processing magnetic skyrmions toward spintronic applications. Page 9 of 80 arXiv:1906.04718 [physics.app-ph] Figure 1. Illustrations of a series of 2D and 3D topological spin textures in magnetic materials. (a) Néel-type skyrmion (Q = -1), (b) Bloch-type skyrmion (Q = -1), (c) antiskyrmion (Q = +1), (d) biskyrmion (Q = -2), (e) vortex (Q = -0.5), (f) meron (Q = -0.5), (g) bimeron (Q = -1), (h) skyrmionium (Q = 0), (i) skyrmion tube, and (j) magnetic bobber. The arrow denotes the spin direction and the out-of-plane spin component (mz) is represented by the color: red is out of the plane, white is in-plane, and blue is into the plane. Figure 2. Illustrations of 2D magnetic skyrmions with different topological charge, vorticity number and helicity number, i.e. (Q, Qv, Qh). The arrow denotes the spin direction and the out-of-plane spin component (mz) is represented by the color: red is out of the plane, white is in-plane, and blue is into the plane. Page 10 of 80 arXiv:1906.04718 [physics.app-ph] 1.2. Magnetic Skyrmions at Low Temperature In the history of magnetism, many new phenomena and magnetic structures were first discovered at low temperature due to the low transition temperature of magnetic materials, where the thermal fluctuations are largely reduced. Likely, ten years ago in 2009, Mühlbauer et al. for the first time experimentally observed the lattice structure of magnetic skyrmions in the chiral itinerant-electron magnet MnSi at a low temperature about 29 K [75] using a neutron scattering measurement [see Fig. 3(a)]. The lattice of magnetic skyrmions, so-called the skyrmion lattice, is a hexagonal ordered array of skyrmions and often referred to as the skyrmion crystal. As shown in Fig. 3(b), in 2010, such a hexagonal skyrmion crystal was directly observed in a thin film of crystalline Fe0.5Co0.5Si at a low temperature of 25 K by Yu et al. [102], where the real-space imaging of magnetic spin textures was acquired by using the Lorentz transmission electron microscopy (LTEM). Using the same technique, in 2011, Yu et al. reported the formation of the skyrmion crystal in FeGe over a wide range of temperature ranging from 60 K to 260 K [103], which is very close to room temperature. In 2012, Yu et al. further demonstrated the current-induced motion of the skyrmion crystal in a FeGe thin film at a near room temperature, ranging from 250 K to 270 K [104]. In 2013, the transition of a skyrmion crystal phases to other magnetic textures (including helimagnetic, conical and ferromagnetic phases) in Fe1−xCoxSi (x = 0.5) was also observed by using the magnetic force microscopy (MFM) at a very low temperature of 10 K [105]. Stabilization of skyrmions in such non-centrosymmetric crystalline materials is known to originate from the bulk-form DM interaction due to the broken inversion symmetry as discussed in Sec. 1.1. However, as was discussed in earlier Sec. 1.1 of this article, the existence of magnetic skyrmions turns out to be possible not only in ferromagnetic bulk metals, but also ultrathin films where the interface-oriented DM interaction is harnessed. In 2011, Heinze et al. experimentally revealed the 2D square skyrmion crystal in a hexagonal monolayer crystalline Fe film grown on the Ir surface using the spin-polarized scanning tunnelling microscopy (SP- STM) at a low temperature of 11 K [8]. Such a topological spin texture stems from the interplay among the Heisenberg exchange, four-spin and DM interactions. In such interface- oriented DM interaction-governed system, Romming et al. for the first time experimentally realized both the writing and deleting of individual skyrmions in a PdFe bilayer on Ir(111) at a low temperature of 4.2 K [see Fig. 3(c)], where skyrmions are controlled by local spin- polarized currents from a STM. At the time, an out-of-plane magnetic field of several Tesla (e.g., B = ~3 T) is required to stabilize individual skyrmions at low temperatures [106]. It is worth to mention that Jonietz et al. [78] and Schulz et al [107] also reported the ultralow Page 11 of 80 arXiv:1906.04718 [physics.app-ph] threshold current, 106 A m-2, for depinning magnetic skyrmions at ~26 K, which is 4-5 orders smaller than the depinning threshold of ferromagnetic domain walls, and these demonstrations excited great interest on using skyrmions for low power spintronic applications together with earlier findings. However, as skyrmions driven by the extremely low depinning current were displaced at a relatively low velocity [1, 107, 108], the required current density for actual devices may be increased as skyrmions need to move with a fast velocity. At low temperatures, the hosting materials of skyrmions are not limited to conventional metallic ferromagnets, and other kinds of topological magnetic textures can also be stabilized. In 2012, Seki and colleagues [109] for the first time identified skyrmions in an insulating multiferroic magnet Cu2OSeO3 through LTEM and magnetic susceptibility measurements below the temperature of 60 K [see Fig. 3(d)]. By utilizing the small-angle neutron scattering (SANS), Seki et al. also found a triangular skyrmion lattice in multiferroic insulator Cu2OSeO3 [110]. In 2016, Matsumoto et al. directly observed skyrmion domain boundaries in FeGe1−xSix (x ~ 0.25) using the differential phase contrast scanning transmission electron microscopy (DPC STEM) equipped with a segmented annular all-field detector at a temperature of 95 K [111], where they found that individual skyrmions at the domain boundary cores can flexibly change their size and shape to stabilize their core structures. The flexibility of the skyrmion structure at domain boundaries could be a basis for creating twisted skyrmion [112] that can be used for applications. Here it is worth mentioning that the DPC STEM technique can directly and precisely observe electromagnetic structures at an ultra-high resolution down to 1 nm. In 2016, McGrouther et al. revealed the internal spin structures of skyrmions in a freestanding nanowedge specimen of cubic B20 structured FeGe by using the DPC STEM technique but equipped with a pixelated detector [113]. In 2018, also by using the DPC STEM with a pixelated detector, McVitie et al. imaged the Néel-type domain wall structure and precisely measured the domain wall width [114]. These early stage experimental observations of skyrmions at low temperatures mainly focused on the existence of skyrmion crystals, skyrmion chain, skyrmion cluster, and individual isolated skyrmions in diverse material platforms stabilized by either bulk or interfacial DM interactions. Although these findings have proved fascinating physical properties of skyrmions and provided in-depth physical understanding behind their topological characteristics, from the viewpoint of practical applications, most commercial electronic devices based on skyrmions would require their room-temperature stabilization (or Page 12 of 80 arXiv:1906.04718 [physics.app-ph] at even higher temperature). In the following, we will review the discovery of skyrmions and relevant topological spin textures at room temperature. Figure 3. Observation of magnetic skyrmions at low temperature. (a) Magnetic phase diagram of MnSi. For B = 0, helimagnetic order develops below Tc = 29.5 K. Reprinted with permission from [75]. Copyright © 2009 AAAS. (b) A hexagonal skyrmion crystal observed using Lorentz TEM in a thin film of crystalline Fe0.5Co0.5Si at a low temperature of 25 K. Scale bar, 100 nm. Reprinted with permission from [102]. Copyright © 2010 Macmillan Publishers Limited. (c) Successive population of the island with skyrmions by injecting higher-energy electrons through local voltage sweeps. Reprinted with permission from [106]. Copyright © 2013 AAAS. (d) Skyrmion crystal in Cu2OSeO3 thin film obtained through the analysis of Lorentz TEM data taken at 5K. Reprinted with permission from [109]. Copyright © 2012 AAAS. 1.3. Discovery of Room-Temperature Skyrmions Between 2015 and 2016, there have been several reports of room-temperature stabilized magnetic skyrmions. The most notable demonstration was the stabilization of Page 13 of 80 arXiv:1906.04718 [physics.app-ph] room-temperature skyrmions in the industry-relevant sputter-grown non-crystalline ferromagnetic heterostructures. This approach was motivated by the theoretical suggestion [9] and the experimental confirmation [13, 115] of homochiral Néel-type domain walls in such ultrathin ferromagnet-based asymmetric heterostructures at room temperature. In 2015, Jiang et al. not only observed stable room-temperature individual skyrmion bubbles in a Ta (5 nm)/CoFeB (1.1 nm)/TaOx (3 nm) trilayer grown by magnetron sputtering [see Fig. 4(a)], but also demonstrated the current-driven transformation of skyrmion bubbles from strip domains in such a trilayer device via a geometrical constriction [116]. It should be noted that the term 'skyrmion bubble' studied in this report is a topologically nontrivial object with a skyrmion number of Q = 1 and a fixed Néel-type chirality induced by the DM interaction. However, unlike the compact skyrmion with 1-100 nm diameter, the skyrmion bubble has a larger size (~ 1 micrometer) comparable to trivial achiral bubbles stabilized by dipolar fields [91, 117- 119]. Both skyrmion bubble and compact skyrmion are topological spin textures and turned out to be equivalent in the topological definition [see Eq. (3)], but they are often differentiated with two different names -- skyrmion bubble or compact skyrmion -- in some contexts due to their significant size difference. Soon after this report, in 2016, some other groups subsequently reported the observation of room-temperature skyrmions in similar ultrathin ferromagnetic asymmetric heterostructures. Moreau-Luchaire et al. fabricated the Ir/Co/Pt asymmetric multilayers [29], where an additive interface-induced DM interaction of about 2 mJ m-2 was achieved, and using the scanning X-ray transmission microscopy (SXTM), sub-100 nm individual skyrmions were imaged in the asymmetric multilayers at room temperature and low magnetic fields (e.g., Bz = 58 mT) [see Fig. 4(b)]. As the thermal stability of magnetization increases with the volume of the magnet, the skyrmions in the ten repetitions of the Ir/Co/Pt films are stable against thermal fluctuations at room temperature [29]. As shown in Fig. 4(c), Boulle et al. also observed stable skyrmions in sputtered ultrathin Pt/Co/MgO nanostructures at room temperature and zero external magnetic field [120]. Based on the in-plane magnetization sensitive X-ray magnetic circular dichroism photoemission electron microscopy (XMCD- PEEM) experiments, it was found that the skyrmions in the Pt/Co/MgO thin films have the left-handed chiral Néel structure [120], which indicates they are stabilized by interface- induced DM interactions. In the same year, Woo et al. reported the observation of stable room-temperature skyrmions and their current-induced motion in [Pt (3 nm)/Co (0.9 nm)/Ta (4 nm)]15 and [Pt (4.5 nm)/CoFeB (0.7 nm)/MgO (1.4 nm)]15 multilayer stacks by using both the full-field Page 14 of 80 arXiv:1906.04718 [physics.app-ph] magnetic transmission soft X-ray microscopy (MTXM) and STXM [121]. Woo et al. demonstrated the stabilization of skyrmion lattice in a confined circular magnetic disk of about 2-µm diameter [see Fig. 4(d)], and further reported the motion of a train of individual skyrmions at a speed up to 100 m s-1 driven by a current density of j ~ 5 × 1011 A m-2 [121]. This is a direct experimental evidence that suggests skyrmions can be used for real-world high-speed spintronic applications. Later in 2016, as shown in Fig. 4(e), Yu et al. also experimentally demonstrated the creation of room-temperature skyrmion bubbles in ultrathin CoFeB films [122], where skyrmion bubbles are stabilized by a delicate competition among the Heisenberg exchange interaction, dipolar interaction, and perpendicular magnetic anisotropy (PMA). The skyrmion bubbles found in this work have a skyrmion number of Q = 1 and typical size close to 1 micrometer. Yu et al. reported that, although the DM interaction strength is only of 0.25 mJ m-2 in the CoFeB films, it is strong enough to fix the Néel-type chirality of skyrmion bubbles, leading to the Néel-type chiral nature. In 2017, Soumyanarayanan et al. [31] found a tunable room-temperature skyrmion platform in the Ir/Fe/Co/Pt multilayer stacks for studying sub-50-nm skyrmions [see Fig. 4(f)], where one can adjust the magnetic interactions governing skyrmions by varying the ferromagnetic layer composition. In 2018, Zhang et al. experimentally realized the control of skyrmion density in Pt/Co/Ta multilayer stacks by varying temperature [123] or in-plane magnetic field [124]. The existence of room-temperature magnetic skyrmions were also demonstrated using non-centrosymmetric crystalline interfaces and bulk materials. As shown in Fig. 4(g), Chen et al. imaged the room-temperature magnetic skyrmions in the Fe/Ni/Cu/Ni/Cu(001) multilayers using the spin-polarized low-energy electron microscopy (SPLEEM) [125], where the exchange coupling across non-magnetic spacer layers is responsible for the stabilization of skyrmions. A new class of cubic chiral magnets hosting room-temperature skyrmion crystals -- β-Mn-type Co-Zn-Mn alloys, was found by Tokunaga and colleagues [126]. The room- temperature magnetic skyrmion crystal was also realized in an artificial manner, as demonstrated by Gilbert et al. via patterning asymmetric magnetic nanodots with controlled circularity on an underlayer with PMA [127]. Such an approach is theoretically predicted by Sun et al. [128], which will be discussed in Sec. 2 as a possible method for writing skyrmions. As mentioned in Sec. 1.1, magnetic skyrmions can be stabilized in different materials by different mechanisms. The room-temperature skyrmions in asymmetric ferromagnetic multilayer stacks are usually stabilized by enhanced interface-induced DM interactions. In 2017, Hou et al. [129] for the first time observed the skyrmion bubbles with variable Page 15 of 80 arXiv:1906.04718 [physics.app-ph] topological charges at room temperature in the frustrated kagome Fe3Sn2 magnet with uniaxial magnetic anisotropy, by using in-situ LTEM. In the same year, antiskyrmions with Qv = -1 were experimentally identified in acentric tetragonal Heusler compounds Mn1.4Pt0.9Pd0.1Sn by Nayak et al. [88]. These two works not only provide more material platforms for hosting magnetic skyrmions and other topological spin textures, but also prove that topological spin textures with different topological charges can be stabilized at room temperature, shedding light on possible future applications based on different topological spin textures. Figure 4. Observation of magnetic skyrmions at room temperature. (a) The current-induced generation of skyrmion bubbles in Ta (5 nm)/Co20Fe60B20 (1.1 nm)/TaOx (3 nm). Reprinted with permission from [116]. Copyright © 2015 AAAS. (b) A 1.5×1.5 µm2 out-of-plane magnetization (mz) map obtained by STXM on a (Ir/Co/Pt)10 multilayer with applying out-of- plane magnetic field of 68 mT. Reprinted with permission from [29]. Copyright © 2016 Macmillan Publishers Limited. (c) XMCD-PEEM image of a 420 nm square dot (indicated by the dotted line) which is patterned with Ta (3 nm)/Pt (3 nm)/Co (0.5 -- 1 nm)/MgOx/Ta (2 nm) thin film. Reprinted with permission from [120]. Copyright © 2016 Macmillan Publishers Limited. (d) An initial labyrinth domain state was generated by static field (first image) and then transformed into a hexagonal skyrmion lattice in a 2µm Pt/Co/Ta disc by applying a bipolar pulse train with Vpp = 10V (second image). The last two images were acquired after applying Vpp =4V and Vpp = 5V, respectively. Dark (light) contrast corresponds to up (down) magnetization in all STXM images except for the last three, where the X-ray magnetic circular dichroism (XMCD) contrast was inverted. Reprinted with permission from [121]. Copyright © 2016 Macmillan Publishers Limited. (e) Polar-MOKE images of CoFeB thin film with PMA field of Hk ≈ 1.1 kOe for 4.8 Oe out-of-plane external magnetic field. Page 16 of 80 arXiv:1906.04718 [physics.app-ph] Reprinted with permission from [122]. Copyright © 2016 American Chemical Society. (f) Microscopic imaging (scale bar, 0.5 µm) of sample Fe (3)/Co (6) (number of atomic layers in braces) with MFM at ~ −100 mT of applied field. Reprinted with permission from [31]. Copyright © 2016 Macmillan Publishers Limited. (g) The out-of-plane SPLEEM images of Fe (2.6 ML)/Ni (2 ML)/Cu (8.6 ML)/Ni (15 ML)/Cu (001) multilayer structures (1 ML ~ 1.8 Å). Reprinted with permission from [125]. Copyright © 2015 AIP Publishing. 1.4. Potential Applications The field of skyrmion-electronics mainly focuses on the design and development of information storage and processing devices [2, 3, 5-7, 130]. The most highlighted skyrmionic information storage device -- skyrmionic racetrack memory is proposed by Fert et al. in 2013 [1], which is an improved design based on the domain-wall racetrack memory proposed by Parkin et al. in 2008 [131]. Later in 2013, Iwasaki et al. [108, 132] and Sampaio et al. [133] independently studied the current-induced dynamics of isolated skyrmions in ferromagnetic racetracks, which provided fundamental insight into the design of skyrmion-based racetrack memory. Since 2013, a number of theoretical and experimental works were carried out on the skyrmion-based racetrack memory and its derived applications, however, the experimental demonstration of such fully functional electrically operating skyrmion-based memory device is awaiting its demonstration. In 2013, the magnetoelectric resonance effect of magnetic skyrmions was found by Mochizuki et al. [134] and Okamura et al. [135], which suggests an opportunity for building the skyrmion-based microwave applications, such as microwave detector [136]. Besides, the skyrmion-based nano-oscillators [137-139] can be built using the oscillation characteristics induced by currents. In 2015, Zhang et al. [140] theoretically proposed the prototype of skyrmion-based logic computing gates, which can perform the logic AND and OR operations. Zhang et al. [141] also proposed a transistor-like functional device based on skyrmions, where the current-induced motion of skyrmions is controlled by a gate voltage. Note that a similar idea on the control of skyrmion dynamics by a gate voltage was also proposed by Upadhyaya et al. [142] in the same year. These potential skyrmion-based information computing devices can be combined with the skyrmion-based racetrack memory devices, which may lead to the invention of an in-memory logic computing circuit based on skyrmions. Some bio-inspired applications based on skyrmions were also proposed in recent two years. Huang et al. [143] designed a skyrmion-based artificial synapse device for Page 17 of 80 arXiv:1906.04718 [physics.app-ph] neuromorphic computing systems in 2017, which was recently experimentally realized by Song et al. [144]. Li et al. [145] also proposed a skyrmion-based artificial neuron that mimics the leaky-integrate-fire function of a biological neuron. In 2018, Bourianoff et al. [146] and Prychynenko et al. [147] proposed the first reservoir computing model based on magnetic skyrmions. Most recently, the thermally activated dynamics of skyrmions was investigated by several groups and found to be useful for computing [148-159]. For example, the thermal skyrmion diffusion-based signal reshuffling device operation was experimentally realized by Zázvorka et al. [160], which could be used in future skyrmion-based probabilistic computing devices. Besides, Nozaki et al. experimentally realized the control of the thermal diffusion of skyrmions by applying a pure voltage [149]. For the successful demonstration of any skyrmion-based potential applications, the energy-efficient writing, deleting, reading and processing processes of skyrmions are prerequisites. In the following, we review the progresses that have been made over the last 10 years in the writing, deleting, and reading of magnetic skyrmions. We will also review several representative skyrmion-based applications in detail, including the skyrmion-based racetrack memory, logic computing gate, transistor-like device and bio-inspired applications. 2. Writing and Deleting Skyrmions Magnetic skyrmion is a promising block for building future spintronic applications, as it can be employed as a non-volatile information carrier in magnetic media. The simplest method to encode binary information using magnetic skyrmions in magnetic materials is based on the writing and deleting of individual isolated skyrmions. Namely, the presence of a skyrmion can stand for the binary information digit "1" or "0". Hence, the controllable and reliable creation and annihilation of skyrmions are prerequisites for any skyrmion-based information storage applications. It is a vital task to find the best method to write and delete skyrmions. 2.1. Magnetic Field Magnetic field is one of the most important and controllable external stimuli that can be rather easily realized in most laboratories. In the early stage of skyrmion research, a number of theoretical and experimental studies were carried out on the magnetic phase diagram, where skyrmion crystals and isolated skyrmions usually form in chiral magnets within the range of certain out-of-plane magnetic fields and temperature [54, 75, 78, 102, 103, 105, 109, 161-163]. That is to say, by applying an external magnetic field Page 18 of 80 arXiv:1906.04718 [physics.app-ph] perpendicularly to the magnetic film with DM interactions, it is possible to control the formation of skyrmion textures [106]. Some of the initial demonstrations of room- temperature skyrmions introduced in Sec. 1.1 were also realized by shrinking maze stripe domains into circular bubble-shaped domains by applying magnetic field. On the other hand, once the skyrmion is created in a magnetic film, its size also depends on the strength of the external out-of-plane magnetic field [11, 164, 165] that interplays with other energetic contributions e.g. DM interaction, magnetic anisotropy, demagnetization energy and exchange interaction. Namely, a small out-of-plane magnetic field is usually required for the formation of skyrmions, while a large out-of-plane magnetic field could lead to the collapse and annihilation of skyrmions. In 2016, Müller et al. predicted and numerically demonstrated that, by applying magnetic field pulses, skyrmions can be created close to the edge of a chiral magnet due to the edge instability [166]. In 2017, M. Mochizuki [167] demonstrated that in a magnetic film with a fabricated hole or notch, individual skyrmions can be created by an external magnetic field in a controlled manner [see Fig. 5(a)]. Most recently, the writing of skyrmions in a uniformly magnetized film by a magnetic dipole was theoretically investigated by Garanin et al. [168], which can be seen as the creation of skyrmions by a localized magnetic field. In 2018, such a creation method of writing skyrmions was experimentally demonstrated by Zhang et al. using a scanning local magnetic field provided by magnetic force microscopy (MFM) tips [169]. Once the magnetic skyrmion is created, it can also be manipulated or even deleted by magnetic fields, similar to the magnetic vortices in magnetic disks [170, 171]. For example, Büttner et al. showed that the gigahertz gyrotropic eigenmode dynamics of a single skyrmion bubble can be excited by an external magnetic field pulse and corresponding magnetic field gradient [172]. Magnetic skyrmions in nanostructures with confined geometries can also be switched by microwave magnetic fields [see Fig. 5(b)] [173] or magnetic field pulses [see Fig. 5(c)] [174, 175]. Note that the microwave magnetic field can also excited other dynamics of skyrmions [176, 177]. Indeed, when the strength of magnetic field is larger than certain threshold, it is destructive for skyrmions and may ultimately lead to the fully polarized magnetic state, i.e. the ferromagnetic state. Hence, the application of a strong magnetic field, either in-plane or out-of-plane, is also the easiest way to delete skyrmions from magnetic media. Page 19 of 80 arXiv:1906.04718 [physics.app-ph] Figure 5. Writing, deleting and switching of skyrmions using magnetic fields. (a) Schematic of skyrmion creation in a magnetic field application. Reprinted from [167]. With permission of AIP Publishing. (b) The skyrmion core reversal induced by an AC magnetic field pulse. Reprinted with permission from [173]. Copyright © 2015 AIP Publishing. (c) The skyrmion core annihilation process with applying an in-plane magnetic field pulse. The yellow symbols indicate the out-of-plane spins. Reprinted with permission from [174]. Copyright © 2014 Macmillan Publishers Limited. 2.2. Spin-Polarized Electric Current Spin-polarized electric current could provide spin-transfer torques (STTs) or spin- orbit torques (SOTs) on magnetic moments, which is an important driving force for magnetization dynamics in modern spintronic devices [97, 178, 179]. In 2011, the interaction between STTs and skyrmion lattices in chiral magnets was theoretically studied by Everschor et al. [180]. Then, the writing of a magnetic skyrmion using the spin-polarized electric current was theoretically predicted by Tchoe et al. in 2012, where skyrmions can be nucleated from a ferromagnetic background by a vertical current injection with the current density of the estimated order of j ~ 1010-1011 A m-2 [181]. In 2013, Iwasaki et al. [132] and Sampaio et al. [133] independently studied the current-induced skyrmion creation in confined geometries using theoretical and numerical approaches. In particular, Iwasaki et al. numerically demonstrated that skyrmions can be nucleated and created from the notch in a nanotrack [132], which is an effective method to write skyrmions at a certain location. In the same year, as was noted in Sec. 2.1, Romming et al. [106] for the first time experimentally realized both the writing and deleting of single isolated magnetic skyrmions by using the Page 20 of 80 arXiv:1906.04718 [physics.app-ph] local tunneling spin-polarized currents (~ 1 nA) from a STM at low temperature [see Fig. 6(a)]. In 2014, Zhou and Ezawa theoretically predicted that a magnetic skyrmion can be converted from a domain-wall pair in a junction geometry (j ~ 1.8×1012 A m-2) [182]. In a similar manner, in 2015, Jiang et al. demonstrated in a room-temperature MOKE experiments that magnetic skyrmion bubbles can be created by utilizing the spin current-driven (j ~ 0.5×1010 A m-2) transformation between strip domains and skyrmion bubbles in a Ta/Co20Fe60B20/TaOx trilayer with a junction constriction [116]. The creation of skyrmion bubbles in such a junction constriction driven by inhomogeneous spin currents was further theoretically studied independently by Heinonen et al. [183], S.-Z. Lin [184], and Liu et al. [185], confirming its efficient mechanism and device compatibility suitable for future applications. The spin-polarized current can also result in the creation of skyrmions or skyrmion bubbles by other different mechanisms. In 2016, Yuan and Wang theoretically demonstrated that a skyrmion can be created in a ferromagnetic nanodisk by applying a nano-second current pulse (j ~ 2.0×1012 A m-2) [186]. Yin et al. also suggested in a theoretical study that it is possible to create a single skyrmion in helimagnetic thin films using the dynamical excitations induced by the Oersted field and the STT given by a vertically injected spin- polarized current (j ~ 1.7×1011 A m-2) [187]. In 2017, as shown in Fig. 6(b), Legrand et al. experimentally realized the creation of magnetic skyrmions by applying a uniform spin current directly into nanotracks (j ~ 2.38×1011 A m-2) [32]. Latter in 2017, Woo et al. experimentally demonstrated the creation of skyrmions at room temperature and zero external magnetic field by applying bipolar spin current pulses (j ~ 1.6×1011 A m-2) directly into a Pt/CoFeB/MgO multilayer [33], which turned out to be thermally-induced skyrmion generation as systematically studied recently by Lemesh et al. [34]. Hrabec et al. also reported the experimental creation of skyrmions induced by applying electric current (j ~ 2.8×1011 A m-2) through an electric contact placed upon a symmetric magnetic bilayer system [188]. The spin-polarized current-induced generation of skyrmions can be deterministic and systematic when pinning sites or patterned notches, where PMA is locally reduced, are used as the source of skyrmion generation as reported by Buttner et al. [35] in 2017 (j ~ 2.6×1011 A m-2) [see Fig. 6(c)] and Woo et al. [36] in 2018 (j ~ 2.5×1010 A m-2) [Fig. 6(d)]. More recently, Finizio et al. reported that localized strong thermal fluctuation could also introduce systematic skyrmion generation at a designed location [189]. Page 21 of 80 arXiv:1906.04718 [physics.app-ph] On the other hand, the spin-polarized current can delete skyrmions in nanostructures. For example, in 2017, De Lucia et al. theoretically studied the annihilation of skyrmions induced by spin current pulses (j ~ 5×1012 A m-2) and suggested that skyrmions can be reliably deleted by designing the pulse shape [190]. In 2018, Woo et al experimentally demonstrated such deterministic deletion of a single skyrmion in ferrimagnetic GdFeCo films with the application of designed current pulses (j ~ 2.5×1010 A m-2) [36]. In Table 1, we summarized the typical current density applied to create and delete skyrmions in the above mentioned theoretical and experimental works for the purposes of reference and comparison. Figure 6. Writing and deleting of skyrmions using spin-polarized electric currents. (a) Creation and annihilation of single skyrmions with local spin-polarized currents at T = 4.2 K in the PdFe bilayer. Reprinted with permission from [106]. (b) MFM images of the 1 μm wide tracks before any pulses and after the injection of 1000 current pulses of 200 ns in 0.6 nm thick Co multilayers. Reprinted with permission from [32]. (c) Single-skyrmion generation and subsequent motion in Pt/CoFeB/MgO multilayers. Reprinted with permission from [35]. Copyright © 2017 Macmillan Publishers Limited. (d) Magnetic skyrmion configuration for writing (d1) and deleting (d2). Scale bar, 500 nm. Reprinted with permission from [36]. Page 22 of 80 arXiv:1906.04718 [physics.app-ph] Table 1. Writing and deleting of skyrmions using spin-polarized electric currents with different current densities in both theoretical and experimental works. Materials Current density Pulse width T (K) B (mT) Reference Experimental Writing PdFe bilayer 1~200 nA Ta/Co20Fe60B20/TaO 0.5×1010 A/m2 1 μs (single) Ta/Co/[Pt/Ir/Co/]10/Pt 2.38×1011 A/m2 200 ns (1000 pulses) [Pt/Co4Fe4B2/MgO]20 1.6×1011 A/m2 20 ns (3.33 MHz, 5s) [Pt/CoFeB/MgO]15 3.0×1011 A/m2 7.5 ns (10000 pulses) Pt/Ni/Co/Ni/Au/Ni/Co/Ni/Pt 2.8×1011 A/m2 7 ns (single) [Pt/Co60Fe20B20/MgO]15 2.6×1011 A/m2 12 ns (single) [Pt/Gd25Fe65.6Co9.4/MgO]20 2.5×1010 A/m2 10 ns (single) 4.2 RT RT RT RT RT RT RT 1800 0.5 20.5 0 40 6 6.1 130 [106] [116] [32] [33] [34] [188] [35] [36] [Pt/Gd25Fe65.6Co9.4/MgO]20 2.5×1010 A/m2 10 ns (single) RT 130 [36] Experimental Deleting CoPt CoFeB Theoretical Writing 1.8×1012 A/m2 1.0×1012 A/m2 40 ns (single) Co20Fe60B20 1.5~3.0×1010 A/m2 CoPt 2.0×1012 A/m2 0.7 ns (single) 1.7×1011 A/m2 3.6×1011 A/m2 Theoretical Deleting 0 0 0 0 50 0 0 0.5 70 0 [182] [183] [185] [186] [187] [132] CoFeB 5~20 ×1012 A/m2 0 0 [190] RT, room temperature. 2.3. Local Electric Field In the last decade, a number of studies have been performed to explore skyrmion dynamics induced by electric current, however, the electric current often accompanies Joule heating that leads to energy losses or even permanent damages to metallic devices. Therefore, it may be necessary to find more reliable ways to drive skyrmion dynamics, and in this Page 23 of 80 arXiv:1906.04718 [physics.app-ph] context, the pure electric field is one of the desired ways to drive skyrmion dynamics because of its ultralow power consumption with negligible Joule heating. A number of reports [109, 134, 135, 191, 192] had shown the possibility to control skyrmions in magnetoelectric materials, such as the chiral magnetic insulator Cu2OSeO3, using a pure electric field (due to the magnetoelectric coupling). In 2015, Mochizuki and Watanabe [193] theoretically proposed the writing of isolated skyrmions in a multiferroic thin film by applying a local electric field [see Fig. 7(a)], which is realized through the magnetoelectric coupling effect in multiferroic compounds. In 2016, the electric-field-controlled transition between the skyrmion lattice phase and conical phase was experimentally realized in magnetoelectric chiral magnet Cu2OSeO3 [194]. In ferromagnetic materials, the writing and deleting of magnetic skyrmions can also be induced by a pure electric field, usually based on the effect of voltage control of magnetic anisotropy (VCMA) [142]. In 2017, as shown in Fig. 7(b), Hsu et al. experimentally demonstrated the reversible transition between the ferromagnetic texture and the magnetic skyrmion by applying a local electric field to the Fe triple layer on Ir(111) [195]. Such a local electric field, which can induce the writing and deleting of magnetic skyrmions, were realized by using the SP-STM. In the same year, the writing and deleting of skyrmion bubbles controlled by electric field in a Pt/Co/oxide trilayer was realized in an efficient and reproducible manner by Schott et al. at room temperature [196]. In 2018, Srivastava et al. experimentally demonstrated the electric field turning of DM interactions in a Ta/FeCoB/TaOx trilayer [197], which can be used as a method to further control the skyrmion chirality. Huang et al. also experimentally realized the writing of skyrmions in the magnetoelectric compound Cu2OSeO3 [192]. In 2019, Ma et al. designed and fabricated a Pt/CoNi/Pt/CoNi/Pt multilayer, which is sandwiched between the indium tin oxide (ITO)/SiO2 bilayer and the glass substrate, in the form of racetracks where the thickness of the films had a slope [198]. In such a nanostructure, Ma et al. demonstrated that many skyrmion bubbles can be created (i.e., corresponding to the skyrmion writing) and directionally displaced about 10 micrometres by applying a pure electric field [see Fig. 7(c)]. When the electric field is removed, the skyrmion bubbles will be annihilated, which can be regarded as the deleting of skyrmion bubbles. Similar to the manipulations of skyrmions by magnetic field and electric current, the polarity of skyrmions can also be switched by a pure electric field, in principle. For example, Bhattacharya et al. numerically demonstrated the electric-field-induced switching of a magnetic skyrmion in a magnetic tunnel junction (MTJ) structure [199]. On the other hand, Page 24 of 80 arXiv:1906.04718 [physics.app-ph] the dynamics of skyrmions can also be controlled by a pure electric field, as experimentally demonstrated by Nozaki et al. in a W/FeB/Ir/MgO multilayer structure recently [149], the thermal motion of skyrmion bubbles can be controlled by a pure voltage. Figure 7. Writing and deleting of skyrmions by electric fields. (a) Schematic of skyrmion creation via local application of an electric field using an electrode tip and simulated spatiotemporal dynamics of the magnetizations for the electrical skyrmion creation process in the Cu2OSeO3 thin film. Reprinted with permission from [193]. (b) Perspective views of subsequent SP -- STM constant-current images of the same Fe triple layer area showing the writing and deleting of individual magnetic skyrmions. Reprinted with permission from [195]. Copyright © 2017 Macmillan Publishers Limited. (c) MOKE microscopy images of the EF-induced creation and motion of a chiral domain wall accompanied by the creation of skyrmion bubbles in Pt/CoNi/Pt/CoNi/Pt multilayer with the thickness gradient at Hz = −0.2 mT. Reprinted with permission from [198]. 2.4. Laser The magnetization dynamics can also be driven by (ultrafast) laser pulses, and laser pulse-induced magnetization excitation has been of great interest for many years. Such excitation enabled the observation of ultrafast demagnetization [200] and all-optical switching [201-204], which could enable the integration of optical spintronics devices. Single laser shot-induced deterministic magnetization switching was also demonstrated using ferrimagnetic thin films, where the ultrafast last-induced heating could lead to the angular momentum transfer between two sublattices [202]. Unlike ferrimagnetic materials, the injected laser-induced local heating in ferromagnetic materials probabilistically lead to the Page 25 of 80 arXiv:1906.04718 [physics.app-ph] switching of magnetization within the heated area due to the thermal perturbation and its interactions with e.g. local dipolar field [130, 205, 206]. In detail, the switching field (i.e. coercive field) of ferromagnets generally decreases with increasing temperature, therefore, the dipolar interactions that favor the formation of multi-domains overcome the switching field threshold, leading to the local magnetization switching. Nevertheless, in chiral magnetic materials, such laser-induced magnetization switching results in the local generation of a single or multiple chiral magnetic textures, including skyrmions. In 2009, Ogasawara et al. experimentally studied the magnetization reversal dynamics in a TbFeCo film induced by a femtosecond laser pulse in the presence of a small external magnetic field [207], where the laser-induced reversed magnetic domain has a size of about 1 micrometer. Such a circular magnetic domain, which is stabilized by the dipolar interaction, can be seen as a achiral skyrmion bubble [208]. In 2013, Finazzi et al. reported the creation of topological spin textures induced by ultrashort single laser pulses in an amorphous thin alloy film of Tb22Fe69Co9 without the help of an external magnetic field [209], where both skyrmion and skyrmionium textures are created [see Fig. 8(a)]. The lateral size of the laser-created magnetic skyrmions can be as small as 150 nm [209]. In 2017, Fujita and Sato theoretically studied the ultrafast creation of skyrmion and skyrmionium textures by applying a vortex laser beam [68], where the switching of magnetization in either a ferromagnetic or antiferromagnetic background is a result of the laser-induced non-uniform temperature. In 2018, Yang et al. numerically studied the manipulation of skyrmions by an all-photonic orbital angular momentum transfer mechanism [210]. In the same year, as shown in Fig. 8(b), the laser-induced writing and deleting of skyrmions in the prototypical itinerant chiral magnet FeGe was experimentally realized by Berruto et al. [211], where the writing and deleting speed was found to be controlled by the cooling rate following the laser-induced temperature increase. Je et al. also experimentally demonstrated the laser-induced creation of disordered hexagonal skyrmion bubble lattices from a ferromagnetic background in an ultrathin Ta/Fe72Co8B20/TaOx trilayer film at room temperature [see Fig. 8(c)], where the density of skyrmion bubbles was controlled by the laser fluence [206]. Such an experiment suggests that the skyrmion bubbles and bubble lattices can possibly be manipulated by lasers in a controlled manner. Apart from the writing and deleting of skyrmions, magnetic skyrmions can be excited to different modes by optical methods. For example, Ogawa et al. reported the laser-induced collective excitation modes of the skyrmion phase in an insulating chiral magnet Cu2OSeO3 Page 26 of 80 arXiv:1906.04718 [physics.app-ph] [212], including the rotation mode, breathing mode, and several additional spin precession modes. Figure 8. Writing of skyrmions by lasers. (a) Near-field Faraday rotation map showing magnetic domains induced in a thin TbFeCo film after single laser pulse irradiation, corresponding skyrmion spin textures, and profile (dots) of the Faraday rotation measured along the red lines. Reprinted with permission from [209]. (b) Lorentz-Fresnel micrographs of the FeGe nanoslab. The skyrmion lattice in #4 is created by near-IR fs laser pulses. Scale bar is 250 nm. Reprinted with permission from [211]. (c) Nucleated skyrmion bubbles from a saturated state (+z) by a 35-fs single laser pulse with a laser with linear polarization, right- handed σ+ polarization, and left-handed σ− polarization. Reprinted with permission from [206]. 2.5. Imprinting Skyrmion-like magnetic vortex domains, which is stabilized by geometrical constrains, can also be created by means of nanoimprinting even in the absence of DM interactions. As shown in Fig. 9(a), in 2013, Sun et al. proposed a method for the creation of a skyrmion lattice based on a combination of a perpendicularly magnetized CoPt film and nanopatterned arrays of magnetic vortices that are geometrically confined within Co Page 27 of 80 arXiv:1906.04718 [physics.app-ph] nanodisks [128]. Such a method is similar to the exchange spring model [213], where the spin textures in a soft phase (i.e., corresponding to the magnetic vortex in Co nanodisk) can be imprinted into the hard phase (i.e., corresponding to the CoPt film) due to the interlayer exchange coupling between the soft and hard phases. In principle, the created skyrmion lattice [see Fig. 9(b)] can be stabilized in a wide temperature and field range, even at room temperature, zero magnetic field, and in the absence of the DM interaction. In 2015, Fraerman et al. experimentally observed the creation of skyrmions in a perpendicularly magnetized Co/Pt multilayer film via the imprinting method [214], where the Co/Pt multilayer film is exchange-coupled with a Co nanodisk with the vortex state. In the same year, by using the transmission soft X-ray microscopy, Streubel et al. experimentally studied the dynamics of imprinted skyrmion textures and found that skyrmionium textures can also be created by imprinting a magnetic vortex from the soft Py layer into the hard Co/Pd layer in the presence of reasonable interlayer exchange coupling [215]. The imprinting skyrmion lattice with controlled circularity and polarity as ground state at room temperature was also experimentally realized by Gilbert et al. [127], where the magnetic vortex state in Co nanodots were imprinted into the Co/Pd underlayer with PMA [see Fig. 9(c)]. On the other hand, Del-Valle et al. theoretically demonstrated the possibility to imprint skyrmions in thin films by superconducting vortices [216]. In 2018, Sun et al. numerically studied the stability and skyrmion Hall effect (SkHE) of magnetic skyrmions in the exchange-coupled system [217], which was used for the creation of skyrmions via imprinting effect in previous reports. They found that the stability of the skyrmion state can be enhanced due to the coupling between skyrmions in the thin film and the vortices in the capping nanodisks. At the same time, the capping nanodisks also act as attracting centers so that the SkHE can be effectively suppressed. The SkHE is a phenomenon related to the topology of skyrmion structures [26, 218, 219], which may be detrimental for the in-line motion of skyrmions. We will discuss the SkHE in detail in Sec. 4. In this section, we have reviewed a series of methods for writing and deleting skyrmions in magnetic materials. From the viewpoint of practical applications, it is desirable to use the spin-polarized electric current to control the generation/deletion of skyrmions, which can be integrated into transistor-based conventional complimentary metal-oxide semiconductor (CMOS) scheme. Spin currents can be rather easily generated either by using a polarizer layer via spin-filtering effect or by using high spin-orbit coupling material with relatively large spin Hall effect (or interfaces with large Rashba effect), which have been extensively explored for the last few years. Such a heavy metal/ferromagnet heterostructure Page 28 of 80 arXiv:1906.04718 [physics.app-ph] are also known to generate certain DM interaction and PMA to stabilize skyrmions. However, the method utilizing electric excitation often suffers from inevitable current- induced Joule heating and leads to a large power consumption, therefore the electric field and laser could also be alternative methods for writing and deleting skyrmions in a low-energy manner in the future. Figure 9. Writing of skyrmions by imprinting. (a)-(b) The theoretical proposal for creating the 2D skyrmion crystal. (a1) Ordered arrays of magnetic submicron disks are prepared on top of a film with perpendicular anisotropy. The arrows represent the magnetization orientation of the local moments. (a2) Skyrmion lattice creation with the field treatment. (b) Top view of magnetic configuration of the CoPt layer in a calculated artificial skyrmion crystal. (a)-(b) are reprinted with permission from [128]. (c) Imprinted artificial skyrmion lattices at room temperature. (c1) Illustration of the hybrid structure consists of Co dots (red) on top of Co/Pd PMA underlayer (grey) where the in-plane spin texture of the Co dots (purple arrows) is imprinted into an irradiated Co/Pd region (light blue) underneath the dots (tilted blue arrows). Remanent-state (c1) MFM and (c2) SEMPA (superimposed onto a scanning electron microscopy image of the dots) images, after saturating the dots in an in- plane field parallel to the flat edge of the dots to the right, indicate circularity control. Scale bar, 2 mm. A key to the magnetization winding direction is shown in the insets. Reprinted with permission from [127]. Page 29 of 80 arXiv:1906.04718 [physics.app-ph] 3. Reading Skyrmions 3.1. Microscopy Imaging of Magnetic Skyrmions: X-ray, TEM, SPM, MOKE The manipulation of magnetic skyrmions as information carriers in device applications requires the precise tracking of their locations and dynamic behaviors. Hence, the efficient reading of skyrmions in real space is an important task that needs to be realized in a reliable manner with high enough on/off ratio. A lot of state-of-art microscopy magnetic imaging techniques have been used to observe the real-space profiles of skyrmions in laboratories for the purpose of studying their static and dynamic properties (see Table 2) [29, 33, 35, 36, 102, 106, 109, 111, 116, 120, 121, 125, 148, 160, 174, 195, 197, 220-233]. However, as a part of the information write-in/read-out system, electrical reading scheme is a prerequisite for any skyrmion-based information storage and computing device applications. Some methods were purposed to read and detect skyrmions in device applications, such as by using the magnetic tunnel junction (MTJ) or reading skyrmion-induced Hall voltages. In this section, we first review recent progress on the real-space imaging/reading of skyrmions in laboratories. The microscopy imaging of magnetic skyrmions in real space is very important for both theoretical and applied reasons. From the theoretical point of view, the imaging of the in-plane and out-of-plane spin textures in real space can directly identify the topological nature of a skyrmion. For example, one may calculate the topological charge based on the detailed in-plane spin texture of a skyrmion, including the vorticity number Qv and helicity number Qh. Moreover, some of techniques offer sub-ns temporal resolution along with real- space magnetization imaging, which plays an important role to understand the ultrafast dynamic properties of skyrmions, e.g. excitation modes. The real-space imaging of a 2D skyrmion lattice, which consists of many Bloch-type skyrmions, was first realized in a thin film of Fe0.5Co0.5Si by using the LTEM in 2010 [102]. Later, skyrmion lattices in thin films of FeGe [41, 103, 220-222], Cu2OSeO3 [109, 192, 234], MnSi [223, 235], Mn1-xFexGe [224], and Fe-Gd alloy [225] were also observed by using the LTEM as discussed in Sec. 1.2. The LTEM is a particularly useful tool for observing the magnetization lying in the film plane, e.g. Bloch-type skyrmions, as the Lorentz interaction is non-vanishing when the magnetic induction component is perpendicular to the electron beam, which will deflect the electron beam [236]. Although typical Frensel-mode LTEM measurement cannot directly observe Néel-type magnetic configurations at zone-axis mode due to the cancellation of magnetic inductions, the observation of Néel-type magnetic domains and skyrmions has also been demonstrated using LTEM by tilting sample plane Page 30 of 80 arXiv:1906.04718 [physics.app-ph] against beam direction or using off-focused images [237, 238]. More recently, as noted in Sec. 1.2. aberration-corrected DPC-STEM imaging technique has also been introduced [111, 239, 240] and used to directly image magnetic textures including skyrmions [111, 113]. The SP-STM can also be used to measure nanometer-scale magnetic textures (see Table 2). In 2013, Romming et al. observed nanoscale isolated skyrmions in a PdFe bilayer on Ir(111) by using SP-STM [106]. The field-dependent size and shape of single isolated skyrmions were then also identified by using the SP-STM [226]. The magnetic properties of nanoscale isolated skyrmions discovered in PdFe/Ir bilayers were also experimentally and theoretically investigated by Leonov et al. in 2016 [230]. In 2017, Hsu et al. further realized and observed the electric-field-induced switching of single isolated skyrmions in a Fe triple layer on Ir(111) using the SP-STM [195]. Recently years, various microscopy methods using X-ray radiation were used to as a powerful tool to image ultra-small spin textures and ultra-fast spin dynamics (see Table 2). In 2014, Li et al. observed the imprinted skyrmions by using the element-specific XMCD and PEEM measurements [174]. In 2016, Moreau-Luchaire et al. realized additive DM interaction in Ir/Co/Pt multilayers at room temperature, where isolated skyrmions with dimeters below 100 nm were directly observed by using the STXM [29]. In the same year, Boulle et al. observed room-temperature skyrmions in sputtered ultrathin Pt/Co/MgO multilayers by using the photoemission electron microscopy combined with XMCD-PEEM [120], which identified the Néel-type nature of the skyrmion chirality. Woo et al. also observed room-temperature skyrmions and revealed their current-driven dynamics in Pt/Co/Ta and Pt/CoFeB/MgO multilayers by using the high-resolution STXM [33, 121]. More recently, Woo et al. directly imaged ferrimagnetic skyrmions in GdFeCo films [227] and observed the writing and deleting dynamics of a ferrimagnetic skyrmion by using the time-resolved X-ray microscopy [36]. Indeed, skyrmion textures can also be directly observed by other methods. For example, by using the SPLEEM, Chen et al. imaged magnetic skyrmions in in-situ grown Fe/Ni bilayers in 2015 [125]. Moreover, X-ray holography offers a unique drift-free diffractive magnetic imaging characteristic with sub-10 nm spatial resolution (see Table 2), and was used to demonstrate ultra-small skyrmions and their current-driven dynamics [35, 241]. Recently, mainly inspired by easy accessibility, a number of experimental studies reported the direct observation of skyrmions by using the polar magneto-optical Kerr effect (MOKE) microscopy [116, 148, 160, 197, 228]. For example, the first observation of room temperature skyrmions and the creation of skyrmions from stripe domain walls was directly Page 31 of 80 arXiv:1906.04718 [physics.app-ph] imaged by using the MOKE microscopy [116]. The spatial resolution of MOKE microscopy is rather limited by the wavelength of the light used for the observation, which usually lead to a resolution of a micrometer (see Table 2). Table 2. List of the representative imaging techniques for the observation of magnetic skyrmions Methods Spatial Advantages Best used for Reference resolution LTEM ~ 1 nm High resolution Thin chiral magnets [102, 109, Easy access with nanoscale Bloch- 220-225] type skyrmions DPC-STEM ~ 1 nm High resolution Thin chiral magnets [111, 113] No post imaging process with nanoscale Néel/Bloch-type skyrmions SP-STM Single Ultrahigh resolution down Crystalline chiral [106, 195, atom to atomic scale magnets with 226] extremely small skyrmions XMCD-PEEM ~ 25 nm In-plane spin resolution Surface magnetic skyrmions [120, 174] STXM ~ 25 nm Easy magnetic/electric Skyrmions in buried [29, 36, excitation magnetic multilayers 227] Dynamics imaging (~70 accompanying ps) excitations SPLEEM ~ 10 nm In-plane spin resolution In-situ deposited [125] High surface sensitivity surface magnetic skyrmion X-ray ~ 10 nm Drift-free technique Nanoscale skyrmions [35] holography Dynamics imaging (~70 in buried magnetic ps) multilayers Page 32 of 80 arXiv:1906.04718 [physics.app-ph] MOKE ~ 1 µm Easy access and custom >1 µm skyrmions [116, 148, modification 160, 197, 228] LTEM, Lorentz transmission electron microscopy; DPC-STEM, differential phase contrast scanning transmission electron microscopy; SP-STM, spin-polarized scanning tunneling microscopy; XMCD-PEEM, photoemission electron microscopy combined with X-ray magnetic circular dichroism; STXM, scanning transmission X-ray microscopy; SPLEEM, spin polarized low-energy electron microscopy; MOKE, magneto-optical Kerr effect microscopy. 3.2. Electrical Reading 1: Topological Hall Resistivity Measurements The realization of reading skyrmions in an electrical manner is a crucial step toward the device applications based on skyrmions. A promising method to read skyrmions electrically is by harnessing the topological Hall effect (THE) of magnetic skyrmions [162, 242-244]. The THE arises from the Berry phase in a magnet with smoothly varying magnetization [245-247], which can occur due to the Berry curvature induced by magnetic skyrmions [162, 244], because conduction electrons must adiabatically obey the unique spin textures of magnetic skyrmions when the s-d coupling is strong [2]. Namely, a prominent feature of a skyrmion is that it produces emergent magnetic field, which originates from the solid angle subtended by the spins called scalar spin chirality. The topological property of skyrmion guarantees that the total flux generated by one skyrmion is one flux quantum, h/e. The size of a skyrmion is ~ 1 nm for atomic Fe layer on Ir(111) surface [8], ~ 3 nm for MnGe [244], ~ 18 nm for MnSi [243], and ~ 70 nm for FeGe [103]. The corresponding emergent magnetic field is ~ 4000 T, 1100 T, 28 T, and 1 T, respectively. In fact, the THE is a manifestation of emergent magnetic field, where the Hall effect occurs in the presence of skyrmions [107, 242-244]. Moreover, the quantized THE has been theoretically proposed [248]. In 2009, Neubauer et al. [242] experimentally studied the THE of the skyrmion crystal phase in MnSi, where a district anomalous contribution to the Hall effect in the skyrmion crystal phases was identified [see Fig. 10(a)]. Such a topological Hall contribution has an opposite sign to the normal Hall effect, and its prefactor is consistent with the Page 33 of 80 arXiv:1906.04718 [physics.app-ph] skyrmion density in a quantitative manner. In 2012, Schulz et al. experimentally measured the temperature-dependent the Hall resistivity in MnSi under different strengths of applied electric current [107], where the depinning and motion of skyrmions were confirmed by direct observation of the emergent electric field of skyrmions. In the same year, magnetic skyrmions in epitaxial B20 FeGe(111) thin films was also identified by measuring the topological Hall resistivity due to the skyrmion textures [162]. In 2014, Yokouchi et al. studied the stability of skyrmions in epitaxial thin films of Mn1−xFexSi with various thicknesses and compositions by measuring the topological Hall resistivity as functions of temperature and applied field [249]. The topological and thermodynamic stability of skyrmions in MnSi was also studied by measuring topological Hall resistivity [250]. In 2016, Matsuno et al. studied the interface DM interaction in epitaxial bilayers consisting of ferromagnetic SrRuO3 and paramagnetic SrIrO3 by measuring the THE [251], which demonstrated that the skyrmion phase is stabilized by interface-induced DM interaction. In 2017, Liu et al. demonstrated the creation and annihilation of skyrmion- induced THE in Mn-doped Bi2Te3 topological insulator films [252]. Although the electrical measurements induced by THE were mostly conducted with crystalline materials where the existence of skyrmion lattice provides large collective electrical signal, recent reports by Maccariello et al. [253] [see Fig. 10(b)] and Zeissler et al. [254] [see Fig. 10(c)] demonstrated the electrical Hall measurements of individual room temperature skyrmion in sputter-grown films and nanostructures. However, the two demonstrations report large difference in the effective magnitude of topological Hall contribution from a single room-temperature skyrmion, which may need further investigations. In 2019, Shao et al. also experimentally observed the THE at above room temperature in a bilayer heterostructure composed of a magnetic insulator (thulium iron garnet, Tm3Fe5O12) in contact with a metal (Pt). On the other hand, it is noteworthy that, in 2016, Hamamoto et al. theoretically proposed a method to detect the skyrmion position in a pure electrical way by measuring the Hall conductance in a constricted geometry [255], where the Hall conductance is found to have a peak when a skyrmion is located at the lead position, while it reduces when a skyrmion is away from the lead. Page 34 of 80 arXiv:1906.04718 [physics.app-ph] Figure 10. Electrical reading of magnetic skyrmions by using the THE and MTJ. (a) Topological Hall effect in the A phase of MnSi. (a1) Hall resistivity ρxy near Tc in the temperature and field range of the A phase. (a2) Additional Hall contribution ρxy in the A phase. Data are shifted vertically for better visibility. Reprinted with permission from [242]. Copyright © 2009 The American Physical Society. (b) Topological Hall effect in skyrmion phase of MnSi. (b) Variation in Hall due to a single skyrmion. Hall resistivity shows a change of 25 nΩ cm associated with the single-skyrmion formation. The insets are MFM images. Reprinted with permission from [253]. Copyright © 2018 Macmillan Publishers Limited, part of Springer Nature. (c) The normalized Hall resistance Rxy/Rxy,sat and the extracted Mz/Msat. The inset is the XMCD image of two skyrmions at -45 mT. Reprinted with permission from [254]. Copyright © 2018 Macmillan Publishers Limited, part of Springer Nature. (d) Electrical detection of magnetic skyrmions by tunnelling non-collinear magnetoresistance. (d1) The signal change caused by the non-collinear magnetoresistance dI/dU map of two skyrmions in PdFe/Ir(111); the inset presents a profile along the arrow. (d2) The dI/dU tunnel spectra in the centre of a skyrmion (red) and outside the skyrmion in the FM background (black). a.u., arbitrary units. Reprinted with permission from [256]. Copyright © 2015 Macmillan Publishers Limited. 3.3. Electrical Reading 2: Magnetic Tunnel Junction (MTJ) Recently, theoretical and experimental studies suggest the efficient large on/off ratio detection of skyrmions by measuring the magnetoresistance [256-259], i.e. by using and embedding MTJ sensors in skyrmion-based devices and circuits. Page 35 of 80 arXiv:1906.04718 [physics.app-ph] In 2015, Du et al. experimentally identified the formation of clusters of individual skyrmions by measuring the magnetoresistance in MnSi nanowires [229, 257], where the number of skyrmions was revealed by quantized jumps in the magnetoresistance curves. In the same year, Hanneken et al. [256] proposed a pure electrical reading scheme of nanoscale skyrmions by measuring the tunnelling non-collinear magnetoresistance [see Fig. 10(d)]. The electric conductance signal due to the non-collinear magnetoresistance is sensitive to the local magnetic environment, and can directly detect and distinguish the collinear and non- collinear magnetic states without using a magnetic electrode [256, 259]. Such a mechanism is different from the well-known giant magnetoresistance (GMR) [260], and tunnel magnetoresistance (TMR) effect [261], which are dependent of local magnetization directions in two magnetic layers. On the other hand, the electric reading of single isolated skyrmions in a current-perpendicular-to-plane geometry was also studied by Crum et al. from first principles [262], which is based on the effect of tunnelling spin-mixing magnetoresistance. In 2017, Tomasello et al. theoretically proposed a protocol for the electrical reading of a magnetic skyrmion by measuring the change of the TMR signal via a point-contact MTJ in a three-terminal device [258]. Similar method can also be used to read other magnetic solitons, such as magnetic bubbles, in racetrack-type devices [263]. Most recently, numerical simulations suggested that skyrmions can be directly created and detected in a MTJ with DM interaction in a pure electrical manner [264], which shows the possibility of designing the skyrmion-MTJ-based multibit storage and artificial neural network computation. In 2019, Penthorn et al. found that a single skyrmion with a diameter smaller than 100 nm can lead to a change in MTJ resistance of almost 10% [265], which is enough for electrical detection. In the same year, Kasai et al. also experimentally realized the electrical detection of skyrmions in MTJ [266], where the skyrmion diameter is about 200 nm. 4. Processing Skyrmions 4.1. Current-Driven Dynamics of Magnetic Skyrmions As briefly mentioned in Sec. 1.2, Jonietz et al. [78] observed the rotation of skyrmions arising from the interplay of current-induced spin-torque effects and thermal gradients. They found that the current density required to create observable STTs in skyrmion lattice phase of MnSi exceeds an ultralow threshold of ∼106 A m−2, which is over five orders of magnitude smaller than those typically applied in experimental studies on current-driven magnetization dynamics in nanostructures (∼1011 A m−2). These results indicate that the Page 36 of 80 arXiv:1906.04718 [physics.app-ph] current-driven motion of magnetic skyrmions could be of low-energy consumption. Therefore, the spin-polarized current serves as an effective method for driving and processing skyrmions. However, it should be noted here that the skyrmion velocity is proportional to the driving current density, and practical skyrmion-based applications may require a skyrmion speed much larger than that induced by the very low depinning current density. At the same time, the energy consumption of skyrmion-based application will increase with increasing magnitude of the current. In 2011, Everschor et al. [180] theoretically proposed that skyrmions can be driven by the spin-polarized current via STTs. The translational mode and rotational mode can be induced by STTs. Using the method of Thiele [267], they projected the magnetization dynamics equation, i.e. the Landau-Lifshitz-Gilbert (LLG) equation onto the translational mode to derive an effective equation of motion -- Thiele equation, where 𝑮 is the gyrocoupling vector and 𝓓 is the dissipative tensor. 𝒗𝒔 is the velocity of the 𝑮 × (𝒗𝒔 − 𝒗𝒅) + 𝓓(𝛽𝒗𝒔 − 𝛼𝒗𝒅) = 𝟎, (15) conduction electrons and 𝒗𝒅 is the drift velocity of skyrmions. In 2011, Zang et al. [26] considered the additional damping term in the LLG equation and theoretically proposed the skyrmion Hall effect (SkHE), where the skyrmion shows a transverse velocity when it is driven by the spin current. In 2013, the current-induced creation and motion of skyrmions are numerically demonstrated by Iwasaki et al. [108, 132] and Sampaio et al. [133] [see Fig. 11(a)]. Iwasaki et al. investigated the current-induced motion of skyrmions in the presence of geometrical boundaries [108, 132]. In a channel with a finite width, the transverse confinement results in steady-state characteristics of the skyrmion velocity as a function of current density, which are similar to those of domain walls in ferromagnets. Sampaio et al. [133] considered two cases. One is the case of spin-polarized current injected under a current-in-plane geometry. The skyrmion is driven by the adiabatic and non-adiabatic STTs, 𝝉adiab = 𝑢𝒎 × ( 𝜕𝒎 𝜕𝑥 × 𝒎) , 𝝉non−adiab = 𝛽𝑢 (𝒎 × 𝜕𝒎 𝜕𝑥 ), (16) where 𝑢 = 𝛾0ℏ 𝜇0𝑒 𝑗𝑃 2𝑀S with the reduced Planck constant ℏ, the electron charge e, the applied current density j, the spin polarization rate P, and the saturation magnetization MS. is the electron velocity and β is the non-adiabaticity factor. The other is the case driven by a vertical spin current, which can be obtained either by using a MTJ or by utilizing the spin Hall effect. The skyrmion is driven by the induced in-plane and out-of-plane torques, Page 37 of 80 arXiv:1906.04718 [physics.app-ph] 𝝉IP = 𝑢 𝑡 𝒎 × (𝒎p × 𝒎), 𝝉OOP = −𝜉 𝑢 𝑡 (𝒎 × 𝒎p), (17) where t is the film thickness and ξ is the amplitude of the out-of-plane torque relative to the in-plane one and 𝒎p is the current polarization vector. The numerical results obtained by Sampaio et al. [133] demonstrated that the efficiency of the vertical spin current to drive skyrmions is much higher than that of the in-plane current. In 2016, Woo et al. [121] experimentally demonstrated the stabilization of magnetic skyrmions and their current-driven motion in thin transition metal ferromagnets at room temperature, where a train of individual skyrmions can be driven into motion at speeds exceeding 100 m s−1 by current pulses (5 × 1011 A m-2) in the Pt/CoFeB/MgO multilayers track. In 2017, Jiang et al. [218] and Litzius et al. [219] directly observed the SkHE [see Fig. 11(b)] in room-temperature experiments where skyrmions are driven by the current-induced SOTs. Like the charged particles in the conventional Hall effect, current-driven skyrmions acquire a transverse velocity component. As shown in Fig. 11(c), Jiang et al. observed a linear dependence of the skyrmion Hall angle on the driving current density, which is possibly resulted by the pinning of skyrmions. They also changed the sign of the topological charge Q and the electric current, and thus, a strong similarity between the conventional Hall effect of the electronic charge was found. Litzius et al. also experimentally observed the dependency of the skyrmion Hall angle on the driving current density, while they suggested that this dependency is induced by the additional effect of the skyrmion deformation as well as the effect of the field-like SOT. In 2017, Legrand et al. [32] investigated the nucleation and spin-torque-induced motion of sub-100 nm skyrmions in Pt/Co/Ir trilayers, which enable additive DM interactions at the Pt/Co and Co/Ir interfaces [29]. In particular, they showed that such small compact skyrmions can be nucleated by applying a uniform current directly into tracks and subsequently be driven into motion via SOTs. In 2016, theoretical studies [155, 268] have predicted enhanced current-driven behaviors of antiferromagnetic skyrmions due to the elimination of SkHE. In 2018, Woo et al. [36] experimentally investigated the skyrmion in ferrimagnetic GdFeCo multilayers, where the stabilization of ferrimagnetic skyrmions and their current-driven dynamics were demonstrated. The distinctive nature is that the SkHE is not totally eliminated in the ferrimagnetic system since the magnetization are different between two antiferromagnetically coupled underlying sub-lattices. Indeed, they demonstrated that the current-driven ferrimagnetic skyrmion shows a small, but non-zero SkHE. They further confirmed that Page 38 of 80 arXiv:1906.04718 [physics.app-ph] ferrimagnetic skyrmions can move at a velocity of ~50 m s−1 with reduced skyrmion Hall angle, θSkHE ~20°. In 2019, Hirata et al. [269] experimentally demonstrated that the SkHE can be totally vanished at the angular momentum compensation temperature of a ferrimagnet [see Fig. 11(d)], i.e. when the ferrimagnet becomes antiferromagnet. Figure 11. (a) Skyrmion velocity v as a function of current density j for in-plane currents with different values of the non-adiabaticity parameter β (0.15, 0.30 and 0.60 in yellow, orange and brown lines and circles, respectively) and for vertical currents (blue line, squares for isolated skyrmion). Reprinted with permission from [133]. (b) Schematic of the skyrmion Hall effect. The spin textures of skyrmions are indicated by the arrows. Reprinted with permission from [218]. (c) Phase diagram of the skyrmion Hall angle as a function of current density/sign of topological charge obtained by tracking the motion of several tens of skyrmions. Reprinted with permission from [270]. (d) Current-driven elongation of magnetic bubble in GdFeCo/Pt films as a function of temperature. Reprinted with permission from [269]. Page 39 of 80 arXiv:1906.04718 [physics.app-ph] 4.2. Information Storage: Racetrack Memory and Skyrmion-MTJ In 2013, Fert et al. [1] proposed that the skyrmion can be used to build the racetrack memory, which is based on the design of domain wall-based racetrack memory [271]. In the skyrmion-based racetrack memory, the information can be coded by the presence and absence of skyrmions, as shown in Fig. 12(a). In 2014, Tomasello et al. [272] showed the technological advantages and limitations of the manipulation of Bloch-type and Néel-type skyrmions by spin currents generated in the ferromagnetic layer or in the heavy-metal substrate arising from the spin Hall effect. They found that the Néel-type skyrmion driven by the spin torques due to the spin Hall effect is a promising strategy for technological implementation of next-generation of skyrmion racetrack memories. In 2017, Yu et al. experimentally demonstrated a room-temperature skyrmion shift device [273], which could serve as a basis for building a fully functional skyrmion-based racetrack memory. In 2014, Zhou et al. numerically demonstrated a current-driven reversible conversion between a skyrmion and a domain-wall pair in a junction geometry [182]. The information encoded in domain walls can be transformed into skyrmions, and then read out by transforming skyrmions back to domain walls after a functional control of the skyrmions. Such a hybrid device has the potential to outperform domain-wall racetrack memory because that it combined advantages of domain walls and skyrmions for spintronic applications. In 2015, Zhang et al. [274] numerically investigated the effects of skyrmion-skyrmion and skyrmion-edge repulsions on the feasibility of skyrmion-based racetrack memory. They suggested that the practicable spacing between consecutive skyrmionic bits is determined by the DM interaction helix length, i.e. 4𝜋𝐴/𝐷, where A is the exchange constant and D is the DM interaction constant. Further, they demonstrated that by fabricating a notched tip at the end of the nanotrack is an effective and simple method to avoid the clogging of skyrmionic bits, which also enables the elimination of useless skyrmionic bits beyond the reading element. As discussed in Sec. 4.1, when the skyrmion is driven into motion by a spin current, the trajectory of skyrmion bends away from the driving current direction due to the topological Magnus force [108, 155, 275], which is always perpendicular to the velocity. Such a SkHE may create an obstacle for the transportation of skyrmions. Consequently, the skyrmions may be destroyed by touching the nanotrack edges. In 2016, Zhang et al. [276] theoretically proposed a antiferromagnetically exchange-coupled bilayer system, i.e. a synthetic antiferromagnetic bilayer nanotrack, in which the SkHE can be suppressed as the Magnus forces in the top and bottom layers are exactly cancelled. This system serves as a Page 40 of 80 arXiv:1906.04718 [physics.app-ph] promising platform where skyrmions can move in a perfectly straight trajectory even at an ultra-fast processing speed. In 2017, Tomasello et al. [277] investigated the performance of synthetic antiferromagnetic racetrack memory. They pointed out that two adjacent racetracks can be more closer as compared with the with single heavy metal/ferromagnet bilayer since the dipole interactions of the two antiferromagnetically coupled layers cancel each other and the disturbance is thus reduced significantly. In 2016, Kang et al. [278] also proposed a complementary skyrmion racetrack memory structure, in which the skyrmions can be selectively driven into two different nanotracks by using a voltage-controlled Y-junction. In such proposed structure, both data bits "0" or "1" are represented with the presence of a skyrmion, therefore improving the data robustness and clock synchronization. In 2018, a skyrmion-based multilevel device with tunneling magnetoresistance was proposed [264], in which a MTJ with stable intermediate states can be realized based on the skyrmionic states in the free layer [see Fig. 12(b)]. This design may also facilitate the electrical detection of skyrmions. Most recently, the deterministic electrical switching from a ferromagnetic state to a skyrmion spin texture in the free layer of a MTJ was experimentally realized by Penthorn et al. [265]. Kasai et al. also experimentally realized the electrical detection of skyrmions in MTJ [266]. Although fully functional skyrmion-based racetrack memory has not been experimentally demonstrated yet, integrating above mentioned findings in a single device scheme may enable the realization of such skyrmion-based information storage devices. Page 41 of 80 arXiv:1906.04718 [physics.app-ph] Figure 12. (a) The illustration of skyrmionic racetrack memory, in which the skyrmions exhibit the same velocity. The colour scale shows the out-of-plane component of magnetization, mz. Reprinted with permission from [1]. (b) Illustration of a skyrmion-based MTJ and Magnetization−current hysteresis loops obtained by simulating the switching of R = 60 nm MTJs with different D and stray field BS. Reprinted with permission from [264]. (c) The top-view of the nanotracks under different spin current density j as well as different voltage-controlled perpendicular magnetic anisotropy Kuv. Initial state: both the electric field and spin current are turned off; the skyrmion keeps its position on the left side of the nanotrack. Off state: both the electric field and spin current are turned on. The spin current drives the skyrmion moving toward the right, while the electric field, which results in the change of PMA in the voltage-gated region, leads to the termination of the skyrmion when it approaches the voltage-gated region. On state: the electric field is turned off but the spin current is turned on. The skyrmion driven by the spin current passes the voltage-gated region and reaches the right side of the nanotrack. Reprinted with permission from [141]. (d) Skyrmion logical AND operation. The skyrmion represents logical 1, and the ferromagnetic ground state represents logical 0. Left panel, the basic operation of AND gate 1 + 0 = 0. Middle panel, the basic operation of the AND gate 0 + 1 = 0. Right panel, the basic operation of the AND gate 1 + 1 = 1. (e) Skyrmion logical OR operation. The skyrmion represents logical 1, and the ferromagnetic ground state represents logical 0. Left panel, the basic operation of OR gate 1 + 0 = 1. Middle panel, the basic operation of the OR gate 0 + 1 = 1. Right panel, the basic operation of the OR gate 1 + 1 = 1. (d) and (e) are reprinted with permission from [140]. 4.3. Information Computing: Transistor-Like Devices and Skyrmionic Logic In 2015, Zhang et al. [141] proposed a skyrmion-based transistor-like functional device [see Fig. 12(c)], where a gate voltage can be used to switch on/off a circuit. The PMA in the gate region is locally controlled by an applied electric field due to the charge accumulations. For the ON state: the spin current is turned on but the electric field is turned off. The skyrmion driven by the spin current passes the voltage-gated region. For the OFF state: both the electric field and the spin current are turned on; the electric field changes the PMA in the voltage-gated region and creates an energy barrier, leading to the termination of the skyrmion when it approaches the voltage-gated region. Zhang et al. numerically demonstrated that the working conditions can be controlled by tuning the amplitude of applied electric field and spin current, while proving the scalability of this transistor-like Page 42 of 80 arXiv:1906.04718 [physics.app-ph] device. In the same year, a similar idea on the control of skyrmion dynamics by a gate voltage was also proposed by Upadhyaya et al. [142], which also suggests the transistor-like function of skyrmions. In 2017, Xia et al. [279] numerically demonstrated the skyrmion- based transistor-like functional device also can be driven by spin waves and the working conditions can be adjusted by the amplitude and frequency of spin waves, as well as the applied voltage on the gated region. Later in 2018, Zhao et al [280] also numerically demonstrated the antiferromagnetic skyrmion transistor-like device based on strain manipulation. In 2015, Zhang et al. [140] theoretically proposed that skyrmions can be reproduced and merged in junction geometries based on the reversible conversion between skyrmions and domain walls [182]. They demonstrated that the logic AND [see Fig. 12(d)] and OR [see Fig. 12(e)] operations can be realized based on the duplication and merging of skyrmions. In their proposed skyrmionic logic devices, binary digit "0" corresponds to the absence of a skyrmion and binary digit "1" corresponds to the presence of a skyrmion. There are two input branches and one output branch in the device. The logic AND operations, "0" + "0" = "0", "0" + "1" = "0", "0" + "1" = "0", "1" + "1" = "1", can be implemented in the designed geometry. For example, "1" + "0" means that there is one skyrmion in input A and no skyrmion in input B. Under the driving force provided by the spin current, the skyrmion in branch A moves toward the output side. The skyrmion is first converted into a domain wall pair and the domain pair fail to convert into skyrmion in the wide Y-junction. Then, no skyrmion exists in the output branch. Thus, the logic operation "1" + "0" = "0" is thus realized. Similarly to the AND operation, the OR operation, "0" + "0" = "0", "0" + "1" = "1", "0" + "1" = "1", and "1" + "1" = "1", can also be implemented in a slightly modified geometry. For example, in the OR operation, "1" + "0" means that there is one skyrmion in input A and no skyrmion in input B. Under the driving force provided by the spin current, the skyrmion in branch A moves toward the output side. The skyrmion is first converted into a domain wall pair and the domain pair is then converted into skyrmion in the narrow Y- junction. Therefore, there is one skyrmion in the output branch finally, and the operation "1" + "0" = "1" is implemented. Soon after the numerical demonstration of the first prototype of skyrmionic logic computing device, several works on skyrmion logic emerged. For example, in 2015, Zhang et al. [281] designed the skyrmion-based NIMP, XOR, and IMP gates. In 2016, Xing et al. [282] numerically demonstrated the NAND and NOR gates by using both domain walls and skyrmions. In 2018, Luo et al. [283] numerically demonstrated logic functions including Page 43 of 80 arXiv:1906.04718 [physics.app-ph] AND, OR, NOT, NAND, NOR, XOR, and XNOR in the ferromagnetic nanotrack by virtue of various effects including SOT, SkHE, skyrmion-edge repulsions, and skyrmion-skyrmion collision. However, it should be noted that the skyrmion-based Boolean logic operations have not yet been demonstrated in an experiment. The complex structures and their co-existences of the already proposed skyrmion logic gates as well as the precise control of the dynamic behaviors of skyrmions may be difficult to be realized in nanoscale dimensions at the current stage. 4.4. Bio-Inspired Computing: Skyrmions for Neuromorphic Devices In 2017, Huang et al. proposed that the skyrmion can be used to build a skyrmion- based artificial synapse device for neuromorphic computing. They numerically demonstrated the short-term plasticity and long-term potentiation functions based on skyrmions in a nanotrack. The synaptic weight of the proposed device can be strengthened/weakened by positive/negative stimuli, mimicking the potentiation/depression process of a biological synapse. Also, the resolution of the synaptic weight can be adjusted based on the nanotrack width and the skyrmion size. The merits of using skyrmions for such artificial synaptic device include several aspects. For example, the skyrmion structure can be regarded as a nanoscale rigid object due to its quasi-particle-like nature. Based on this feature, many nanoscale skyrmions can accumulate in a given device and controlled by external stimuli, of which the properties are approximately analogous to those of a biological synapse and could be used to build nanoscale computing devices with good variability and scalability [3, 106]. Moreover, the extremely low-threshold for skyrmion excitation [1, 78, 107, 108] may promise the reduced power consumption of skyrmion synapse-based neuromorphic computing, which is highly desired for such device to be used for massive parallel computations. Later in 2017, Li et al. [145] presented a magnetic skyrmion-based artificial neuron device concept based on current-induced skyrmion motion in a nanotrack. The neuronal activity of a biologic neuron was realized with the tunable current-driven skyrmion motion dynamics of a single skyrmion. In 2018, Chen et al. [284] also proposed a magnetic skyrmion-based device to emulate the core functionality of neurons and synapses for an all-spin spiking deep neural network. The synaptic weight can be adjusted by the number of skyrmions under the read MTJs, and the resolution can be improved by having multiple branches with various conductance ranges. Recently, utilizing current-induced skyrmion generation, motion, deletion and detection together in a single device scheme, Song et al. experimentally demonstrated the basic operations of a magnetic skyrmion-based artificial synapse including potentiation and Page 44 of 80 arXiv:1906.04718 [physics.app-ph] depression, and further demonstrated artificial skyrmion synapse-based neuromorphic pattern recognition computing using simulations [144]. Besides, in 2018, Prychynenko et al. [147] proposed the skyrmion-based reservoir computing applications, in which a single skyrmion is embedded in a ferromagnetic ribbon. The reservoir computing was realized via a two-terminal device with nonlinear voltage characteristics originating from magnetoresistive effects. Most recently in 2019, Zázvorka et al. experimentally demonstrated a thermal skyrmion diffusion-based signal reshuffling device [160]. In this work, using both experiment and simulation, Zázvorka et al. uncovered the dynamics of thermally activated skyrmion diffusions using low-pinning multilayer materials, and further analysis revealed the possibility of using skyrmions for an efficient future probabilistic computing device [160]. 5. Summary and Outlook 5.1. Potential Novel Materials: Antiferromagnet, Synthetic Antiferromagnet, Ferrimagnet, Frustrated Magnet and 2D van der Waals Magnet In this section, we give an outlook on magnetic skyrmions in terms of potential novel materials and other possible topological spin textures. From the viewpoint of possible skyrmion-hosting materials, many previous works have focused on the ferromagnet, which is the most commonly known and important class of magnetic materials that has been employed in commercial products, such as modern hard disk drives. There are also two other technologically promising classes of magnetic materials, namely, the ferrimagnets and antiferromagnets, which can be used to host skyrmions. For example, several theoretical studies have demonstrated that skyrmion textures can be stabilized and manipulated in antiferromagnets [155, 268, 285-287]. For the last few years, antiferromagnetic spintronics has received much attention from the magnetism community as the antiferromagnetic materials have several intrinsic properties that may lead to a better performance of spin dynamics suitable for practical applications [288, 289]. Since 2016, several theoretical works have suggested that the magnetic skyrmions in antiferromagnets may have better mobility in compared to that in ferromagnets [155, 268, 290-293]. In ferromagnets, the magnetic skyrmion driven by an external force, such as the spin current, usually shows an undesired transverse shift, which may result in the destruction of skyrmion at sample edges. Such a phenomenon is referred to as the SkHE as mentioned in Sec. 4, which is attributed to the Magnus force induced by the topological nature of the skyrmion. In antiferromagnets, as shown in Fig. 13(a), a magnetic skyrmion can be regarded Page 45 of 80 arXiv:1906.04718 [physics.app-ph] as two coupled sublattice skyrmions with opposite topological charges, which means that the Magnus forces acted on the two sublattice skyrmions can be exactly cancelled and therefore, leading to the straight motion of the antiferromagnetic skyrmion along the direction force direction [155, 268, 290]. Compared to the current-driven skyrmions in ferromagnets, the speed of antiferromagnetic skyrmions can also be significant improved [155, 268, 290]. Similar to the skyrmions in antiferromagnets, the bilayer skyrmions in the synthetic antiferromagnets [see Fig. 13(b)] also have a zero topological charge and show no SkHE when they are driven by external driving forces [276, 277, 294]. Theoretical works also suggest that the multilayer skyrmions in a synthetic antiferromagnetic multilayer packed with even constituent ferromagnetic layers are also immune from the SkHE [30]. However, the efficient electrical manipulation (in particular -- detection) of antiferromagnetic spin textures with zero net magnetization remains as an important challenge for the practical application of antiferromagnetic materials for skyrmion-electronics. On the other hand, ferrimagnets are also composed of two sublattices which favor antiparallel alignment of spins between each other, where the magnetic moments of the two sublattices are not equal and lead to a net magnetic moment. Therefore, the ferrimagnetic skyrmions [see Fig. 13(c)] should have intermediate properties between ferromagnetic and antiferromagnetic ones. Indeed, ferrimagnetic skyrmions have non-zero SkHE, but their SkHE is much reduced in compared to the ferromagnetic skyrmions [227]. Recent reports also demonstrated that compensated ferrimagnets can have zero SkHE [269] as well as largely improved current-driven velocity suggested by the current-driven motion of domain walls in compensated ferrimagnets [241]. Due to the absence or reduction of the SkHE and their inherent fast dynamics, both antiferromagnetic and ferrimagnetic skyrmions can reach a very high speed without being destroyed at sample edges on a racetrack-like device scheme. That is to say, the antiferromagnetic skyrmion can strictly move along the nanotrack without showing a transverse motion toward the edge. For the ferrimagnetic skyrmion, although it still shows a transverse motion (i.e., a shift) toward the edge, the skyrmion-edge interaction becomes very strong only when the skyrmion velocity is extremely large. Therefore, antiferromagnets, synthetic antiferromagnets, and ferrimagnets are considered as possible skyrmion-hosting materials, where the dynamic performance of skyrmions can be remarkably improved in compared with their counterparts in conventional ferromagnetic materials. Recently, another type of magnetic materials with exotic and particular properties, i.e. the frustrated magnet, has been reported to host magnetic skyrmions. The frustration [295] in magnets can be raised from competing exchange interactions, and usually leads to complex Page 46 of 80 arXiv:1906.04718 [physics.app-ph] magnetic phase diagrams. In 2012, Okubo et al. theoretically discovered that the skyrmion lattice can be stabilized in a frustrated magnet by an order-from-disorder mechanism [89]. In 2015, Leonov and Mostovoy theoretically investigated a rich phase diagram of an anisotropic frustrated magnet and properties of frustrated skyrmions with arbitrary vorticity and helicity [296]. Then, many exotic and unique properties of frustrated skyrmions have also been studied theoretically [42, 297-300]. For example, the energy of a skyrmion with a vorticity number of Qv = +1 is identical to that of an antiskyrmion with a vorticity number of Qv = -1. Namely, the skyrmion energy in frustrated magnets can be irrespective to the skyrmion helicity [301]. Also, a frustrated skyrmion may show coupled dynamics of its helicity and center of mass, which can result in the rotational skyrmion motion [301-303]. Indeed, the coupled dynamics of a frustrated skyrmion depends on many factors. In 2017, Zhang et al. [302] theoretically reported the current-induced helicity locking-unlocking transition phenomenon of a frustrated skyrmion, which suggests the dipolar interaction may also play an important role in the dynamics of skyrmions in the frustrated magnets, especially at low temperature. Owing to the multiple degrees of freedom of skyrmions in frustrated magnets, it is anticipated that frustrated skyrmions can be used as versatile information carriers to perform information storage and computing. As shown in Fig. 13(d), in 2017, Hou et al. successfully observed various and spontaneous skyrmion bubbles in a frustrated kagome magnet Fe3Sn2 by using the LTEM at room temperature [129]. In 2019, Kurumaji et al. [304] experimentally found a Bloch-type skyrmion state in a frustrated centrosymmetric triangular-lattice magnet Gd2PdSi3, where a giant topological Hall response is observed, indicating the field-induced skyrmion phase. Very recently, it has been reported that 2D van der Waals (vdW) crystals can also have long-range intrinsic ferromagnetism in few materials with strong magnetic anisotropy [305, 306], e.g. Cr2Ge2Te3 or CrI3, whereas such magnetic order is usually strongly suppressed by thermal fluctuations [see Fig. 13(e)], as predicted by the Mermin-Wagner theorem [307]. These studies on 2D magnets opened a whole new door toward 2D heterostructure-based spintronics, as a result, there appeared significant following interests and efforts revealing the possibility of such devices by demonstrating the magnetoresistance effect in 2D magnets [308], gate-tunable room-temperature 2D magnets [309], current-driven magnetization switching [310, 311], skyrmions in 2D magnets [312], and so on. Due to the broken inversion symmetry and expected large SOC (induced by heavy atomic component) of some representative 2D magnets, e.g. Fe3GeTe2, it was suggested that 2D magnets might Page 47 of 80 arXiv:1906.04718 [physics.app-ph] permit the presence of DM interaction that could stabilize magnetic skyrmions [313, 314]. Recently, several experimental studies indeed demonstrated the stabilization of Néel-type magnetic skyrmions in a van der Waals ferromagnet such as Fe3GeTe2 [315-317] and Cr3Ge2Te6 [318], which may excite large research efforts on 2D van der Waals magnet-based skyrmions. Nevertheless, continuous studies on potential novel skyrmion-hosting materials could reveal more unexpected but remarkable properties and features of skyrmions that can be utilized to perform more efficient information processing in the future. Figure 13. Skyrmions in potential novel materials. (a) The AFM Skyrmion is composed of two topological objects with opposite topological charge; hence, the Magnus force acts in opposite directions. The strong coupling between the sublattices leads to a perfect cancellation of the two opposing forces, and so, the AFM Skyrmion has no transverse motion. Reprinted with permission from [155] and [268]. (b) Illustration of a bilayer skyrmion in an AFM-coupled nanodisk and the current-induced motion of skyrmions in the top and bottom FM layers of an AFM-coupled bilayer nanotrack. Reprinted with permission from [276]. (c) Schematic of antiferromagnetically exchange-coupled ferrimagnetic skyrmion on a magnetic track as observed in GdFeCo films and sequential STXM images showing the responses of multiple ferrimagnetic after injecting unipolar current pulses. Reprinted with permission from [227]. (d) Skyrmionic bubbles in a frustrated kagome Fe3Sn2 magnet imaged using LTEM at 300 K. Reprinted with permission from [129]. (e) Temperature- dependent magnetization of the bulk crystal Cr2Ge2Te6 measured by SQUID under fields of 0.025 T (red) and 0.3 T (blue). The inset shows the crystal structure of Cr2Ge2Te6. Bulk Cr2Ge2Te6 has a layered structure with interlayer vdW spacing of 3.4 Å. Reprinted with permission from [305]. Page 48 of 80 arXiv:1906.04718 [physics.app-ph] 5.2. Other Topological Spin Textures: Antiskyrmion, Skyrmionium, Biskyrmion, Meron, Antimeron, and Bimeron During extensive studies on magnetic skyrmions since a decade ago, a number of derivative concepts on skyrmion-like objects were created and envisioned for practical applications. As reviewed in Sec. 1, the skyrmion structure and its topological charge are determined by both in-plane and out-of-plane spin configurations. That is to say, there could be a variety of different skyrmion-like objects due to the modification of either the in-plane or out-of-plane spin configuration in principle. For example, most studies have focused on Néel-type and Bloch-type skyrmions with topological charge of Q = 1, however, several recent theoretical and experimental reports have revealed that antiskyrmions [81, 88, 205, 300, 319-322] and skyrmioniums [58, 61, 71, 72, 209, 323] have special dynamic properties and can be employed as information carriers similar to conventional skyrmions. In compared to skyrmions with Qv = +1, the antiskyrmion has different in-plane spin textures, which leads to an opposite-sign vorticity of Qv = -1 as well as largely different current-induced motion as reported in several recent studies [301-303, 314, 324]. In 2017, Nayak et al. [88] for the first time observed the antiskyrmions in tetragonal Heusler materials above room temperate, as shown in Fig. 14(a). Also, both skyrmions and antiskyrmions are stable or meta-stable solutions in frustrated magnets [301-303]. The skyrmion with Qv = +1 and antiskyrmion with Qv = -1 can be used to carry different binary information bits in a same spintronic device. Namely, the skyrmion stands for digital "1", and the antiskyrmion stands for digital "0". In such a case, it is expected that information can be computed based on the manipulation of skyrmions and antiskyrmions. Indeed, there should be a detection scheme for the system containing both skyrmions and antiskyrmion. An effective detection method should be sensitive to the variation of in-plane spin textures of skyrmions, such as the electric Hall measurement (see Sec. 3.2). Namely, the skyrmions and antiskyrmions will be differentiated by the sign of the THE since the THE is proportional to the skyrmion number. As mentioned in Sec. 1.1, the skyrmionium [24] has a topological charge of Q = 0. In 2013, the skyrmionium structure was experimentally created by a laser and observed in real space by Finazzi et al. [209]. Recently, Zhang et al. [323] also realized the real-space observation of a skyrmionium in ferromagnetic thin films coupled to a magnetic topological insulator [see Fig. 14(b)]. As the skyrmionium has a zero topological charge, it is immune from the SkHE and can be used to reliably delivery information in narrow and long nanotracks. Recently theoretical and simulation works have demonstrated that both spin Page 49 of 80 arXiv:1906.04718 [physics.app-ph] currents and spin waves are able to drive skyrmioniums into motion in narrow nanotracks [58, 71, 72, 263]. The skyrmion structure with a topological charge of Q = +2 is referred to as the biskyrmion. Theoretically, the skyrmion with Qv = +2 can be excited from a skyrmion with Qv = +1 by external stimuli, such as the STT [325]. The biskyrmion with Qv = +2 can also be formed by merging two skyrmions with Qv = +1, as theoretically demonstrated in frustrated magnets [302]. Recent years, some experiments revealed the rich static and dynamic properties of biskyrmions in different type of magnetic materials [326-330]. In 2014, Yu et al. [330] observed biskyrmions in a layered manganite by using the LTEM [see Fig. 14(c)], and realized the current-driven motion of biskyrmions. However, Loudon et al. pointed out in a recent report that the images of biskyrmions observed by LTEM can be explained as type-II magnetic bubbles viewed at an angle to their axes [331]. Note that the type-II magnetic bubble has a topological charge of zero and thus is a topologically trivial object. So, future works on biskyrmions could focus on their topology-dependent dynamics, which should be different from that of type-II bubbles. The above mentioned antiskyrmion, skyrmionium, and biskyrmion are possible solutions in easy-axis magnets. In easy-plane magnets (i.e., magnets in which magnetization favor in-plane configuration), there is also a counterpart of skyrmions, i.e. the so-called bimeron [140]. The bimeron consists of a meron and an antimeron, which forms an asymmetric spin texture carrying a topological charge of Q = 1. For this reason, the bimeron is also referred to as a meron-antimeron pair [84, 332, 333] or an asymmetric skyrmion in easy-plane magnets [334, 335]. The meron and antimeron were originated in classical field theory [336], of which the concept was then studied in different systems. For example, the bimeron was theoretically studied in quantum Hall systems in 2010 [337, 338], where a skyrmion in the bilayer system with imbalanced the electron density may be deformed into a bimeron when the parallel magnetic field penetrates between the two layers. As a counterpart of the skyrmions in easy-axis magnets (i.e., magnets where magnetization favor out-of-plane configuration), the bimeron is a localized and compact quasi-particle like spin textures in easy-plane magnets. Recently theoretical works have suggested that an isolated bimeron in either easy-plane ferromagnets or easy-plane antiferromagnets can be driven into motion by STTs [334, 335, 339]. Similar to the transformation from strip magnetic domains to magnetic skyrmions or bubbles via the "pinching mechanism" [32, 34, 116, 184, 340], the mutual conversion between a skyrmion and a bimeron is also an important issue, which is essential for future spintronic circuits Page 50 of 80 arXiv:1906.04718 [physics.app-ph] based on different topological spin textures. In 2014, Zhang et al. numerically demonstrated the possibility that a single isolated skyrmion in an easy-axis magnet can be transformed to a single isolated bimeron in an easy-plane magnet through a bridge-like narrow nanotrack junction connecting the easy-axis and easy-plane magnets [140]. In 2018, as shown in Fig. 14(d), the magnetic field-driven transformation between meron-antimeron pair lattice and skyrmion lattice was experimentally realized in a chiral magnet by Yu et al. [341, 342]. Figure 14. Other topological spin textures similar to skyrmions. (a) Under-focused LTEM images of antiskyrmions taken at fields applied along [001] of 0.29 T for antiskyrmions in Mn1.4Pt0.9Pd0.1Sn. Reprinted with permission from [88]. (b) XPEEM image of an isolated skyrmionium obtained at the Fe L3 edge of the NiFe top layer at a temperature of 44 K in zero applied magnetic field. Reprinted with permission from [323]. (c) The spin texture of biskyrmion lattice in a bilayered manganese oxide La2-2x Sr1+2x Mn2 O7 with x = 0.315. Reprinted with permission from [330]. (d) Real-space magnetization textures of a square lattice of merons and antimerons in a thin plate of the chiral-lattice magnet Co8Zn9Mn3. Reprinted with permission from [341]. 5.3. Summary In this review, we have mainly focused on key device application-relevant findings and advances that have been made in the emerging field of skyrmion-electronics since the first experimental identification of skyrmions in 2009, which might indeed enable practical applications in the future. We have reviewed the writing and deleting of skyrmions by using different methods, including magnetic field, electric field, electric current, and laser. The imaging and electrical reading of skyrmions by different state-of-art experimental techniques are also discussed. Moreover, as the processing of skyrmions is fundamental to any skyrmion-based functional device applications, relevant findings have been reviewed with a focus on the implementation of skyrmion-based racetrack memory, logic computing devices and other emerging devices. The skyrmion-electronics is a rapidly growing field, in which a Page 51 of 80 arXiv:1906.04718 [physics.app-ph] number of main topics have been expanded and many promising new topics have been introduced. Therefore, we have also introduced and put forward an overview on recent skyrmion research of different material systems as well as different skyrmion counterparts. We envision that skyrmion-electronics will continue to be an active and intriguing field of study, where more theoretical and experimental findings on the manipulation of skyrmions and their counterparts will emerge to support the development of practical applications and products for industry. Finally, we note that no review articles can cover this topic exhaustively, and therefore, here we refer to other published reviews where some of the aspects of chiral magnetic skyrmions are discussed in more detail [2, 3, 5-7, 94, 98, 100, 157, 163, 295, 343- 345]. 6. Acknowledgements X.Z. acknowledges the support by the Presidential Postdoctoral Fellowship of The Chinese University of Hong Kong, Shenzhen (CUHKSZ). Y.Z. acknowledges the support by the President's Fund of CUHKSZ, Longgang Key Laboratory of Applied Spintronics, National Natural Science Foundation of China (Grant Nos. 11974298 and 61961136006), Shenzhen Fundamental Research Fund (Grant No. JCYJ20170410171958839), and Shenzhen Peacock Group Plan (Grant No. KQTD20180413181702403). K.M.S. and T.-E.P. acknowledge the support by KIST Institutional Program (2E29410) and the support from the National Research Council of Science and Technology (NST) (Grant No. CAP-16-01-KIST) by the Korean Government (MSIP). M.E. acknowledges the support by the Grants-in-Aid for Scientific Research from JSPS KAKENHI (Grant Nos. JP18H03676, JP17K05490, and JP15H05854) and also the support by CREST, JST (Grant Nos. JPMJCR16F1 and JPMJCR1874). W.Z. acknowledges the support by the National Natural Science Foundation of China (Grant Nos. 61871008, 61571023, 61627813), and the National Key Technology Program of China (Grant No. 2017ZX01032101). G.Z. acknowledges the support by the National Natural Science Foundation of China (Grant Nos. 51771127, 51571126 and 51772004) of China, the Scientific Research Fund of Sichuan Provincial Education Department (Grant Nos. 18TD0010 and 16CZ0006). X.L. acknowledges the support by the Grants-in-Aid for Scientific Research from JSPS KAKENHI (Grant Nos. 17K19074, 26600041 and 22360122). S.W. acknowledges the support from IBM Research and management support from Guohan Hu and Daniel Worledge. Page 52 of 80 arXiv:1906.04718 [physics.app-ph] 7. References [1] A. Fert, V. Cros, and J. Sampaio, Skyrmions on the track, Nat. Nanotech. 8, 152 (2013). [2] N. Nagaosa and Y. Tokura, Topological properties and dynamics of magnetic skyrmions, Nat. Nanotech. 8, 899 (2013). [3] R. Wiesendanger, Nanoscale magnetic skyrmions in metallic films and multilayers: a new twist for spintronics, Nat. Rev. Mater. 1, 16044 (2016). [4] J. Seidel, R. K. Vasudevan, and N. Valanoor, Topological Structures in Multiferroics -- Domain Walls, Skyrmions and Vortices, Advanced Electronic Materials 2, 1500292 (2016). [5] W. Kang, Y. Huang, X. Zhang, Y. Zhou, and W. Zhao, Skyrmion-electronics: an overview and outlook, Proc. IEEE 104, 2040 (2016). [6] G. Finocchio, F. Büttner, R. Tomasello, M. Carpentieri, and M. Kläui, Magnetic skyrmions: from fundamental to applications, J. Phys. D: Appl. Phys. 49, 423001 (2016). [7] A. Fert, N. Reyren, and V. Cros, Magnetic skyrmions: advances in physics and potential applications, Nat. Rev. Mater. 2, 17031 (2017). [8] S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer, and S. Blügel, Spontaneous atomic-scale magnetic skyrmion lattice in two dimensions, Nature Physics 7, 713 (2011). [9] A. Thiaville, S. Rohart, É. Jué, V. Cros, and A. Fert, Dynamics of Dzyaloshinskii domain walls in ultrathin magnetic films, Europhys. Lett. 100, 57002 (2012). [10] F. N. Rybakov, A. B. Borisov, and A. N. Bogdanov, Three-dimensional skyrmion states in thin films of cubic helimagnets, Physical Review B 87, 094424 (2013). [11] S. Rohart and A. Thiaville, Skyrmion confinement in ultrathin film nanostructures in the presence of Dzyaloshinskii-Moriya interaction, Phys. Rev. B 88, 184422 (2013). [12] W. Wang, M. Albert, M. Beg, M.-A. Bisotti, D. Chernyshenko, D. Cortés-Ortuño, I. Hawke, and H. Fangohr, Magnon-driven domain-wall motion with the Dzyaloshinskii- Moriya Interaction, Phys. Rev. Lett. 114, 087203 (2015). [13] K. S. Ryu, L. Thomas, S. H. Yang, and S. Parkin, Chiral spin torque at magnetic domain walls, Nat. Nanotech. 8, 527 (2013). [14] S. H. Yang, K. S. Ryu, and S. Parkin, Domain-wall velocities of up to 750 m s-1 driven by exchange-coupling torque in synthetic antiferromagnets, Nat. Nanotech. 10, 221 (2015). [15] I. E. Dzyaloshinskii, Theory of Helicoidal Structures in Antiferromagnets. I. Nonmetals Sov. Phys. JETP 19, 960 (1964). [16] A. N. Bogdanov and D. A. Yablonskii, Thermodynamically stable "vortices" in magnetically ordered crystals. The mixed state of magnets, Sov. Phys. JETP 68, 101 (1989). Page 53 of 80 arXiv:1906.04718 [physics.app-ph] [17] A. N. Bogdanov, New localized solutions of the nonlinear field equations, JETP Letters 62, 247 (1995). [18] A. N. Bogdanov and U. K. Rossler, Chiral symmetry breaking in magnetic thin films and multilayers, Phys. Rev. Lett. 87, 037203 (2001). [19] A. N. Bogdanov, U. K. Rössler, M. Wolf, and K. H. Müller, Magnetic structures and reorientation transitions in noncentrosymmetric uniaxial antiferromagnets, Physical Review B 66, 214410 (2002). [20] U. K. Rossler, A. N. Bogdanov, and C. Pfleiderer, Spontaneous skyrmion ground states in magnetic metals, Nature 442, 797 (2006). [21] A. B. Butenko, A. A. Leonov, A. N. Bogdanov, and U. K. Rössler, Theory of vortex states in magnetic nanodisks with induced Dzyaloshinskii--Moriya interactions, Physical Review B 80, 134410 (2009). [22] H. Murakawa, Y. Onose, and Y. Tokura, Electric-Field Switching of a Magnetic Propagation Vector in a Helimagnet, Phys. Rev. Lett. 103, 147201 (2009). [23] A. Bogdanov and A. Hubert, Thermodynamically stable magnetic vortex states in magnetic crystals, J. Magn. Magn. Mater. 138, 255 (1994). [24] A. Bogdanov and A. Hubert, The stability of vortex-like structures in uniaxial ferromagnets, Journal of Magnetism and Magnetic Materials 195, 182 (1999). [25] A. N. Bogdanov, U. K. Rössler, and C. Pfleiderer, Modulated and localized structures in cubic helimagnets, Physica B: Condensed Matter 359-361, 1162 (2005). [26] J. Zang, M. Mostovoy, J. H. Han, and N. Nagaosa, Dynamics of Skyrmion crystals in metallic thin films, Phys Rev Lett 107, 136804 (2011). [27] E. A. Karhu, U. K. Rössler, A. N. Bogdanov, S. Kahwaji, B. J. Kirby, H. Fritzsche, M. D. Robertson, C. F. Majkrzak, and T. L. Monchesky, Chiral modulations and reorientation effects in MnSi thin films, Physical Review B 85, 094429 (2012). [28] B. Dupé, G. Bihlmayer, M. Böttcher, S. Blügel, and S. Heinze, Engineering skyrmions in transition-metal multilayers for spintronics, Nat. Commun. 7, 11779 (2016). [29] C. Moreau-Luchaire, C. Moutafis, N. Reyren, J. Sampaio, C. A. F. Vaz, N. Van Horne, K. Bouzehouane, K. Garcia, C. Deranlot, P. Warnicke, P. Wohlhüter, J. M. George, M. Weigand, J. Raabe, V. Cros, and A. Fert, Additive interfacial chiral interaction in multilayers for stabilization of small individual skyrmions at room temperature, Nat. Nanotech. 11, 444 (2016). [30] X. Zhang, M. Ezawa, and Y. Zhou, Thermally stable magnetic skyrmions in multilayer synthetic antiferromagnetic racetracks, Phys. Rev. B 94, 064406 (2016). Page 54 of 80 arXiv:1906.04718 [physics.app-ph] [31] A. Soumyanarayanan, M. Raju, A. L. Gonzalez Oyarce, A. K. C. Tan, M.-Y. Im, A. P. Petrović, P. Ho, K. H. Khoo, M. Tran, C. K. Gan, F. Ernult, and C. Panagopoulos, Tunable room-temperature magnetic skyrmions in Ir/Fe/Co/Pt multilayers, Nat. Mater. 16, 898 (2017). [32] W. Legrand, D. Maccariello, N. Reyren, K. Garcia, C. Moutafis, C. Moreau-Luchaire, S. Collin, K. Bouzehouane, V. Cros, and A. Fert, Room-Temperature Current-Induced Generation and Motion of sub-100 nm Skyrmions, Nano Lett 17, 2703 (2017). [33] S. Woo, K. M. Song, H.-S. Han, M.-S. Jung, M.-Y. Im, K.-S. Lee, K. S. Song, P. Fischer, J.-I. Hong, J. W. Choi, B.-C. Min, H. C. Koo, and J. Chang, Spin-orbit torque-driven skyrmion dynamics revealed by time-resolved X-ray microscopy, Nat. Commun. 8, 15573 (2017). [34] I. Lemesh, K. Litzius, M. Böttcher, P. Bassirian, N. Kerber, D. Heinze, J. Zázvorka, F. Büttner, L. Caretta, M. Mann, M. Weigand, S. Finizio, J. Raabe, M.-Y. Im, H. Stoll, G. Schütz, B. Dupé, M. Kläui, and G. S. D. Beach, Current-Induced Skyrmion Generation through Morphological Thermal Transitions in Chiral Ferromagnetic Heterostructures, Adv. Mater. 30, 1870372 (2018). [35] F. Büttner, I. Lemesh, M. Schneider, B. Pfau, C. M. Günther, P. Hessing, J. Geilhufe, L. Caretta, D. Engel, B. Krüger, J. Viefhaus, S. Eisebitt, and G. S. D. Beach, Field-free deterministic ultrafast creation of magnetic skyrmions by spin -- orbit torques, Nat. Nanotech. 12, 1040 (2017). [36] S. Woo, K. M. Song, X. Zhang, M. Ezawa, Y. Zhou, X. Liu, M. Weigand, S. Finizio, J. Raabe, M.-C. Park, K.-Y. Lee, J. W. Choi, B.-C. Min, H. C. Koo, and J. Chang, Deterministic creation and deletion of a single magnetic skyrmion observed by direct time- resolved X-ray microscopy, Nat. Electron. 1, 288 (2018). [37] Y. Dovzhenko, F. Casola, S. Schlotter, T. X. Zhou, F. Büttner, R. L. Walsworth, G. S. D. Beach, and A. Yacoby, Magnetostatic twists in room-temperature skyrmions explored by nitrogen-vacancy center spin texture reconstruction, Nat. Commun. 9, 2712 (2018). [38] F. Zheng, F. N. Rybakov, A. B. Borisov, D. Song, S. Wang, Z.-A. Li, H. Du, N. S. Kiselev, J. Caron, A. Kovács, M. Tian, Y. Zhang, S. Blügel, and R. E. Dunin-Borkowski, Experimental observation of chiral magnetic bobbers in B20-type FeGe, Nat. Nanotech. 13, 451 (2018). [39] J. A. Fernandez-Roldan, R. Perez del Real, C. Bran, M. Vazquez, and O. Chubykalo- Fesenko, Magnetization pinning in modulated nanowires: from topological protection to the "corkscrew" mechanism, Nanoscale 10, 5923 (2018). Page 55 of 80 arXiv:1906.04718 [physics.app-ph] [40] M. Redies, F. R. Lux, J. P. Hanke, P. M. Buhl, G. P. Müller, N. S. Kiselev, S. Blügel, and Y. Mokrousov, Distinct magnetotransport and orbital fingerprints of chiral bobbers, Physical Review B 99, 140407 (2019). [41] A. S. Ahmed, J. Rowland, B. D. Esser, S. R. Dunsiger, D. W. McComb, M. Randeria, and R. K. Kawakami, Chiral bobbers and skyrmions in epitaxial FeGe/Si(111) films, Physical Review Materials 2, 041401 (2018). [42] P. Sutcliffe, Skyrmion Knots in Frustrated Magnets, Phys. Rev. Lett. 118, 247203 (2017). [43] P. Sutcliffe, Hopfions in chiral magnets, Journal of Physics A: Mathematical and Theoretical 51, 375401 (2018). [44] X. Wang, A. Qaiumzadeh, and A. Brataas, Current-Driven Dynamics of Magnetic Hopfions, arXiv preprint arXiv:1905.09154 (2019). [45] Y. Liu, R. K. Lake, and J. Zang, Binding a hopfion in a chiral magnet nanodisk, Physical Review B 98, 174437 (2018). [46] J.-S. B. Tai and I. I. Smalyukh, Static Hopf Solitons and Knotted Emergent Fields in Solid-State Noncentrosymmetric Magnetic Nanostructures, Phys. Rev. Lett. 121, 187201 (2018). [47] J.-S. B. Tai, P. J. Ackerman, and I. I. Smalyukh, Topological transformations of Hopf solitons in chiral ferromagnets and liquid crystals, Proceedings of the National Academy of Sciences 115, 921 (2018). [48] Y. Shiomi, N. Kanazawa, K. Shibata, Y. Onose, and Y. Tokura, Topological Nernst effect in a three-dimensional skyrmion-lattice phase, Physical Review B 88, 064409 (2013). [49] F. N. Rybakov, A. B. Borisov, S. Blügel, and N. S. Kiselev, New spiral state and skyrmion lattice in 3D model of chiral magnets, New Journal of Physics 18, 045002 (2016). [50] A. O. Leonov, T. L. Monchesky, J. C. Loudon, and A. N. Bogdanov, Three-dimensional chiral skyrmions with attractive interparticle interactions, Journal of Physics: Condensed Matter 28, 35LT01 (2016). [51] R. Cheng, M. Li, A. Sapkota, A. Rai, A. Pokhrel, T. Mewes, C. Mewes, D. Xiao, M. De Graef, and V. Sokalski, Magnetic domain wall skyrmions, Physical Review B 99, 184412 (2019). [52] Y. Togawa, T. Koyama, K. Takayanagi, S. Mori, Y. Kousaka, J. Akimitsu, S. Nishihara, K. Inoue, A. S. Ovchinnikov, and J. Kishine, Chiral Magnetic Soliton Lattice on a Chiral Helimagnet, Phys. Rev. Lett. 108, 107202 (2012). Page 56 of 80 arXiv:1906.04718 [physics.app-ph] [53] T. H. R. Skyrme, A unified field theory of mesons and baryons, Nucl. Phys. 31, 556 (1962). [54] A. Bocdanov and A. Hubert, The Properties of Isolated Magnetic Vortices, Phys. Status Solidi B 186, 527 (1994). [55] H.-B. Brauna, Topological effects in nanomagnetism: from superparamagnetism to chiral quantum solitons, Adv. Phys. 61, 1 (2012). [56] Z. F. Ezawa, Quantum Hall Effects (World Scientific, Singapore, 2013). [57] A. A. Belavin and A. M. Polyakov, Metastable states of two-dimensional isotropic ferromagnets, JETP Lett. 22, 245 (1975). [58] X. Zhang, J. Xia, Y. Zhou, D. Wang, X. Liu, W. Zhao, and M. Ezawa, Control and manipulation of a magnetic skyrmionium in nanostructures, Physical Review B 94, 094420 (2016). [59] F. N. Rybakov and N. S. Kiselev, Chiral magnetic skyrmions with arbitrary topological charge, Physical Review B 99, 064437 (2019). [60] D. Foster, C. Kind, P. J. Ackerman, J.-S. B. Tai, M. R. Dennis, and I. I. Smalyukh, Two- dimensional skyrmion bags in liquid crystals and ferromagnets, Nature Physics 15, 655 (2019). [61] A. O. Leonov, U. K. Rössler, and M. Mostovoy, Target-skyrmions and skyrmion clusters in nanowires of chiral magnets, EPJ Web of Conferences 75, 05002 (2014). [62] X. Liu, Q. Zhu, S. Zhang, Q. Liu, and J. Wang, Static property and current-driven precession of 2π-vortex in nano-disk with Dzyaloshinskii-Moriya interaction, AIP Adv. 5, 087137 (2015). [63] S. Komineas and N. Papanicolaou, Skyrmion dynamics in chiral ferromagnets, Physical Review B 92, 064412 (2015). [64] S. Komineas and N. Papanicolaou, Skyrmion dynamics in chiral ferromagnets under spin-transfer torque, Physical Review B 92, 174405 (2015). [65] Y. Liu, H. Du, M. Jia, and A. Du, Switching of a target skyrmion by a spin-polarized current, Physical Review B 91, 094425 (2015). [66] M. Beg, R. Carey, W. Wang, D. Cortés-Ortuño, M. Vousden, M.-A. Bisotti, M. Albert, D. Chernyshenko, O. Hovorka, R. L. Stamps, and H. Fangohr, Ground state search, hysteretic behaviour, and reversal mechanism of skyrmionic textures in confined helimagnetic nanostructures, Sci. Rep. 5, 17137 (2015). [67] J. Mulkers, M. V. Milošević, and B. Van Waeyenberge, Cycloidal versus skyrmionic states in mesoscopic chiral magnets, Physical Review B 93, 214405 (2016). Page 57 of 80 arXiv:1906.04718 [physics.app-ph] [68] H. Fujita and M. Sato, Ultrafast generation of skyrmionic defects with vortex beams: Printing laser profiles on magnets, Physical Review B 95, 054421 (2017). [69] H. Fujita and M. Sato, Encoding orbital angular momentum of light in magnets, Physical Review B 96, 060407 (2017). [70] J. Hagemeister, A. Siemens, L. Rózsa, E. Y. Vedmedenko, and R. Wiesendanger, Controlled creation and stability of k  skyrmions on a discrete lattice, Physical Review B 97, 174436 (2018). [71] A. G. Kolesnikov, M. E. Stebliy, A. S. Samardak, and A. V. Ognev, Skyrmionium - high velocity without the skyrmion Hall effect, Scientific reports 8, 16966 (2018). [72] S. Li, J. Xia, X. Zhang, M. Ezawa, W. Kang, X. Liu, Y. Zhou, and W. Zhao, Dynamics of a magnetic skyrmionium driven by spin waves, Appl. Phys. Lett. 112, 142404 (2018). [73] M. Shen, Y. Zhang, J. Ou-Yang, X. Yang, and L. You, Motion of a skyrmionium driven by spin wave, Appl. Phys. Lett. 112, 062403 (2018). [74] J. M. D. Coey, Magnetism and Magnetic Materials (Cambridge University Press, Cambridge, 2010). [75] S. Mühlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Böni, Skyrmion lattice in a chiral magnet, Science 323, 915 (2009). [76] S. Zhang, G. van der Laan, J. Müller, L. Heinen, M. Garst, A. Bauer, H. Berger, C. Pfleiderer, and T. Hesjedal, Reciprocal space tomography of 3D skyrmion lattice order in a chiral magnet, Proceedings of the National Academy of Sciences 115, 6386 (2018). [77] Y. Fujishiro, N. Kanazawa, T. Nakajima, X. Z. Yu, K. Ohishi, Y. Kawamura, K. Kakurai, T. Arima, H. Mitamura, A. Miyake, K. Akiba, M. Tokunaga, A. Matsuo, K. Kindo, T. Koretsune, R. Arita, and Y. Tokura, Topological transitions among skyrmion- and hedgehog-lattice states in cubic chiral magnets, Nat. Commun. 10, 1059 (2019). [78] F. Jonietz, S. Mühlbauer, C. Pfleiderer, A. Neubauer, W. Münzer, A. Bauer, T. Adams, R. Georgii, P. Böni, R. A. Duine, K. Everschor, M. Garst, and A. Rosch, Spin Transfer Torques in MnSi at Ultralow Current Densities, Science 330, 1648 (2010). [79] S. W. Lovesey, Theory of neutron scattering from condensed matter (Clarendon Press, Oxford (UK), 1984). [80] I. Kezsmarki, S. Bordacs, P. Milde, E. Neuber, L. M. Eng, J. S. White, H. M. Ronnow, C. D. Dewhurst, M. Mochizuki, K. Yanai, H. Nakamura, D. Ehlers, V. Tsurkan, and A. Loidl, Neel-type skyrmion lattice with confined orientation in the polar magnetic semiconductor GaV4S8, Nat. Mater. 14, 1116 (2015). Page 58 of 80 arXiv:1906.04718 [physics.app-ph] [81] W. Koshibae and N. Nagaosa, Theory of antiskyrmions in magnets, Nat. Commun. 7, 10542 (2016). [82] I. Dzyaloshinskii, A thermodynamic theory of ``weak'' ferromagnetism of antiferromagnetics, Journal of Physics and Chemistry of Solids 4, 241 (1958). [83] T. Moriya, Anisotropic superexchange interaction and weak ferromagnetism, Physical Review 120, 91 (1960). [84] S.-Z. Lin, A. Saxena, and C. D. Batista, Skyrmion fractionalization and merons in chiral magnets with easy-plane anisotropy, Physical Review B 91, 224407 (2015). [85] K.-W. Kim, K.-W. Moon, N. Kerber, J. Nothhelfer, and K. Everschor-Sitte, Asymmetric skyrmion Hall effect in systems with a hybrid Dzyaloshinskii-Moriya interaction, Physical Review B 97, 224427 (2018). [86] J. Rowland, S. Banerjee, and M. Randeria, Skyrmions in chiral magnets with Rashba and Dresselhaus spin-orbit coupling, Physical Review B 93, 020404 (2016). [87] U. Güngördü, R. Nepal, O. A. Tretiakov, K. Belashchenko, and A. A. Kovalev, Stability of skyrmion lattices and symmetries of quasi-two-dimensional chiral magnets, Physical Review B 93, 064428 (2016). [88] A. K. Nayak, V. Kumar, T. Ma, P. Werner, E. Pippel, R. Sahoo, F. Damay, U. K. Rössler, C. Felser, and S. S. P. Parkin, Magnetic antiskyrmions above room temperature in tetragonal Heusler materials, Nature 548, 561 (2017). [89] T. Okubo, S. Chung, and H. Kawamura, Multiple-q States and the Skyrmion Lattice of the Triangular-Lattice Heisenberg Antiferromagnet under Magnetic Fields, Phys. Rev. Lett. 108, 017206 (2012). [90] Y. S. Lin, P. J. Grundy, and E. A. Giess, Bubble domains in magnetostatically coupled garnet films, Appl. Phys. Lett. 23, 485 (1973). [91] A. Malozemoff and J. Slonczewski, Magnetic Domain Walls in Bubble Materials (Academic press, New York, United States, 1979). [92] T. Garel and S. Doniach, Phase transitions with spontaneous modulation-the dipolar Ising ferromagnet, Physical Review B 26, 325 (1982). [93] T. Suzuki, A study of magnetization distribution of submicron bubbles in sputtered Ho- Co thin films, Journal of Magnetism and Magnetic Materials 31-34, 1009 (1983). [94] Y.-H. Liu and Y.-Q. Li, Dynamics of magnetic skyrmions, Chin. Phys. B 24, 017506 (2015). [95] A. Soumyanarayanan, N. Reyren, A. Fert, and C. Panagopoulos, Emergent phenomena induced by spin -- orbit coupling at surfaces and interfaces, Nature 539, 509 (2016). Page 59 of 80 arXiv:1906.04718 [physics.app-ph] [96] N. Kanazawa, S. Seki, and Y. Tokura, Noncentrosymmetric Magnets Hosting Magnetic Skyrmions, Adv. Mater. 29, 1603227 (2017). [97] F. Hellman, A. Hoffmann, Y. Tserkovnyak, G. S. D. Beach, E. E. Fullerton, C. Leighton, A. H. MacDonald, D. C. Ralph, D. A. Arena, H. A. Dürr, P. Fischer, J. Grollier, J. P. Heremans, T. Jungwirth, A. V. Kimel, B. Koopmans, I. N. Krivorotov, S. J. May, A. K. Petford-Long, J. M. Rondinelli, N. Samarth, I. K. Schuller, A. N. Slavin, M. D. Stiles, O. Tchernyshyov, A. Thiaville, and B. L. Zink, Interface-induced phenomena in magnetism, Rev. Mod. Phys. 89, 025006 (2017). [98] W. Jiang, G. Chen, K. Liu, J. Zang, S. G. E. te Velthuis, and A. Hoffmann, Skyrmions in magnetic multilayers, Phys. Rep. 704, 1 (2017). [99] B. Dieny and M. Chshiev, Perpendicular magnetic anisotropy at transition metal/oxide interfaces and applications, Rev. Mod. Phys. 89, 025008 (2017). [100] K. Everschor-Sitte, J. Masell, R. M. Reeve, and M. Kläui, Perspective: Magnetic skyrmions -- Overview of recent progress in an active research field, Journal of Applied Physics 124, 240901 (2018). [101] Y. Zhou, Magnetic skyrmions: intriguing physics and new spintronic device concepts, National Science Review 6, 210 (2018). [102] X. Z. Yu, Y. Onose, N. Kanazawa, J. H. Park, J. H. Han, Y. Matsui, N. Nagaosa, and Y. Tokura, Real-space observation of a two-dimensional skyrmion crystal, Nature 465, 901 (2010). [103] X. Z. Yu, N. Kanazawa, Y. Onose, K. Kimoto, W. Z. Zhang, S. Ishiwata, Y. Matsui, and Y. Tokura, Near room-temperature formation of a skyrmion crystal in thin-films of the helimagnet FeGe, Nat. Mater. 10, 106 (2011). [104] X. Z. Yu, N. Kanazawa, W. Z. Zhang, T. Nagai, T. Hara, K. Kimoto, Y. Matsui, Y. Onose, and Y. Tokura, Skyrmion flow near room temperature in an ultralow current density, Nat. Commun. 3, 988 (2012). [105] P. Milde, D. Köhler, J. Seidel, L. M. Eng, A. Bauer, A. Chacon, J. Kindervater, S. Mühlbauer, C. Pfleiderer, S. Buhrandt, C. Schütte, and A. Rosch, Unwinding of a Skyrmion Lattice by Magnetic Monopoles, Science 340, 1076 (2013). [106] N. Romming, C. Hanneken, M. Menzel, J. E. Bickel, B. Wolter, K. von Bergmann, A. Kubetzka, and R. Wiesendanger, Writing and deleting single magnetic skyrmions, Science 341, 636 (2013). Page 60 of 80 arXiv:1906.04718 [physics.app-ph] [107] T. Schulz, R. Ritz, A. Bauer, M. Halder, M. Wagner, C. Franz, C. Pfleiderer, K. Everschor, M. Garst, and A. Rosch, Emergent electrodynamics of skyrmions in a chiral magnet, Nat. Phys. 8, 301 (2012). [108] J. Iwasaki, M. Mochizuki, and N. Nagaosa, Universal current-velocity relation of skyrmion motion in chiral magnets, Nat. Commun. 4, 1463 (2013). [109] S. Seki, X. Z. Yu, S. Ishiwata, and Y. Tokura, Observation of skyrmions in a multiferroic material, Science 336, 198 (2012). [110] S. Seki, J. H. Kim, D. S. Inosov, R. Georgii, B. Keimer, S. Ishiwata, and Y. Tokura, Formation and rotation of skyrmion crystal in the chiral-lattice insulator Cu2OSeO3, Physical Review B 85, 220406(R) (2012). [111] T. Matsumoto, Y.-G. So, Y. Kohno, H. Sawada, Y. Ikuhara, and N. Shibata, Direct observation of Σ7 domain boundary core structure in magnetic skyrmion lattice, Science Advances 2, e1501280 (2016). [112] H. Yang, C. Wang, X. Wang, X. S. Wang, Y. Cao, and P. Yan, Twisted skyrmions at domain boundaries and the method of image skyrmions, Physical Review B 98, 014433 (2018). [113] D. McGrouther, R. J. Lamb, M. Krajnak, S. McFadzean, S. McVitie, R. L. Stamps, A. O. Leonov, A. N. Bogdanov, and Y. Togawa, Internal structure of hexagonal skyrmion lattices in cubic helimagnets, New Journal of Physics 18, 095004 (2016). [114] S. McVitie, S. Hughes, K. Fallon, S. McFadzean, D. McGrouther, M. Krajnak, W. Legrand, D. Maccariello, S. Collin, K. Garcia, N. Reyren, V. Cros, A. Fert, K. Zeissler, and C. H. Marrows, A transmission electron microscope study of Néel skyrmion magnetic textures in multilayer thin film systems with large interfacial chiral interaction, Sci. Rep. 8, 5703 (2018). [115] S. Emori, U. Bauer, S. M. Ahn, E. Martinez, and G. S. Beach, Current-driven dynamics of chiral ferromagnetic domain walls, Nat. Mater. 12, 611 (2013). [116] W. Jiang, P. Upadhyaya, W. Zhang, G. Yu, M. B. Jungfleisch, F. Y. Fradin, J. E. Pearson, Y. Tserkovnyak, K. L. Wang, O. Heinonen, S. G. E. te Velthuis, and A. Hoffmann, Blowing magnetic skyrmion bubbles, Science 349, 283 (2015). [117] A. H. Bobeck and E. Della Torre, Magnetic bubbles (North-Holland, Amsterdam, 1975). [118] A. A. Thiele, Theory of the Static Stability of Cylindrical Domains in Uniaxial Platelets, Journal of Applied Physics 41, 1139 (1970). Page 61 of 80 arXiv:1906.04718 [physics.app-ph] [119] A. Thiaville and K. Pátek, Conical bubbles, Journal of Magnetism and Magnetic Materials 124, 355 (1993). [120] O. Boulle, J. Vogel, H. Yang, S. Pizzini, D. de Souza Chaves, A. Locatelli, T. O. Menteş, A. Sala, L. D. Buda-Prejbeanu, O. Klein, M. Belmeguenai, Y. Roussigné, A. Stashkevich, S. M. Chérif, L. Aballe, M. Foerster, M. Chshiev, S. Auffret, I. M. Miron, and G. Gaudin, Room-temperature chiral magnetic skyrmions in ultrathin magnetic nanostructures, Nat. Nanotech. 11, 449 (2016). [121] S. Woo, K. Litzius, B. Kruger, M.-Y. Im, L. Caretta, K. Richter, M. Mann, A. Krone, R. M. Reeve, M. Weigand, P. Agrawal, I. Lemesh, M.-A. Mawass, P. Fischer, M. Klaui, and G. S. D. Beach, Observation of room-temperature magnetic skyrmions and their current- driven dynamics in ultrathin metallic ferromagnets, Nat. Mater. 15, 501 (2016). [122] G. Yu, P. Upadhyaya, X. Li, W. Li, S. K. Kim, Y. Fan, K. L. Wong, Y. Tserkovnyak, P. K. Amiri, and K. L. Wang, Room-temperature creation and spin -- orbit torque manipulation of skyrmions in thin films with engineered asymmetry, Nano Lett. 16, 1981 (2016). [123] S. Zhang, J. Zhang, Y. Wen, E. M. Chudnovsky, and X. Zhang, Creation of a thermally assisted skyrmion lattice in Pt/Co/Ta multilayer films, Appl. Phys. Lett. 113, 192403 (2018). [124] S. Zhang, J. Zhang, Y. Wen, E. M. Chudnovsky, and X. Zhang, Determination of chirality and density control of Néel-type skyrmions with in-plane magnetic field, Communications Physics 1, 36 (2018). [125] G. Chen, A. Mascaraque, A. T. N'Diaye, and A. K. Schmid, Room temperature skyrmion ground state stabilized through interlayer exchange coupling, Appl. Phys. Lett. 106, 242404 (2015). [126] Y. Tokunaga, X. Z. Yu, J. S. White, H. M. Rønnow, D. Morikawa, Y. Taguchi, and Y. Tokura, A new class of chiral materials hosting magnetic skyrmions beyond room temperature, Nat. Commun. 6, 7638 (2015). [127] D. A. Gilbert, B. B. Maranville, A. L. Balk, B. J. Kirby, P. Fischer, D. T. Pierce, J. Unguris, J. A. Borchers, and K. Liu, Realization of ground-state artificial skyrmion lattices at room temperature, Nat. Commun. 6, 8462 (2015). [128] L. Sun, R. X. Cao, B. F. Miao, Z. Feng, B. You, D. Wu, W. Zhang, A. Hu, and H. F. Ding, Creating an artificial two-dimensional skyrmion crystal by nanopatterning, Phys. Rev. Lett. 110, 167201 (2013). [129] Z. Hou, W. Ren, B. Ding, G. Xu, Y. Wang, B. Yang, Q. Zhang, Y. Zhang, E. Liu, F. Xu, W. Wang, G. Wu, X. Zhang, B. Shen, and Z. Zhang, Observation of Various and Page 62 of 80 arXiv:1906.04718 [physics.app-ph] Spontaneous Magnetic Skyrmionic Bubbles at Room Temperature in a Frustrated Kagome Magnet with Uniaxial Magnetic Anisotropy, Adv. Mater. 29, 1701144 (2017). [130] W. Koshibae, Y. Kaneko, J. Iwasaki, M. Kawasaki, Y. Tokura, and N. Nagaosa, Memory functions of magnetic skyrmions, Jpn. J. Appl. Phys. 54, 053001 (2015). [131] S. S. P. Parkin, M. Hayashi, and L. Thomas, Magnetic domain-wall racetrack memory, Science 320, 190 (2008). [132] J. Iwasaki, M. Mochizuki, and N. Nagaosa, Current-induced skyrmion dynamics in constricted geometries, Nat. Nanotech. 8, 742 (2013). [133] J. Sampaio, V. Cros, S. Rohart, A. Thiaville, and A. Fert, Nucleation, stability and current-induced motion of isolated magnetic skyrmions in nanostructures, Nat. Nanotech. 8, 839 (2013). [134] M. Mochizuki and S. Seki, Magnetoelectric resonances and predicted microwave diode effect of the skyrmion crystal in a multiferroic chiral-lattice magnet, Physical Review B 87, 134403 (2013). [135] Y. Okamura, F. Kagawa, M. Mochizuki, M. Kubota, S. Seki, S. Ishiwata, M. Kawasaki, Y. Onose, and Y. Tokura, Microwave magnetoelectric effect via skyrmion resonance modes in a helimagnetic multiferroic, Nat. Commun. 4, 2391 (2013). [136] G. Finocchio, M. Ricci, R. Tomasello, A. Giordano, M. Lanuzza, V. Puliafito, P. Burrascano, B. Azzerboni, and M. Carpentieri, Skyrmion based microwave detectors and harvesting, Appl. Phys. Lett. 107, 262401 (2015). [137] R. H. Liu, W. L. Lim, and S. Urazhdin, Dynamical Skyrmion State in a Spin Current Nano-Oscillator with Perpendicular Magnetic Anisotropy, Phys. Rev. Lett. 114, 137201 (2015). [138] S. F. Zhang, J. B. Wang, Q. Zheng, Q. Y. Zhu, X. Y. Liu, S. J. Chen, C. D. Jin, Q. F. Liu, C. L. Jia, and D. S. Xue, Current-induced magnetic skyrmions oscillator, New J. Phys. 17, 023061 (2015). [139] C. Navau, N. Del-Valle, and A. Sanchez, Analytical trajectories of skyrmions in confined geometries: Skyrmionic racetracks and nano-oscillators, Physical Review B 94, 184104 (2016). [140] X. Zhang, M. Ezawa, and Y. Zhou, Magnetic skyrmion logic gates: conversion, duplication and merging of skyrmions, Sci. Rep. 5, 9400 (2015). [141] X. Zhang, Y. Zhou, M. Ezawa, G. P. Zhao, and W. Zhao, Magnetic skyrmion transistor: skyrmion motion in a voltage-gated nanotrack, Sci. Rep. 5, 11369 (2015). Page 63 of 80 arXiv:1906.04718 [physics.app-ph] [142] P. Upadhyaya, G. Yu, P. K. Amiri, and K. L. Wang, Electric-field guiding of magnetic skyrmions, Phys. Rev. B 92, 134411 (2015). [143] Y. Huang, W. Kang, X. Zhang, Y. Zhou, and W. Zhao, Magnetic skyrmion-based synaptic devices, Nanotechnology 28, 08LT02 (2017). [144] K. M. Song, J.-S. Jeong, S. K. Cha, T.-E. Park, K. Kim, S. Finizio, J. Raabe, J. Chang, H. Ju, and S. Woo, Magnetic skyrmion artificial synapse for neuromorphic computing, arXiv preprint arXiv:1907.00957 (2019). [145] S. Li, W. Kang, Y. Huang, X. Zhang, Y. Zhou, and W. Zhao, Magnetic skyrmion-based artificial neuron device, Nanotechnology 28, 31LT01 (2017). [146] G. Bourianoff, D. Pinna, M. Sitte, and K. Everschor-Sitte, Potential implementation of reservoir computing models based on magnetic skyrmions, AIP Adv. 8, 055602 (2018). [147] D. Prychynenko, M. Sitte, K. Litzius, B. Krüger, G. Bourianoff, M. Kläui, J. Sinova, and K. Everschor-Sitte, Magnetic Skyrmion as a Nonlinear Resistive Element: A Potential Building Block for Reservoir Computing, Phys. Rev. Appl. 9, 014034 (2018). [148] L. Zhao, Z. Wang, X. Zhang, J. Xia, K. Wu, H.-A. Zhou, Y. Dong, G. Yu, K. L. Wang, and X. Liu, Spin-Topology Dependent Brownian Diffusion of Skyrmions, arXiv preprint arXiv:1901.08206 (2019). [149] T. Nozaki, Y. Jibiki, M. Goto, E. Tamura, T. Nozaki, H. Kubota, A. Fukushima, S. Yuasa, and Y. Suzuki, Brownian motion of skyrmion bubbles and its control by voltage applications, Appl. Phys. Lett. 114, 012402 (2019). [150] C. Reichhardt and C. J. O. Reichhardt, Thermal creep and the skyrmion Hall angle in driven skyrmion crystals, Journal of Physics: Condensed Matter 31, 07LT01 (2018). [151] R. E. Troncoso and A. S. Núñez, Thermally assisted current-driven skyrmion motion, Phys. Rev. B 89, 224403 (2014). [152] R. E. Troncoso and Á. S. Núñez, Brownian motion of massive skyrmions in magnetic thin films, Annals of Physics 351, 850 (2014). [153] D. Pinna, F. Abreu Araujo, J. V. Kim, V. Cros, D. Querlioz, P. Bessiere, J. Droulez, and J. Grollier, Skyrmion Gas Manipulation for Probabilistic Computing, Physical Review Applied 9, 064018 (2018). [154] J. Miltat, S. Rohart, and A. Thiaville, Brownian motion of magnetic domain walls and skyrmions, and their diffusion constants, Phys. Rev. B 97, 214426 (2018). [155] J. Barker and O. A. Tretiakov, Static and Dynamical Properties of Antiferromagnetic Skyrmions in the Presence of Applied Current and Temperature, Phys. Rev. Lett. 116, 147203 (2016). Page 64 of 80 arXiv:1906.04718 [physics.app-ph] [156] L. Y. Kong and J. D. Zang, Dynamics of an Insulating Skyrmion under a Temperature Gradient, Phys. Rev. Lett. 111, 067203 (2013). [157] C. Reichhardt and C. J. Olson Reichhardt, Depinning and nonequilibrium dynamic phases of particle assemblies driven over random and ordered substrates: a review, Rep. Prog. Phys. 80, 026501 (2016). [158] P. F. Bessarab, G. P. Müller, I. S. Lobanov, F. N. Rybakov, N. S. Kiselev, H. Jónsson, V. M. Uzdin, S. Blügel, L. Bergqvist, and A. Delin, Lifetime of racetrack skyrmions, Sci. Rep. 8, 3433 (2018). [159] S. Z. Lin, C. D. Batista, C. Reichhardt, and A. Saxena, ac Current Generation in Chiral Magnetic Insulators and Skyrmion Motion induced by the Spin Seebeck Effect, Phys. Rev. Lett. 112, 187203 (2014). [160] J. Zázvorka, F. Jakobs, D. Heinze, N. Keil, S. Kromin, S. Jaiswal, K. Litzius, G. Jakob, P. Virnau, D. Pinna, K. Everschor-Sitte, L. Rózsa, A. Donges, U. Nowak, and M. Kläui, Thermal skyrmion diffusion used in a reshuffler device, Nat. Nanotech. 14, 658 (2019). [161] U. K. Rössler, A. A. Leonov, and A. N. Bogdanov, Chiral Skyrmionic matter in non- centrosymmetric magnets, Journal of Physics: Conference Series 303, 012105 (2011). [162] S. X. Huang and C. L. Chien, Extended skyrmion phase in epitaxial FeGe(111) thin films, Phys. Rev. Lett. 108, 267201 (2012). [163] M. Mochizuki and S. Seki, Dynamical magnetoelectric phenomena of multiferroic skyrmions, Journal of Physics: Condensed Matter 27, 503001 (2015). [164] R. Tomasello, K. Y. Guslienko, M. Ricci, A. Giordano, J. Barker, M. Carpentieri, O. Chubykalo-Fesenko, and G. Finocchio, Origin of temperature and field dependence of magnetic skyrmion size in ultrathin nanodots, Phys. Rev. B 97, 060402 (2018). [165] X. S. Wang, H. Y. Yuan, and X. R. Wang, A theory on skyrmion size, Communications Physics 1, 31 (2018). [166] M. Jan, R. Achim, and G. Markus, Edge instabilities and skyrmion creation in magnetic layers, New J. Phys. 18, 065006 (2016). [167] M. Mochizuki, Controlled creation of nanometric skyrmions using external magnetic fields, Appl. Phys. Lett. 111, 092403 (2017). [168] D. A. Garanin, D. Capic, S. Zhang, X. Zhang, and E. M. Chudnovsky, Writing skyrmions with a magnetic dipole, Journal of Applied Physics 124, 113901 (2018). [169] S. Zhang, J. Zhang, Q. Zhang, C. Barton, V. Neu, Y. Zhao, Z. Hou, Y. Wen, C. Gong, O. Kazakova, W. Wang, Y. Peng, D. A. Garanin, E. M. Chudnovsky, and X. Zhang, Direct Page 65 of 80 arXiv:1906.04718 [physics.app-ph] writing of room temperature and zero field skyrmion lattices by a scanning local magnetic field, Appl. Phys. Lett. 112, 132405 (2018). [170] A. Wachowiak, J. Wiebe, M. Bode, O. Pietzsch, M. Morgenstern, and R. Wiesendanger, Direct Observation of Internal Spin Structure of Magnetic Vortex Cores, Science 298, 577 (2002). [171] B. Van Waeyenberge, A. Puzic, H. Stoll, K. W. Chou, T. Tyliszczak, R. Hertel, M. Fähnle, H. Brückl, K. Rott, G. Reiss, I. Neudecker, D. Weiss, C. H. Back, and G. Schütz, Magnetic vortex core reversal by excitation with short bursts of an alternating field, Nature 444, 461 (2006). [172] F. Buttner, C. Moutafis, M. Schneider, B. Kruger, C. M. Gunther, J. Geilhufe, C. v. K. Schmising, J. Mohanty, B. Pfau, S. Schaffert, A. Bisig, M. Foerster, T. Schulz, C. A. F. Vaz, J. H. Franken, H. J. M. Swagten, M. Klaui, and S. Eisebitt, Dynamics and inertia of skyrmionic spin structures, Nat. Phys. 11, 225 (2015). [173] B. Zhang, W. Wang, M. Beg, H. Fangohr, and W. Kuch, Microwave-induced dynamic switching of magnetic skyrmion cores in nanodots, Appl. Phys. Lett. 106, 102401 (2015). [174] J. Li, A. Tan, K. W. Moon, A. Doran, M. A. Marcus, A. T. Young, E. Arenholz, S. Ma, R. F. Yang, C. Hwang, and Z. Q. Qiu, Tailoring the topology of an artificial magnetic skyrmion, Nat. Commun. 5, 4704 (2014). [175] C. Heo, N. S. Kiselev, A. K. Nandy, S. Blügel, and T. Rasing, Switching of chiral magnetic skyrmions by picosecond magnetic field pulses via transient topological states, Sci. Rep. 6, 27146 (2016). [176] Y. Dai, H. Wang, T. Yang, W. Ren, and Z. Zhang, Flower-like dynamics of coupled Skyrmions with dual resonant modes by a single-frequency microwave magnetic field, Sci. Rep. 4, 6153 (2014). [177] M. Ikka, A. Takeuchi, and M. Mochizuki, Resonance modes and microwave-driven translational motion of a skyrmion crystal under an inclined magnetic field, Physical Review B 98, 184428 (2018). [178] D. C. Ralph and M. D. Stiles, Spin transfer torques, Journal of Magnetism and Magnetic Materials 320, 1190 (2008). [179] J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Spin Hall effects, Rev. Mod. Phys. 87, 1213 (2015). [180] K. Everschor, M. Garst, R. A. Duine, and A. Rosch, Current-induced rotational torques in the skyrmion lattice phase of chiral magnets, Phys. Rev. B 84, 064401 (2011). Page 66 of 80 arXiv:1906.04718 [physics.app-ph] [181] Y. Tchoe and J. H. Han, Skyrmion generation by current, Phys. Rev. B 85, 174416 (2012). [182] Y. Zhou and M. Ezawa, A reversible conversion between a skyrmion and a domain- wall pair in junction geometry, Nat. Commun. 5, 4652 (2014). [183] O. Heinonen, W. Jiang, H. Somaily, S. G. E. te Velthuis, and A. Hoffmann, Generation of magnetic skyrmion bubbles by inhomogeneous spin Hall currents, Phys. Rev. B 93, 094407 (2016). [184] S.-Z. Lin, Edge instability in a chiral stripe domain under an electric current and skyrmion generation, Phys. Rev. B 94, 020402 (2016). [185] Y. Liu, H. Yan, M. Jia, H. Du, and A. Du, Topological analysis of spin-torque driven magnetic skyrmion formation, Appl. Phys. Lett. 109, 102402 (2016). [186] H. Y. Yuan and X. R. Wang, Skyrmion Creation and Manipulation by Nano-Second Current Pulses, Sci. Rep. 6, 22638 (2016). [187] G. Yin, Y. Li, L. Kong, R. K. Lake, C. L. Chien, and J. Zang, Topological charge analysis of ultrafast single skyrmion creation, Physical Review B 93, 174403 (2016). [188] A. Hrabec, J. Sampaio, M. Belmeguenai, I. Gross, R. Weil, S. M. Chérif, A. Stashkevich, V. Jacques, A. Thiaville, and S. Rohart, Current-induced skyrmion generation and dynamics in symmetric bilayers, Nat. Commun. 8, 15765 (2017). [189] S. Finizio, K. Zeissler, S. Wintz, S. Mayr, T. Wessels, A. J. Huxtable, G. Burnell, C. H. Marrows, and J. Raabe, Deterministic field-free skyrmion nucleation at a nano-engineered injector device, arXiv preprint arXiv:1902.10435 (2019). [190] A. De Lucia, K. Litzius, B. Krüger, O. A. Tretiakov, and M. Kläui, Multiscale simulations of topological transformations in magnetic-skyrmion spin structures, Physical Review B 96, 020405 (2017). [191] S. Seki, S. Ishiwata, and Y. Tokura, Magnetoelectric nature of skyrmions in a chiral magnetic insulator Cu2OSeO3, Physical Review B 86, 060403 (2012). [192] P. Huang, M. Cantoni, A. Kruchkov, J. Rajeswari, A. Magrez, F. Carbone, and H. M. Rønnow, In Situ Electric Field Skyrmion Creation in Magnetoelectric Cu2OSeO3, Nano. Lett. 18, 5167 (2018). [193] M. Mochizuki and Y. Watanabe, Writing a skyrmion on multiferroic materials, Appl. Phys. Lett. 107, 082409 (2015). [194] Y. Okamura, F. Kagawa, S. Seki, and Y. Tokura, Transition to and from the skyrmion lattice phase by electric fields in a magnetoelectric compound, Nat. Commun. 7, 12669 (2016). Page 67 of 80 arXiv:1906.04718 [physics.app-ph] [195] P.-J. Hsu, A. Kubetzka, A. Finco, N. Romming, K. von Bergmann, and R. Wiesendanger, Electric-field-driven switching of individual magnetic skyrmions, Nat. Nanotech. 12, 123 (2017). [196] M. Schott, A. Bernand-Mantel, L. Ranno, S. Pizzini, J. Vogel, H. Béa, C. Baraduc, S. Auffret, G. Gaudin, and D. Givord, The skyrmion switch: turning magnetic skyrmion bubbles on and off with an electric field, Nano. Lett. 17, 3006 (2017). [197] T. Srivastava, M. Schott, R. Juge, V. Křižáková, M. Belmeguenai, Y. Roussigné, A. Bernand-Mantel, L. Ranno, S. Pizzini, S.-M. Chérif, A. Stashkevich, S. Auffret, O. Boulle, G. Gaudin, M. Chshiev, C. Baraduc, and H. Béa, Large-Voltage Tuning of Dzyaloshinskii -- Moriya Interactions: A Route toward Dynamic Control of Skyrmion Chirality, Nano. Lett. 18, 4871 (2018). [198] C. Ma, X. Zhang, J. Xia, M. Ezawa, W. Jiang, T. Ono, S. N. Piramanayagam, A. Morisako, Y. Zhou, and X. Liu, Electric Field-Induced Creation and Directional Motion of Domain Walls and Skyrmion Bubbles, Nano. Lett. 19, 353 (2019). [199] D. Bhattacharya, M. M. Al-Rashid, and J. Atulasimha, Voltage controlled core reversal of fixed magnetic skyrmions without a magnetic field, Sci. Rep. 6, 31272 (2016). [200] E. Beaurepaire, J. C. Merle, A. Daunois, and J. Y. Bigot, Ultrafast Spin Dynamics in Ferromagnetic Nickel, Phys. Rev. Lett. 76, 4250 (1996). [201] C. D. Stanciu, F. Hansteen, A. V. Kimel, A. Kirilyuk, A. Tsukamoto, A. Itoh, and T. Rasing, All-Optical Magnetic Recording with Circularly Polarized Light, Phys. Rev. Lett. 99, 047601 (2007). [202] I. Radu, K. Vahaplar, C. Stamm, T. Kachel, N. Pontius, H. A. Dürr, T. A. Ostler, J. Barker, R. F. L. Evans, R. W. Chantrell, A. Tsukamoto, A. Itoh, A. Kirilyuk, T. Rasing, and A. V. Kimel, Transient ferromagnetic-like state mediating ultrafast reversal of antiferromagnetically coupled spins, Nature 472, 205 (2011). [203] C. H. Lambert, S. Mangin, B. S. D. C. S. Varaprasad, Y. K. Takahashi, M. Hehn, M. Cinchetti, G. Malinowski, K. Hono, Y. Fainman, M. Aeschlimann, and E. E. Fullerton, All- optical control of ferromagnetic thin films and nanostructures, Science 345, 1337 (2014). [204] A. Stupakiewicz, K. Szerenos, D. Afanasiev, A. Kirilyuk, and A. V. Kimel, Ultrafast nonthermal photo-magnetic recording in a transparent medium, Nature 542, 71 (2017). [205] W. Koshibae and N. Nagaosa, Creation of skyrmions and antiskyrmions by local heating, Nat. Commun. 5, 5148 (2014). [206] S.-G. Je, P. Vallobra, T. Srivastava, J.-C. Rojas-Sánchez, T. H. Pham, M. Hehn, G. Malinowski, C. Baraduc, S. Auffret, G. Gaudin, S. Mangin, H. Béa, and O. Boulle, Creation Page 68 of 80 arXiv:1906.04718 [physics.app-ph] of Magnetic Skyrmion Bubble Lattices by Ultrafast Laser in Ultrathin Films, Nano. Lett. 18, 7362 (2018). [207] T. Ogasawara, N. Iwata, Y. Murakami, H. Okamoto, and Y. Tokura, Submicron-scale spatial feature of ultrafast photoinduced magnetization reversal in TbFeCo thin film, Appl. Phys. Lett. 94, 162507 (2009). [208] M. Ezawa, Giant Skyrmions Stabilized by Dipole-Dipole Interactions in Thin Ferromagnetic Films, Phys. Rev. Lett. 105, 197202 (2010). [209] M. Finazzi, M. Savoini, A. R. Khorsand, A. Tsukamoto, A. Itoh, L. Duò, A. Kirilyuk, T. Rasing, and M. Ezawa, Laser-induced magnetic nanostructures with tunable topological Properties, Phys. Rev. Lett. 110, 177205 (2013). [210] W. Yang, H. Yang, Y. Cao, and P. Yan, Photonic orbital angular momentum transfer and magnetic skyrmion rotation, Optics Express 26, 8778 (2018). [211] G. Berruto, I. Madan, Y. Murooka, G. M. Vanacore, E. Pomarico, J. Rajeswari, R. Lamb, P. Huang, A. J. Kruchkov, Y. Togawa, T. LaGrange, D. McGrouther, H. M. Rønnow, and F. Carbone, Laser-Induced Skyrmion Writing and Erasing in an Ultrafast Cryo-Lorentz Transmission Electron Microscope, Phys. Rev. Lett. 120, 117201 (2018). [212] N. Ogawa, S. Seki, and Y. Tokura, Ultrafast optical excitation of magnetic skyrmions, Sci. Rep. 5, 9552 (2015). [213] E. F. Kneller and R. Hawig, The exchange-spring magnet: a new material principle for permanent magnets, IEEE Transactions on Magnetics 27, 3588 (1991). [214] A. A. Fraerman, O. L. Ermolaeva, E. V. Skorohodov, N. S. Gusev, V. L. Mironov, S. N. Vdovichev, and E. S. Demidov, Skyrmion states in multilayer exchange coupled ferromagnetic nanostructures with distinct anisotropy directions, Journal of Magnetism and Magnetic Materials 393, 452 (2015). [215] R. Streubel, P. Fischer, M. Kopte, O. G. Schmidt, and D. Makarov, Magnetization dynamics of imprinted non-collinear spin textures, Appl. Phys. Lett. 107, 112406 (2015). [216] N. Del-Valle, S. Agramunt-Puig, A. Sanchez, and C. Navau, Imprinting skyrmions in thin films by ferromagnetic and superconducting templates, Appl. Phys. Lett. 107, 133103 (2015). [217] L. Sun, H. Z. Wu, B. F. Miao, D. Wu, and H. F. Ding, Tuning the stability and the skyrmion Hall effect in magnetic skyrmions by adjusting their exchange strengths with magnetic disks, Journal of Magnetism and Magnetic Materials 455, 39 (2018). Page 69 of 80 arXiv:1906.04718 [physics.app-ph] [218] W. Jiang, X. Zhang, G. Yu, W. Zhang, X. Wang, M. Benjamin Jungfleisch, J. E. Pearson, X. Cheng, O. Heinonen, K. L. Wang, Y. Zhou, A. Hoffmann, and S. G. E. te Velthuis, Direct observation of the skyrmion Hall effect, Nat. Phys. 13, 162 (2017). [219] K. Litzius, I. Lemesh, B. Kruger, P. Bassirian, L. Caretta, K. Richter, F. Buttner, K. Sato, O. A. Tretiakov, J. Forster, R. M. Reeve, M. Weigand, I. Bykova, H. Stoll, G. Schutz, G. S. D. Beach, and M. Klaui, Skyrmion Hall effect revealed by direct time-resolved X-ray microscopy, Nat. Phys. 13, 170 (2017). [220] X. Zhao, C. Jin, C. Wang, H. Du, J. Zang, M. Tian, R. Che, and Y. Zhang, Direct imaging of magnetic field-driven transitions of skyrmion cluster states in FeGe nanodisks, Proceedings of the National Academy of Sciences 113, 4918 (2016). [221] H. Du, R. Che, L. Kong, X. Zhao, C. Jin, C. Wang, J. Yang, W. Ning, R. Li, C. Jin, X. Chen, J. Zang, Y. Zhang, and M. Tian, Edge-mediated skyrmion chain and its collective dynamics in a confined geometry, Nat. Commun. 6, 8504 (2015). [222] C. Wang, H. Du, X. Zhao, C. Jin, M. Tian, Y. Zhang, and R. Che, Enhanced Stability of the Magnetic Skyrmion Lattice Phase under a Tilted Magnetic Field in a Two-Dimensional Chiral Magnet, Nano. Lett. 17, 2921 (2017). [223] X. Z. Yu, J. P. DeGrave, Y. Hara, T. Hara, S. Jin, and Y. Tokura, Observation of the Magnetic Skyrmion Lattice in a MnSi Nanowire by Lorentz TEM, Nano Lett. 13, 3755 (2013). [224] K. Shibata, X. Z. Yu, T. Hara, D. Morikawa, N. Kanazawa, K. Kimoto, S. Ishiwata, Y. Matsui, and Y. Tokura, Towards control of the size and helicity of skyrmions in helimagnetic alloys by spin -- orbit coupling, Nat. Nanotech. 8, 723 (2013). [225] J. C. T. Lee, J. J. Chess, S. A. Montoya, X. Shi, N. Tamura, S. K. Mishra, P. Fischer, B. J. McMorran, S. K. Sinha, E. E. Fullerton, S. D. Kevan, and S. Roy, Synthesizing skyrmion bound pairs in Fe-Gd thin films, Appl. Phys. Lett. 109, 022402 (2016). [226] N. Romming, A. Kubetzka, C. Hanneken, K. von Bergmann, and R. Wiesendanger, Field-Dependent Size and Shape of Single Magnetic Skyrmions, Phys. Rev. Lett. 114, 177203 (2015). [227] S. Woo, K. M. Song, X. Zhang, Y. Zhou, M. Ezawa, X. Liu, S. Finizio, J. Raabe, N. J. Lee, S.-I. Kim, S.-Y. Park, Y. Kim, J.-Y. Kim, D. Lee, O. Lee, J. W. Choi, B.-C. Min, H. C. Koo, and J. Chang, Current-driven dynamics and inhibition of the skyrmion Hall effect of ferrimagnetic skyrmions in GdFeCo films, Nat. Commun. 9, 959 (2018). Page 70 of 80 arXiv:1906.04718 [physics.app-ph] [228] R. Tolley, S. A. Montoya, and E. E. Fullerton, Room-temperature observation and current control of skyrmions in Pt/Co/Os/Pt thin films, Physical Review Materials 2, 044404 (2018). [229] C.-M. Jin and H.-F. Du, Real-space observation of individual skyrmions in helimagnetic nanostripes, Chinese Physics B 24, 128501 (2015). [230] A. O. Leonov, T. L. Monchesky, N. Romming, A. Kubetzka, A. N. Bogdanov, and R. Wiesendanger, The properties of isolated chiral skyrmions in thin magnetic films, New Journal of Physics 18, 065003 (2016). [231] T. Nakajima and T.-h. Arima, Observation of Magnetic Skyrmions by Neutron Scattering, J. Phys. Soc. Jpn. 88, 081006 (2019). [232] H. Luo, W. Xia, H. Du, and J. P. Liu, in Skyrmions (CRC Press, Boca Raton, 2016), p. 53. [233] H. Hopster and H. P. Oepen, Magnetic microscopy of nanostructures (Springer, Berlin, 2005). [234] J. Müller, J. Rajeswari, P. Huang, Y. Murooka, H. M. Rønnow, F. Carbone, and A. Rosch, Magnetic Skyrmions and Skyrmion Clusters in the Helical Phase of Cu2OSeO3, Phys. Rev. Lett. 119, 137201 (2017). [235] A. Tonomura, X. Yu, K. Yanagisawa, T. Matsuda, Y. Onose, N. Kanazawa, H. S. Park, and Y. Tokura, Real-Space Observation of Skyrmion Lattice in Helimagnet MnSi Thin Samples, Nano Lett. 12, 1673 (2012). [236] S. McVitie, D. McGrouther, S. McFadzean, D. A. MacLaren, K. J. O'Shea, and M. J. Benitez, Aberration corrected Lorentz scanning transmission electron microscopy, Ultramicroscopy 152, 57 (2015). [237] M. J. Benitez, A. Hrabec, A. P. Mihai, T. A. Moore, G. Burnell, D. McGrouther, C. H. Marrows, and S. McVitie, Magnetic microscopy and topological stability of homochiral Néel domain walls in a Pt/Co/AlOx trilayer, Nat. Commun. 6, 8957 (2015). [238] S. D. Pollard, J. A. Garlow, J. Yu, Z. Wang, Y. Zhu, and H. Yang, Observation of stable Néel skyrmions in cobalt/palladium multilayers with Lorentz transmission electron microscopy, Nat. Commun. 8, 14761 (2017). [239] N. Dekkers and H. De Lang, Differential phase contrast in a STEM, Optik 41, 452 (1974). [240] J. N. Chapman, P. E. Batson, E. M. Waddell, and R. P. Ferrier, The direct determination of magnetic domain wall profiles by differential phase contrast electron microscopy, Ultramicroscopy 3, 203 (1978). Page 71 of 80 arXiv:1906.04718 [physics.app-ph] [241] L. Caretta, M. Mann, F. Büttner, K. Ueda, B. Pfau, C. M. Günther, P. Hessing, A. Churikova, C. Klose, M. Schneider, D. Engel, C. Marcus, D. Bono, K. Bagschik, S. Eisebitt, and G. S. D. Beach, Fast current-driven domain walls and small skyrmions in a compensated ferrimagnet, Nat. Nanotech. 13, 1154 (2018). [242] A. Neubauer, C. Pfleiderer, B. Binz, A. Rosch, R. Ritz, P. G. Niklowitz, and P. Böni, Topological Hall Effect in the $A$ Phase of MnSi, Phys. Rev. Lett. 102, 186602 (2009). [243] M. Lee, W. Kang, Y. Onose, Y. Tokura, and N. P. Ong, Unusual Hall Effect Anomaly in MnSi under Pressure, Phys. Rev. Lett. 102, 186601 (2009). [244] N. Kanazawa, Y. Onose, T. Arima, D. Okuyama, K. Ohoyama, S. Wakimoto, K. Kakurai, S. Ishiwata, and Y. Tokura, Large Topological Hall Effect in a Short-Period Helimagnet MnGe, Phys. Rev. Lett. 106, 156603 (2011). [245] P. Bruno, V. K. Dugaev, and M. Taillefumier, Topological Hall Effect and Berry Phase in Magnetic Nanostructures, Phys. Rev. Lett. 93, 096806 (2004). [246] B. Binz, A. Vishwanath, and V. Aji, Theory of the Helical Spin Crystal: A Candidate for the Partially Ordered State of MnSi, Phys. Rev. Lett. 96, 207202 (2006). [247] B. Binz and A. Vishwanath, Chirality induced anomalous-Hall effect in helical spin crystals, Physica B: Condensed Matter 403, 1336 (2008). [248] K. Hamamoto, M. Ezawa, and N. Nagaosa, Quantized topological Hall effect in skyrmion crystal, Phys. Rev. B 92, 115417 (2015). [249] T. Yokouchi, N. Kanazawa, A. Tsukazaki, Y. Kozuka, M. Kawasaki, M. Ichikawa, F. Kagawa, and Y. Tokura, Stability of two-dimensional skyrmions in thin films of Mn1-xFexSi investigated by the topological Hall effect, Physical Review B 89, 064416 (2014). [250] H. Oike, A. Kikkawa, N. Kanazawa, Y. Taguchi, M. Kawasaki, Y. Tokura, and F. Kagawa, Interplay between topological and thermodynamic stability in a metastable magnetic skyrmion lattice, Nature Physics 12, 62 (2015). [251] J. Matsuno, N. Ogawa, K. Yasuda, F. Kagawa, W. Koshibae, N. Nagaosa, Y. Tokura, and M. Kawasaki, Interface-driven topological Hall effect in SrRuO3-SrIrO3 bilayer, Science Advances 2, e1600304 (2016). [252] C. Liu, Y. Zang, W. Ruan, Y. Gong, K. He, X. Ma, Q.-K. Xue, and Y. Wang, Dimensional Crossover-Induced Topological Hall Effect in a Magnetic Topological Insulator, Phys. Rev. Lett. 119, 176809 (2017). [253] D. Maccariello, W. Legrand, N. Reyren, K. Garcia, K. Bouzehouane, S. Collin, V. Cros, and A. Fert, Electrical detection of single magnetic skyrmions in metallic multilayers at room temperature, Nat. Nanotech. 13, 233 (2018). Page 72 of 80 arXiv:1906.04718 [physics.app-ph] [254] K. Zeissler, S. Finizio, K. Shahbazi, J. Massey, F. A. Ma'Mari, D. M. Bracher, A. Kleibert, M. C. Rosamond, E. H. Linfield, T. A. Moore, J. Raabe, G. Burnell, and C. H. Marrows, Discrete Hall resistivity contribution from Néel skyrmions in multilayer nanodiscs, Nat. Nanotech. 13, 1161 (2018). [255] K. Hamamoto, M. Ezawa, and N. Nagaosa, Purely electrical detection of a skyrmion in constricted geometry, Appl. Phys. Lett. 108, 112401 (2016). [256] C. Hanneken, F. Otte, A. Kubetzka, B. DupÃ©, N. Romming, K. von Bergmann, R. Wiesendanger, and S. Heinze, Electrical detection of magnetic skyrmions by tunnelling non- collinear magnetoresistance, Nat. Nanotech. 10, 1039 (2015). [257] H. Du, D. Liang, C. Jin, L. Kong, M. J. Stolt, W. Ning, J. Yang, Y. Xing, J. Wang, R. Che, J. Zang, S. Jin, Y. Zhang, and M. Tian, Electrical probing of field-driven cascading quantized transitions of skyrmion cluster states in MnSi nanowires, Nat. Commun. 6, 7637 (2015). [258] R. Tomasello, M. Ricci, P. Burrascano, V. Puliafito, M. Carpentieri, and G. Finocchio, Electrical detection of single magnetic skyrmion at room temperature, AIP Adv. 7, 056022 (2017). [259] A. Kubetzka, C. Hanneken, R. Wiesendanger, and K. von Bergmann, Impact of the skyrmion spin texture on magnetoresistance, Physical Review B 95, 104433 (2017). [260] M. N. Baibich, J. M. Broto, A. Fert, F. N. Van Dau, F. Petroff, P. Etienne, G. Creuzet, A. Friederich, and J. Chazelas, Giant Magnetoresistance of (001)Fe/(001)Cr Magnetic Superlattices, Phys. Rev. Lett. 61, 2472 (1988). [261] S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma, H. D. Gan, M. Endo, S. Kanai, J. Hayakawa, F. Matsukura, and H. Ohno, A perpendicular-anisotropy CoFeB -- MgO magnetic tunnel junction, Nat. Mater. 9, 721 (2010). [262] D. M. Crum, M. Bouhassoune, J. Bouaziz, B. Schweflinghaus, S. Blugel, and S. Lounis, Perpendicular reading of single confined magnetic skyrmions, Nat. Commun. 6, 8541 (2015). [263] R. P. Loreto, W. A. Moura-Melo, A. R. Pereira, X. Zhang, Y. Zhou, M. Ezawa, and C. I. L. de Araujo, Creation, transport and detection of imprinted magnetic solitons stabilized by spin-polarized current, Journal of Magnetism and Magnetic Materials 455, 25 (2018). [264] X. Zhang, W. Cai, X. Zhang, Z. Wang, Z. Li, Y. Zhang, K. Cao, N. Lei, W. Kang, Y. Zhang, H. Yu, Y. Zhou, and W. Zhao, Skyrmions in Magnetic Tunnel Junctions, ACS Applied Materials & Interfaces 10, 16887 (2018). Page 73 of 80 arXiv:1906.04718 [physics.app-ph] [265] N. E. Penthorn, X. Hao, Z. Wang, Y. Huai, and H. W. Jiang, Experimental Observation of Single Skyrmion Signatures in a Magnetic Tunnel Junction, Phys. Rev. Lett. 122, 257201 (2019). [266] S. Kasai, S. Sugimoto, Y. Nakatani, R. Ishikawa, and Y. K. Takahashi, Voltage- controlled magnetic skyrmions in magnetic tunnel junctions, Appl. Phys. Express. 12, 083001 (2019). [267] A. A. Thiele, Steady-state motion of magnetic domains, Phys. Rev. Lett. 30, 230 (1973). [268] X. Zhang, Y. Zhou, and M. Ezawa, Antiferromagnetic Skyrmion: Stability, Creation and Manipulation, Sci. Rep. 6, 24795 (2016). [269] Y. Hirata, D.-H. Kim, S. K. Kim, D.-K. Lee, S.-H. Oh, D.-Y. Kim, T. Nishimura, T. Okuno, Y. Futakawa, H. Yoshikawa, A. Tsukamoto, Y. Tserkovnyak, Y. Shiota, T. Moriyama, S.-B. Choe, K.-J. Lee, and T. Ono, Vanishing skyrmion Hall effect at the angular momentum compensation temperature of a ferrimagnet, Nat. Nanotech. 14, 232 (2019). [270] G. Chen, Skyrmion Hall effect, Nature Physics 13, 112 (2017). [271] S. Parkin and S. H. Yang, Memory on the racetrack, Nat. Nanotech. 10, 195 (2015). [272] R. Tomasello, E. Martinez, R. Zivieri, L. Torres, M. Carpentieri, and G. Finocchio, A strategy for the design of skyrmion racetrack memories, Sci. Rep. 4, 6784 (2014). [273] G. Yu, P. Upadhyaya, Q. Shao, H. Wu, G. Yin, X. Li, C. He, W. Jiang, X. Han, P. K. Amiri, and K. L. Wang, Room-temperature skyrmion shift device for memory application, Nano Lett. 17, 261 (2017). [274] X. Zhang, G. P. Zhao, H. Fangohr, J. P. Liu, W. X. Xia, J. Xia, and F. J. Morvan, Skyrmion-skyrmion and skyrmion-edge repulsions on the skyrmion-based racetrack memory, Sci. Rep. 5, 7643 (2015). [275] S.-Z. Lin, C. Reichhardt, C. D. Batista, and A. Saxena, Particle model for skyrmions in metallic chiral magnets: Dynamics, pinning, and creep, Phys. Rev. B 87, 214419 (2013). [276] X. Zhang, Y. Zhou, and M. Ezawa, Magnetic bilayer-skyrmions without skyrmion Hall effect, Nat. Commun. 7, 10293 (2016). [277] R. Tomasello, V. Puliafito, E. Martinez, A. Manchon, M. Ricci, M. Carpentieri, and G. Finocchio, Performance of synthetic antiferromagnetic racetrack memory: domain wall versus skyrmion, J. Phys. D: Appl. Phys. 50, 325302 (2017). [278] W. Kang, C. Zheng, Y. Huang, X. Zhang, Y. Zhou, W. Lv, and W. Zhao, Complementary Skyrmion Racetrack Memory With Voltage Manipulation, IEEE Electron Device Letters 37, 924 (2016). Page 74 of 80 arXiv:1906.04718 [physics.app-ph] [279] J. Xia, Y. Huang, X. Zhang, W. Kang, C. Zheng, X. Liu, W. Zhao, and Y. Zhou, A microwave field-driven transistor-like skyrmionic device with the microwave current-assisted skyrmion creation, Journal of Applied Physics 122, 153901 (2017). [280] X. Zhao, R. Ren, G. Xie, and Y. Liu, Single antiferromagnetic skyrmion transistor based on strain manipulation, Appl. Phys. Lett. 112, 252402 (2018). [281] S. Zhang, A. A. Baker, S. Komineas, and T. Hesjedal, Topological computation based on direct magnetic logic communication, Scientific reports 5, 15773 (2015). [282] X. Xing, P. W. T. Pong, and Y. Zhou, Skyrmion domain wall collision and domain wall-gated skyrmion logic, Physical Review B 94, 054408 (2016). [283] S. Luo, M. Song, X. Li, Y. Zhang, J. Hong, X. Yang, X. Zou, N. Xu, and L. You, Reconfigurable Skyrmion Logic Gates, Nano Lett 18, 1180 (2018). [284] M. Chen, A. Sengupta, and K. Roy, Magnetic Skyrmion as a Spintronic Deep Learning Spiking Neuron Processor, IEEE Transactions on Magnetics 54, 1 (2018). [285] B. Göbel, A. Mook, J. Henk, and I. Mertig, Antiferromagnetic skyrmion crystals: Generation, topological Hall, and topological spin Hall effect, Physical Review B 96, 060406 (2017). [286] P. M. Buhl, F. Freimuth, S. Blügel, and Y. Mokrousov, Topological spin Hall effect in antiferromagnetic skyrmions, Physica Status Solidi (RRL) -- Rapid Research Letters 11, 1700007 (2017). [287] Q. L. He, G. Yin, A. J. Grutter, L. Pan, X. Che, G. Yu, D. A. Gilbert, S. M. Disseler, Y. Liu, P. Shafer, B. Zhang, Y. Wu, B. J. Kirby, E. Arenholz, R. K. Lake, X. Han, and K. L. Wang, Exchange-biasing topological charges by antiferromagnetism, Nat. Commun. 9, 2767 (2018). [288] T. Jungwirth, X. Marti, P. Wadley, and J. Wunderlich, Antiferromagnetic spintronics, Nat. Nanotech. 11, 231 (2016). [289] V. Baltz, A. Manchon, M. Tsoi, T. Moriyama, T. Ono, and Y. Tserkovnyak, Antiferromagnetic spintronics, Rev. Mod. Phys. 90, 015005 (2018). [290] L. Shen, J. Xia, G. Zhao, X. Zhang, M. Ezawa, O. A. Tretiakov, X. Liu, and Y. Zhou, Dynamics of the antiferromagnetic skyrmion induced by a magnetic anisotropy gradient, Phys. Rev. B 98, 134448 (2018). [291] H. Velkov, O. Gomonay, M. Beens, G. Schwiete, A. Brataas, J. Sinova, and R. A. Duine, Phenomenology of current-induced skyrmion motion in antiferromagnets, New Journal of Physics 18, 075016 (2016). Page 75 of 80 arXiv:1906.04718 [physics.app-ph] [292] C. Jin, C. Song, J. Wang, and Q. Liu, Dynamics of antiferromagnetic skyrmion driven by the spin Hall effect, Appl. Phys. Lett. 109, 182404 (2016). [293] R. L. Silva, R. C. Silva, A. R. Pereira, and W. A. Moura-Melo, Antiferromagnetic skyrmions overcoming obstacles in a racetrack, Journal of Physics: Condensed Matter 31, 225802 (2019). [294] Q. Sheng, X. L. Liu, W. J. Chen, M. Y. Li, L. J. Liu, and Y. Zheng, Realization of skyrmion subtracter and diverter in a voltage-gated synthetic antiferromagnetic racetrack, Journal of Applied Physics 125, 064502 (2019). [295] T. H. Diep, Phase Transition in Frustrated Magnetic Thin Film -- Physics at Phase Boundaries, Entropy 21, 175 (2019). [296] A. O. Leonov and M. Mostovoy, Multiply periodic states and isolated skyrmions in an anisotropic frustrated magnet, Nat. Commun. 6, 8275 (2015). [297] A. O. Leonov and M. Mostovoy, Edge states and skyrmion dynamics in nanostripes of frustrated magnets, Nat. Commun. 8, 14394 (2017). [298] H. Y. Yuan, O. Gomonay, and M. Kläui, Skyrmions and multisublattice helical states in a frustrated chiral magnet, Physical Review B 96, 134415 (2017). [299] S. von Malottki, B. Dupe, P. F. Bessarab, A. Delin, and S. Heinze, Enhanced skyrmion stability due to exchange frustration, Scientific reports 7, 12299 (2017). [300] J. J. Liang, J. H. Yu, J. Chen, M. H. Qin, M. Zeng, X. B. Lu, X. S. Gao, and J. M. Liu, Magnetic field gradient driven dynamics of isolated skyrmions and antiskyrmions in frustrated magnets, New Journal of Physics 20, 053037 (2018). [301] S.-Z. Lin and S. Hayami, Ginzburg-Landau theory for skyrmions in inversion- symmetric magnets with competing interactions, Phys. Rev. B 93, 064430 (2016). [302] X. Zhang, J. Xia, Y. Zhou, X. Liu, H. Zhang, and M. Ezawa, Skyrmion dynamics in a frustrated ferromagnetic film and current-induced helicity locking-unlocking transition, Nat. Commun. 8, 1717 (2017). [303] J. Xia, X. Zhang, M. Ezawa, Z. Hou, W. Wang, X. Liu, and Y. Zhou, Current-Driven Dynamics of Frustrated Skyrmions in a Synthetic Antiferromagnetic Bilayer, Physical Review Applied 11, 044046 (2019). [304] T. Kurumaji, T. Nakajima, M. Hirschberger, A. Kikkawa, Y. Yamasaki, H. Sagayama, H. Nakao, Y. Taguchi, T.-h. Arima, and Y. Tokura, Skyrmion lattice with a giant topological Hall effect in a frustrated triangular-lattice magnet, Science, 10.1126/science.aau0968 (2019). Page 76 of 80 arXiv:1906.04718 [physics.app-ph] [305] C. Gong, L. Li, Z. Li, H. Ji, A. Stern, Y. Xia, T. Cao, W. Bao, C. Wang, Y. Wang, Z. Q. Qiu, R. J. Cava, S. G. Louie, J. Xia, and X. Zhang, Discovery of intrinsic ferromagnetism in two-dimensional van der Waals crystals, Nature 546, 265 (2017). [306] B. Huang, G. Clark, E. Navarro-Moratalla, D. R. Klein, R. Cheng, K. L. Seyler, D. Zhong, E. Schmidgall, M. A. McGuire, D. H. Cobden, W. Yao, D. Xiao, P. Jarillo-Herrero, and X. Xu, Layer-dependent ferromagnetism in a van der Waals crystal down to the monolayer limit, Nature 546, 270 (2017). [307] N. D. Mermin and H. Wagner, Absence of Ferromagnetism or Antiferromagnetism in One- or Two-Dimensional Isotropic Heisenberg Models, Phys. Rev. Lett. 17, 1133 (1966). [308] T. Song, X. Cai, M. W.-Y. Tu, X. Zhang, B. Huang, N. P. Wilson, K. L. Seyler, L. Zhu, T. Taniguchi, K. Watanabe, M. A. McGuire, D. H. Cobden, D. Xiao, W. Yao, and X. Xu, Giant tunneling magnetoresistance in spin-filter van der Waals heterostructures, Science 360, 1214 (2018). [309] Y. Deng, Y. Yu, Y. Song, J. Zhang, N. Z. Wang, Z. Sun, Y. Yi, Y. Z. Wu, S. Wu, J. Zhu, J. Wang, X. H. Chen, and Y. Zhang, Gate-tunable room-temperature ferromagnetism in two-dimensional Fe3GeTe2, Nature 563, 94 (2018). [310] X. Wang, J. Tang, X. Xia, C. He, J. Zhang, Y. Liu, C. Wan, C. Fang, C. Guo, and W. Yang, Current-driven magnetization switching in a van der Waals ferromagnet Fe3GeTe2, arXiv preprint arXiv:1902.05794 (2019). [311] M. Alghamdi, M. Lohmann, J. Li, P. R. Jothi, Q. Shao, M. Aldosary, T. Su, B. P. T. Fokwa, and J. Shi, Highly Efficient Spin -- Orbit Torque and Switching of Layered Ferromagnet Fe3GeTe2, Nano. Lett. 19, 4400 (2019). [312] Q. Tong, F. Liu, J. Xiao, and W. Yao, Skyrmions in the Moiré of van der Waals 2D Magnets, Nano. Lett. 18, 7194 (2018). [313] Q. Li, M. Yang, C. Gong, R. V. Chopdekar, A. T. N'Diaye, J. Turner, G. Chen, A. Scholl, P. Shafer, E. Arenholz, A. K. Schmid, S. Wang, K. Liu, N. Gao, A. S. Admasu, S.-W. Cheong, C. Hwang, J. Li, F. Wang, X. Zhang, and Z. Qiu, Patterning-Induced Ferromagnetism of Fe3GeTe2 van der Waals Materials beyond Room Temperature, Nano. Lett. 18, 5974 (2018). [314] M. Gibertini, M. Koperski, A. F. Morpurgo, and K. S. Novoselov, Magnetic 2D materials and heterostructures, Nat. Nanotech. 14, 408 (2019). [315] T.-E. Park, L. Peng, X. Zhang, S. J. Kim, K. M. Song, K. Kim, M. Weigand, G. Schütz, S. Finizio, J. Raabe, J. Xia, Y. Zhou, M. Ezawa, X. Liu, J. Chang, H. C. Koo, Y. D. Kim, X. Page 77 of 80 arXiv:1906.04718 [physics.app-ph] Yu, and S. Woo, Observation of magnetic skyrmion crystals in a van der Waals ferromagnet Fe3GeTe2, arXiv preprint arXiv:1907.01425 (2019). [316] H. Wang, C. Wang, Y. Zhu, Z.-A. Li, H. Zhang, H. Tian, Y. Shi, H. Yang, and J. Li, Direct observations of chiral spin textures in van der Waals magnet Fe3GeTe2 nanolayers, arXiv preprint arXiv:1907.08382 (2019). [317] Y. Wu, S. Zhang, J. Zhang, W. Wang, Y. L. Zhu, J. Hu, K. Wong, C. Fang, C. Wan, X. Han, Q. Shao, T. Taniguchi, K. Watanabe, Z. Mao, X. Zhang, and K. L. Wang, N\'eel-type skyrmion in WTe2/Fe3GeTe2 van der Waals heterostructure, arXiv preprint arXiv:1907.11349 (2019). [318] M.-G. Han, J. A. Garlow, Y. Liu, H. Zhang, J. Li, D. DiMarzio, M. Knight, C. Petrovic, D. Jariwala, and Y. Zhu, Topological magnetic-spin textures in two-dimensional van der Waals Cr2Ge2Te6, arXiv preprint arXiv:1907.05983 (2019). [319] S. Zhang, A. K. Petford-Long, and C. Phatak, Creation of artificial skyrmions and antiskyrmions by anisotropy engineering, Sci. Rep. 6, 31248 (2016). [320] U. Ritzmann, S. von Malottki, J.-V. Kim, S. Heinze, J. Sinova, and B. Dupé, Trochoidal motion and pair generation in skyrmion and antiskyrmion dynamics under spin -- orbit torques, Nature Electronics 1, 451 (2018). [321] L. Camosi, N. Rougemaille, O. Fruchart, J. Vogel, and S. Rohart, Micromagnetics of antiskyrmions in ultrathin films, Physical Review B 97, 134404 (2018). [322] A. Raeliarijaona, R. Nepal, and A. A. Kovalev, Boundary twists, instabilities, and creation of skyrmions and antiskyrmions, Physical Review Materials 2, 124401 (2018). [323] S. Zhang, F. Kronast, G. van der Laan, and T. Hesjedal, Real-Space Observation of Skyrmionium in a Ferromagnet-Magnetic Topological Insulator Heterostructure, Nano Lett 18, 1057 (2018). [324] S. Woo, Skyrmions learn some new moves, Nature Electronics 1, 434 (2018). [325] X. Zhang, Y. Zhou, and M. Ezawa, High-topological-number magnetic skyrmions and topologically protected dissipative structure, Physical Review B 93, 024415 (2016). [326] L. Peng, Y. Zhang, M. He, B. Ding, W. Wang, H. Tian, J. Li, S. Wang, J. Cai, G. Wu, J. P. Liu, M. J. Kramer, and B.-g. Shen, Generation of high-density biskyrmions by electric current, npj Quantum Materials 2, 30 (2017). [327] L. Peng, Y. Zhang, W. Wang, M. He, L. Li, B. Ding, J. Li, Y. Sun, X. G. Zhang, J. Cai, S. Wang, G. Wu, and B. Shen, Real-Space Observation of Nonvolatile Zero-Field Biskyrmion Lattice Generation in MnNiGa Magnet, Nano. Lett. 17, 7075 (2017). Page 78 of 80 arXiv:1906.04718 [physics.app-ph] [328] W. Wang, Y. Zhang, G. Xu, L. Peng, B. Ding, Y. Wang, Z. Hou, X. Zhang, X. Li, E. Liu, S. Wang, J. Cai, F. Wang, J. Li, F. Hu, G. Wu, B. Shen, and X.-X. Zhang, A Centrosymmetric Hexagonal Magnet with Superstable Biskyrmion Magnetic Nanodomains in a Wide Temperature Range of 100 -- 340 K, Adv. Mater. 28, 6887 (2016). [329] L. Peng, Y. Zhang, M. He, B. Ding, W. Wang, J. Li, J. Cai, S. Wang, G. Wu, and B. Shen, Multiple tuning of magnetic biskyrmions using in situ L-TEM in centrosymmetric MnNiGa alloy, Journal of Physics: Condensed Matter 30, 065803 (2018). [330] X. Z. Yu, Y. Tokunaga, Y. Kaneko, W. Z. Zhang, K. Kimoto, Y. Matsui, Y. Taguchi, and Y. Tokura, Biskyrmion states and their current-driven motion in a layered manganite, Nat. Commun. 5, 3198 (2014). [331] J. C. Loudon, A. C. Twitchett-Harrison, D. Cortes-Ortuno, M. T. Birch, L. A. Turnbull, A. Stefancic, F. Y. Ogrin, E. O. Burgos-Parra, N. Bukin, A. Laurenson, H. Popescu, M. Beg, O. Hovorka, H. Fangohr, P. A. Midgley, G. Balakrishnan, and P. D. Hatton, Do Images of Biskyrmions Show Type-II Bubbles?, Adv Mater 31, e1806598 (2019). [332] Y. A. Kharkov, O. P. Sushkov, and M. Mostovoy, Bound States of Skyrmions and Merons near the Lifshitz Point, Phys Rev Lett 119, 207201 (2017). [333] A. Tan, J. Li, A. Scholl, E. Arenholz, A. T. Young, Q. Li, C. Hwang, and Z. Q. Qiu, Topology of spin meron pairs in coupled Ni/Fe/Co/Cu(001) disks, Physical Review B 94, 014433 (2016). [334] A. O. Leonov and I. Kézsmárki, Asymmetric isolated skyrmions in polar magnets with easy-plane anisotropy, Physical Review B 96, 014423 (2017). [335] R. Murooka, A. O. Leonov, K. Inoue, and J.-i. Ohe, Current-induced shuttlecock-like movement of non-axisymmetric chiral skyrmions, arXiv preprint arXiv:1812.02939 (2018). [336] V. De Alfaro, S. Fubini, and G. Furlan, A new classical solution of the Yang-Mills field equations, Phys. Lett. B 65, 163 (1976). [337] Z. F. Ezawa and G. Tsitsishvili, Skyrmion and Bimeron Excitations in Imbalanced Bilayer Quantum Hall Systems, AIP Conf. Proc. 1399, 605 (2011). [338] Z. F. Ezawa and G. Tsitsishvili, Skyrmion and bimeron excitations in bilayer quantum Hall systems, Physica E: Low-dimensional Systems and Nanostructures 42, 1069 (2010). [339] B. Göbel, A. Mook, J. Henk, I. Mertig, and O. A. Tretiakov, Magnetic bimerons as skyrmion analogues in in-plane magnets, Physical Review B 99, 060407(R) (2019). [340] T. J. Gallagher and F. B. Humphrey, Bubble‐collapse and stripe‐chop mechanism in magnetic bubble garnet materials, Appl. Phys. Lett. 31, 235 (1977). Page 79 of 80 arXiv:1906.04718 [physics.app-ph] [341] X. Z. Yu, W. Koshibae, Y. Tokunaga, K. Shibata, Y. Taguchi, N. Nagaosa, and Y. Tokura, Transformation between meron and skyrmion topological spin textures in a chiral magnet, Nature 564, 95 (2018). [342] S. Woo, Elusive spin textures discovered, Nature 564, 43 (2018). [343] M. Garst, J. Waizner, and D. Grundler, Collective spin excitations of helices and magnetic skyrmions: review and perspectives of magnonics in non-centrosymmetric magnets, J. Phys. D: Appl. Phys. 50, 293002 (2017). [344] J. Wang, Mechanical Control of Magnetic Order: From Phase Transition to Skyrmions, Annual Review of Materials Research 49, 361 (2019). [345] A. A. Kovalev and S. Sandhoefner, Skyrmions and Antiskyrmions in Quasi-Two- Dimensional Magnets, Frontiers in Physics 6, 98 (2018). Page 80 of 80
1912.07763	1	1912	2019-12-17T00:10:29	Effects of Mid-infrared Graphene Plasmons on Photothermal Heating	[ "physics.app-ph", "cond-mat.mes-hall" ]	We theoretically investigate the plasmonic heating of graphene-based systems under the mid-infrared laser irradiation, where periodic arrays of graphene plasmonic resonators are placed on dielectric thin films. Optical resonances are sensitive to structural parameters and the number of graphene layers. Under mid-infrared laser irradiation, the steady-state temperature gradients are calculated. We find that graphene plasmons significantly enhance the confinement of electromagnetic fields in the system and lead to a large temperature rise compared to the case without graphene. The correlations between temperature change and the optical power, laser spot, and thermal conductivity of dielectric layer in these systems are discussed. Our numerical results are in accordance with experiments.	physics.app-ph	physics	Effects of Mid-infrared Graphene Plasmons on Photothermal Heating Anh D. Phan,1, 2, ∗ Do T. Nga,3 Do C. Nghia,4 Vu D. Lam,5 and Katsunori Wakabayashi2 1Phenikaa Institute for Advanced Study, Artificial Intelligence Laboratory, Faculty of Information Technology, Materials Science and Engineering, Phenikaa University, Hanoi 12116, Vietnam 2Department of Nanotechnology for Sustainable Energy, School of Science and Technology, Kwansei Gakuin University, Sanda, Hyogo 669-1337, Japan 3Institute of Physics Vietnam Academy of Science and Technology, 10 Dao Tan, Ba Dinh, Hanoi 12116, Vietnam 4Hanoi Pedagogical University 2, Nguyen Van Linh Street, Vinh Phuc, Vietnam 5Graduate University of Science and Technology, Vietnam Academy of Science and Technology, 18 Hoang Quoc Viet, Hanoi, Vietnam (Dated: December 18, 2019) We theoretically investigate the plasmonic heating of graphene-based systems under the mid- infrared laser irradiation, where periodic arrays of graphene plasmonic resonators are placed on dielectric thin films. Optical resonances are sensitive to structural parameters and the number of graphene layers. Under mid-infrared laser irradiation, the steady-state temperature gradients are calculated. We find that graphene plasmons significantly enhance the confinement of electromagnetic fields in the system and lead to a large temperature rise compared to the case without graphene. The correlations between temperature change and the optical power, laser spot, and thermal conductivity of dielectric layer in these systems are discussed. Our numerical results are in accordance with experiments. Photothermal effects have been of interest in efficient solar vapor generation [1], optical data storage [2], al- ternative cancer therapy [3, 4], antibacterial activities [5], and radiative cooling [6]. The incident light excites surface plasmons of metal-like materials in photothermal agents and thus is effectively confined in the nanostruc- tures. The confinement of electromagnetic waves causes strong light-absorbance. The absorbed energy converts efficiently into heat. Manipulating the electron density or plasmonic structures in systems leads to an inhomo- geneity of electric fields and generates local heat to match with desired purposes. Typically, photothermal applica- tions have employed noble metals because of their large concentration of free electrons and high efficiency of light- to-heat conversion [7, 8]. However, large inelastic losses of noble metals limit plasmon lifetimes in metal nanos- tructures and reduce service life of optical confinement. Graphene has been recently considered as a novel plas- monic material [9 -- 11], which strongly confines electro- magnetic fields but provides relatively low loss [12]. The optical and electrical properties of graphene can be easily tuned by doping, applying external fields, and injecting charge carriers [11]. The small number of free electrons in graphene leads to occurance of plasmon resonance at the infrared regime and significantly reduces the heat dissipa- tion compared to metals. Consequently, graphene-based metamaterials are of interest and potentially display var- ious intriguing behaviors. There are several methods to investigate the perfor- mance of graphene-based devices. While experimental implementation is very expensive and simulation is time- consuming, theoretical approaches are scalable and pro- ∗Electronic address: [email protected] vide good insights into the physics underlying design of metamaterials. Furthermore, the scalability is important in generating huge amount of data for machine learning and deep learning analyses. Thus, reliable theoretical models are useful for enhancing the highly accurate fab- rications in both size and morphologies. In this letters, we present a theoretical approach to calculate plasmonic properties of graphene-based nanos- tructures and temperature distribution in the systems when irradiated by a mid-infrared laser light. The sys- tems are modeled to be similar to mid-infrared graphene detectors fabricated in Ref. [9]. These calculations al- low us to determine roles of graphene plasmons on the light confinement and heat generation. In addition, ef- fects of dielectric layers on the temperature gradient are also discussed. Our graphene-based systems as depicted in Figure 1 composed of a square lattice of graphene nanodisks on a diamond-like carbon thin film grown on a silicon sub- strate. The square lattice is characterized by a lattice period, a = 270 nm, a resonator size, D = 210 nm, and a number of graphene resonator layers, N . A thickness of the diamond-like carbon layer is h = 60 nm. Since the energy of optical phonon of the diamond is about 165 meV [13], one can expect that this energy of diamond- like carbon thin film has the same order. The value is higher than the phonon energy of SiO2 (55 meV) [14] and other conventional substrates. Consequently, the surface of diamond-like carbon films has a lower trap density which leads to be non-polar and chemically inert surface. When the size is much smaller than the wavelength of incident light, we can apply quasi-static approxima- tion to analytically determine the polarizability of a sin- gle graphene resonator [11, 15 -- 17]. To capture effects of interactions between graphene resonators on the polar- izability, the dipole approximation is employed for sim- The reflection and transmission coefficient of the graphene-based nanostructure are [11, 17] 2 t13 = r13 = √ c t12t23ei( ω ε2h) √ 1 + r12r23e2i( ω √ r12 + r23e2i( ω ε2h) √ 1 + r12r23e2i( ω c c c ε2h) ε2h) , where √ √ r0 = ε2 − √ √ ε2 + ε1 ε1 , t0 = √ √ ε1 √ 2 ε2 + , ε1 (3) , , r12 = r0 − is(1 − r0) α−1 − γ ε3 − √ √ ε22 √ √ ε2 √ ω/c ε2 + ε3 + √ r23 = 4π a2 s = ε1 , , t12 = t0 + √ 2 ε3 + √ ist0 α−1 − γ ε2 √ , ε2 t23 = γ ≈ g a3 2 ε1 + ε2 + is, (4) where c is the speed of light, g ≈ 4.52 is the net dipolar interaction over the whole square lattice, rpq and tpq are the bulk reflection and transmission coef- ficients, respectively, when electromagnetic fields strike from medium p to q. From these, we compute the trans- mission t132 ≡ t13(N )2 for N > 0 and N = 0 cor- responding to systems with and without graphene plas- monic resonators. The extinction spectra measured in the experiments is the relative difference in these trans- missions 1-t13(N )2/t13(N = 0)2. The calculations clearly determine confinement effects of electromagnetic fields due to graphene plasmons. Under infrared laser irradiation, the diamon-like car- bon layers and graphene-disk resonators absorb electro- magnetic energy and are heated up. The temperature rise in cylinderical coordinate, ∆T ≡ ∆T (ρ, z), obeys the heat diffusion equation. Since the graphene resonator is a two-dimensional material, the thermal conductivity is anisotropic. This diffusion equation in the layer including all stacked plasmonic disks is κ1(cid:107) 1 ρ d dρ ρ d∆T dρ + κ1⊥ d2∆T dz2 = p0e − 2ρ2 w2 e−ν1z, (5) (cid:18) (cid:19) where p0 is the laser power per unit volume, w is the laser spot, ν1 is the absorption coefficient, κ1(cid:107) and κ1⊥ are the effective in-plane and out-of-plane thermal con- ductivity, respectively. To consider a uniform illumina- tion condition, one simply takes w = ∞ and follows the same analysis as in Ref. [22]. These conductivities are strongly dependent on the fraction of graphene plasmons in the plasmonic layer, which is fp = πD2/4a2. By adopting Maxwell-Garnett (MG) theory for an effective medium approximation, κ1(cid:107) and κ1⊥ are κ1(cid:107),⊥ = κm (1 − fp)(κbulk(cid:107),⊥ + 2κm) + 3fpκbulk(cid:107),⊥ (1 − fp)(κbulk(cid:107),⊥ + 2κm) + 3fpκm , (6) FIG. 1: (Color online) (a) The top-down view and (b) the side view of graphene based system including structural pa- rameters. plicity. This approximation works well as a/D ≥ 1.5 [18, 19]. We suppose that the accuracy is still accept- able for a/D ≈ 1.29 in our system, which describes the practical graphene detector in Ref. [9]. α(ω) = ε1 + ε2 2 D3 ζ 2 −iωD ε1 + ε2 2σ(ω) − 1 η , (1) where ε1 = 1, ε2 = 6.25, ε3 = 11.56, ω is frequency, and ζ and η are geometric parameters calculated using the lowest-order dipole plasmon model. The analytical dependence of ζ and η on the morphology and thickness can be found in Ref. [16]. According to the random-phase approximation with zero-parallel wave vector, the N -layers graphene conduc- tivity in mid-infrared regime is [11] N e2iEF σ(ω) = π2 (ω + iτ−1) , (2) where e is the electron charge, is the reduced Planck constant, τ is the carrier relaxation time, and EF is a chemical potential. Typically, σ(ω) contains both inter- band and intraband transitions. However, in the mid- infrared regime, the intraband conductivity is completely dominant. The in-plane optical conductivity of graphene in Eq. (2) ignores contributions of interband transitions. The approximation is only valid in the low frequency regime [11]. To effectively combine Eq. (1) and (2), it is necessary to require the strong coupling condition [20, 21], where the spacing distance between graphene disks is much smaller than the diameter D. In addition, since graphene disk layers are separated by a dielectric layer [21], the vertical optical conductivity is ignored and the horizontal optical conductivity is simply additive. (a)Top-down viewaDwidthSide viewN layers(b)Diamond-like carbonSiliconhε1ε2ε3Graphene nanodiskρzκ1κ2ν1ν2L3 ≈ 630 W/m/K [23, 24] and κbulk⊥ ≈ 6 where κbulk(cid:107) W/m/K [24] are the corresponding conductivity of bulk graphene, and κm ≈ 0.6 W/m/K is the thermal conduc- tivity of medium (air). Here, we ignore effects of plasmon and phonon coupling between graphene layers. A prior work [25] indicated that the vertical coupling of graphene nanodisks is relatively strong and plays an important role in the frequency range from 140 cm−1 to 350 cm−1. How- ever, resonant frequencies in our optical spectra in Fig. 2 are greater than 0.57 eV (∼ 460 cm−1). Thus, this strong coupling between vertically stacked graphene nanodisks may not (or may) happen in this work. to Beer-Lambert's can According √ in which Qabs = estimate 4 the absorption cross section, L = N δ is the thickness of graphene nanostruc- ν1 ≈ Qabs/a2L, is √ 2πωIm(α(ω))/c ε1 + ε2 law, one (cid:18) (cid:20) 1 ρ (cid:19) (cid:21) tures with δ = 0.335 nm, and 1/(a2L) is the density of graphene resonators on the surface of the dielectric film. In the diamond-like layer, the diffusion equation is κ2 d dρ ρ d∆T dρ + d2∆T dz2 = p0e − 2ρ2 w2 e−ν1L−ν2(z−L),(7) where κ2 ≈ 0.6 W/K/m [26] is the thermal conductiv- ity of the diamond-like layer and ν2 ≈ 1.5 µm−1 is the absorption coefficient. For simplification purpose, we as- sume that the silicon substrate is kept at ambient temper- ature by contacting with a thermostat. This boundary condition was used to successfully analyze experiments in Ref. [27]. To solve these differential equations, we take the Han- kel transform of the above equations in ρ and it gives −κ1(cid:107)u2Θ(u, z) + κ1⊥ −u2Θ(u, z) + (cid:20) where ∆T (ρ, z) = (cid:82) ∞ κ2 d2Θ(u, z) dz2 d2Θ(u, z) (cid:21) dz2 ν1P0(1 − R) ν2P0(1 − R) 2π 2π = = e− u2w2 8 e−ν1z, 0 ≤ z ≤ L e− u2w2 8 e−ν1Le−ν2(z−L), L ≤ z ≤ L + h (8) (9) 0 Θ(u, z)J0(ρu)udu with J0 being the Bessel function of the first kind. P0 is the power of the incident flux. A correction factor (1 − R) implies that only the absorption and transmission component of light play a role in the heating process. The analysis is consistent with a recent work [22]. The reflection is calculated by R = r132. At ω ≈ 0.1 eV, except for R ≈ 0.376 as N = 3, R ≈ 0.3 when N changes from 0 to 10. By solving Eqs. (8) and (9), one obtains the solutions (cid:114) κ1(cid:107) κ1⊥ uz − Θ(u, z) = A1(u)e + B1(u)e (cid:114) κ1(cid:107) + κ1⊥ uz ν2P0(1 − R) 2πκ2(u2 − ν2 2 ) Θ(u, z) = A2(u)e−uz + B2(u)euz + e− u2w2 8 e−ν1Le−ν2(z−L), L ≤ z ≤ L + h ν1P0(1 − R) 2π(k1(cid:107)u2 − κ1⊥ν2 1 ) e− u2w2 8 e−ν1z, 0 ≤ z ≤ L (10) (11) where A1(u), B1(u), A2(u), and B2(u) are parameters determined by boundary conditions (cid:12)(cid:12)(cid:12)(cid:12)z=0 (cid:12)(cid:12)(cid:12)(cid:12)z=L− −κ1⊥ ∂Θ(u, z) = 0, ∂z Θ(u, L−) = Θ(u, L+), −κ1⊥ ∂Θ(u, z) ∂z = −κ2 ∂Θ(u, z) ∂z (cid:12)(cid:12)(cid:12)(cid:12)z=L+ , Θ(u, z = L + h) = 0. (12) Figure 2 shows theoretical infrared extinction spec- tra of graphene-based systems with several values of graphene plasmon layers calculated using Eqs. (3) and (4). The numerical results indicate the plasmonic peak for a square lattice of three-layers-graphene disks is roughly located at 0.1 eV (∼ 806 cm−1), which quantita- tively agrees with experiment in Ref. [9]. The presence of graphene plasmons reduces the transmission of electro- magnetic fields through these systems. An inrease of N blue-shifts the plasmonic resonance and enhances the am- plitude signal in the optical spectra. More mid-infrared optical energy is confined in the system as increasing the layer number of graphene plasmons. The amount of the trapped energy can be indirectly measured via tempera- ture caused by the light-to-heat conversion process. Other parameters can tuned to modify the extinction 4 FIG. 2: (Color online) Theoretical extinction spectra for sys- tems having a graphene-disk array with EF = 0.45 eV and τ−1 = 0.03 eV at several numbers of graphene layers. FIG. 3: (Color online) Spatial contour plots of the steady- state temperature rise in Kelvin units in graphene-based sys- tems having (a) N = 1, (b) N = 3, and (c) N = 10 under illumination of a quantum cascade laser light. spectrum. Since α(ω) ∼ D3, a reduction of the diame- ter weakens the in-plane plasmonic coupling among res- onators and lowers the optical peak when fixing a. Sim- ilar behaviors can be observed if the diameter remains unchanged and we increase the lattice period a. Mean- while due to σ(ω) ∼ EF , effects of graphene plasmons on the optical spectrum becomes less important when decreasing EF . To investigate effects of the graphene plasmon layers on the heating of graphene-disk systems, we use Eq. (10), (11) and (12) to calculate the spatial distributions of steady-state temperature when exposed by an infrared laser beam. This laser operates at ∼ 0.1 eV with the in- coming power P0 = 630 µW and w = 2.3 µm. Numerical results are shown in Fig. 3 for different N . The incident photons are highly localized at the surface and spatially decay towards the bottom of the diamond-like carbon layer. The temperature increase at the hottest spot area (z = 0 and ρ = 0) in the case of N = 10 is 45 K, while that of systems having N = 3 and N = 1 are approxi- mately 33 K and 34.5 K, respectively. These tempera- ture rises are much higher than ∆T (ρ = 0, z = 0) ≈ 1.67 K for the system without graphene plasmons. This re- sult clearly indicates that the dielectric loss is dominated by the ohmic loss (Joule heating) on graphene resonators under the mid-infrared irradiation. The temperature rise strongly depends on the inci- dent power P0 and the laser spot w. On the basis of Eqs. (10), (11) and (12), one analytically finds that ∆T (ρ, z) is proportional to P0. This finding is consis- tent with many prior works [4, 23, 28]. For N = 3, our numerical results give ∆T (0, 0) = 0.0521P0 and pre- dicts ∆T (0, 0) ≈ 330 K as P0 = 6.3 mW , which is close to ∆T ≈ 365 K [23]. Meanwhile, at fixed in- coming power, an increase of w leads to a decrease of the laser intensity (2P0/πw2). It is well-known that the absorbed optical energy of a plasmonic resonator is ap- proximately 2QabsP0/πw2. This analysis suggests that ∆Tmax ∼ 1/w2 and/or log10(∆T (0, 0)) is inversely linear to log10(w). This correlation is in a good agreement with numerical calculations. FIG. 4: (Color online) Spatial contour plots of the steady- state temperature rise in Kelvin units in graphene-based sys- tems having N = 3 and (a) κ2 = 0.6 W/m/K, (b) κ2 = 1.2 W/m/K, and (c) κ2 = 2.0 W/m/K under illumination of a quantum cascade laser light. The thermal conductivity of the thin film (κ2) be- low graphene plasmons has a significant influence on the temperature rise of graphene-based systems. To zeroth- order approximation, we assume a decoupling between κ2 and ε2. This assumption may be reasonable since the phonon scattering, which is very important in the lattice thermal conductivity, strongly depends on the interfacial roughness [29, 30]. Effects of the grain boundary on the thermal conductivity is frequency independent [29, 30]. 0.000.080.160.240.32020406080N = 20N = 15N = 10N = 5N = 31-t13(N)/t13(N=0)2 (%)(eV)N = 1-3-2-101236050403020100z (nm) (m)N = 1-3-2-10123 (m)N = 3-3-2-10123(c)(b) (m)0.09.018273645N = 10T (K)(a)-3-2-101236050403020100(c)2 = 0.6 W/m/Kz (nm) (m)(a)(b)-3-2-10123 (m)2 = 1.2 W/m/K-3-2-101232 = 2 W/m/K (m)0.0112233T (K)While a prior work [31] indicated that the grain bound- ary does not alter the dielectric and infrared response of a ceramic material at room temperature. Thus, one can modify the grain boundary scattering without changing the dielectric constant of the host material. Figure 4 shows that the steady temperature profile of the system for different values of κ2 illuminated by the mid-infrared laser light. We use the same laser as the calculations in Figure 3. The object having a larger κ2 requires more thermal energy to be heated. Thus, the temperature rise is depressed. We have proposed to investigate the plasmonic heating of graphene-based systems under irradiation of a mid- infrared laser. The nanostructures comprise a square ar- ray of multilayer graphene nanodisks deposited on the diamond-like carbon thin film, which is supported by a silicon substrate. We employ the dipole model associ- ated with the random-phase approximation to calculate the polarizability of graphene on the substance and its absorption cross section. From these, the reflection and transmission coefficients are theoretically computed to estimate the confined power of the incident light. Our numerical results are in accordance with experiments [9]. When illuminating the systems by the laser light, 5 plasmonic nanodisks absorb more optical energy than the counterparts without graphene and convert to ther- mal dissipation. This finding indicates that the Ohmic loss is much larger than the dielectric loss in the mid- infrared regime. An increase of graphene plasmonic lay- ers enhances the thermal gradients. At fixed number of graphene layers, the temperature rise is linearly pro- portional to the optical power and decays as the inverse square of the laser spot. Furthermore, a decrease of the heated temperature as increasing the thermal conductiv- ity of the thin film layer is also calcualted and discussed. Acknowledgments This research is funded by Vietnam National Founda- tion for Science and Technology Development (NAFOS- TED) under grant number 103.01-2018.337. This work was supported by JSPS KAKENHI Grant Numbers JP19F18322 and JP18H01154. Conflict of Interest: The authors declare that they have no conflict of interest. [1] K. Bae, G. Kang, S. K. Cho, W. Park, K. Kim, and W. J. Padilla, Nat. Commun. 2015, 6, 1-9. and K. L. Shepard, Nat. Nanotechnol. 2008, 3, 654-659. [15] X. Chen, Y. Chen, M. Yan, and M. Qiu, ACS Nano 2016, [2] T. S. Kao, S. D. Jenkins, J. Ruostekoski, and N. I. Zhe- 63 2550-2557. ludev, Phys. Rev. Lett. 2011, 106, 085501. [16] R. Yu, J. D. Cox, J. R. M. Saavedra, and F. J. G. de [3] S. Lal, S. E. Clare and N. J. Halas, Acc. Chem. Res. Abajo, ACS Photonics 2017, 4, 3106-3114. 2008, 41, 1842-1851. [4] V. T. T. Duong, A. D. Phan, N. T. H. Lien, D. T. Hue, D. Q. Hoa, D. T. Nga, T. H. Nhung and N. A. Viet, Phys. Status Solidi A 2018, 215, 1700564. [5] J.-W. Xu, K. Yao, and Z.-K. Xu, Nanoscale 2019, 11, 8680. [6] E. Rephaeli, A. Raman, S. Fan, Nano Lett. 2013, 13, 1457-1461. [7] X. Huang, I. H. El-Sayed, W. Qian, M. A. El-Sayed, J. Am. Chem. Soc. 2006, 128, 2115 (2006). [8] R. Huschka, J. Zuloaga, M. W. Knight, L. V. Brown, P. Nordlander, N. J. Halas, J. Am. Chem. Soc. 2011, 133, 12247. [9] Q. Guo, R. Yu, C. Li, S. Yuan, B. Deng, F. J. G. de Abajo, and F. Xia, Nat. Phys. 2018, 17, 986-992. [10] D. Rodrigo, O. Limaj, D. Janner, D. Etezadi, F. J. G. de Abajo, V. Pruneri, and H. Altug, Science 2015, 349, 165-168. [11] F. J. G. de Abajo, ACS Photonics 2014, 1, 135. [12] A. Woessner, M. B. Lundeberg, Y. Gao, A. Prin- cipi, P. Alonso-Gonzalez, M. Carrega, K. Watanabe, T. Taniguchi, G. Vignale, M. Polini, J. Hone, R. Hillen- brand, and F. H. L. Koppens, Nat. Mater. 2015, 13, 421-425. [13] Y. Wu, Y.-M. Lin, A. A. Bol, K. A. Jenkins, F. Xia, D. B. Farmer, Y. Zhu, and P. Avouris, Nature 2011, 472, 74. [17] S. Thongrattanasiri, F. H. Koppens, and F. J. G. De Abajo, Phys. Rev. Lett. 2012, 108, 047401. [18] J. Christensen, A. Manjavacas, S. Thongrattanasiri, F. H. L. Koppens, and F. J. G. de Abajo, ACS Nano 2012, 6, 431-440. [19] F. J. G. de Abajo, Rev. Mod. Phys. 2007, 79, 1267. [20] F. Ramirez, B. Liu, and S. Shen, J. Quant. Spectrosc. Radiat. Transf. 2015, 158, 27-35. [21] H. Yan, X. Li, B. Chandra, G. Tulevski, Y. Wu, M. Fre- itag, W. Zhu, P. Avouris and F. Xia, Nat. Nanotechnol. 2012, 7, 330-334. [22] A. D. Phan, N. B. Le, T. H. L. Nghiem, L. M. Woods, S. Ishii, and K. Wakabayashi, Phys. Chem. Chem. Phys. 2019, 21, 19915. [23] C. Faugeras, B. Faugeras, M. Orlita, M. Potemski, R. R. Nair, and A. K. Geim, ACS Nano 2010, 4, 1889-1892. [24] E. Pop, V. Varshney, and A. K. Roy, MRS Bulletin 2012, 37, 1273-1281. [25] H. Yan, F. Xia, Z. Li, and P. Avouris, New J. Phys. 2012, 14, 125001. [26] M. Shamsa, W. L. Liu, A. A. Balandin, C. Casiraghi, W. I. Milne, and A. C. Ferrari, Appl. Phys. Lett. 2006, 89, 161921. [27] S. Shihab, L. Thevenard, A. Lemaitre, J.-Y. Duquesne, and C. Gourdon, J. Appl. Phys. 2016, 119, 153904. [28] H. H. Richardson, M. T. Carlson, P. J. Tandler, P. Her- nandez, A. O. Govorov, Nano Lett. 2009, 9, 1139. [14] I. Meric, M. Y. Han, A. F. Young, B. Ozyilmaz, P. Kim, [29] J. Callaway, Phys. Rev. 1959, 113, 1046. [30] J. M. Ziman, Electrons and Phonons: The Theory of Transport Phenomena in Solids, Oxford University Press, Oxford, 2001. [31] J. Petzelt, T. Ostapchuk, I. Gregora, D. Nuzhnyy, I. Rychetsky, K. Maca, and Z. Shen, Ferroelectrics 2008, 363, 227-244. 6
1802.04827	1	1802	2018-02-13T19:28:35	Ultracompact integrated polarizers using bent asymmetric coupled waveguides	[ "physics.app-ph", "physics.optics" ]	A new type of integrated polarizer based on bent coupled optical waveguides is proposed. The simplicity of the device geometry permits its realization using only basic standard fabrication steps. A deep etched waveguide silicon nitride TE-pass implementation has been studied using a full vector 3D mode solver and beam propagation techniques. The characteristics of the underlying physical mechanism bring in flexibility to the design of broadband devices robust against fabrication tolerances. The possibility of incorporating almost design-transparent polarizers in photonic integrated circuits is also discussed.	physics.app-ph	physics	Ultracompact integrated polarizers using bent asymmetric coupled waveguides P. Chamorro-Posada Departmento de Teor´ıa de la Senal y Comunicaciones e Ingenier´ıa Telem´atica, Universidad de Valladolid, ETSI Telecomunicaci´on, Paseo Bel´en 15, E-47011 Valladolid, Spain. February 15, 2018 Abstract A new type of integrated polarizer based on bent coupled opti- cal waveguides is proposed. The simplicity of the device geometry permits its realization using only basic standard fabrication steps. A deep etched waveguide silicon nitride TE-pass implementation has been studied using a full vector 3D mode solver and beam propagation techniques. The characteristics of the underlying physical mechanism bring in flexibility to the design of broadband devices robust against fabrication tolerances. The possibility of incorporating almost design- transparent polarizers in photonic integrated circuits is also discussed. Polarization handling is a key issue in photonic integrated circuits (PICs) [1]. Related devices include polarization beam splitters, polarization rota- tors and polarizers. Integrated polarizers are based on different propagation mechanisms that permit to discriminate either the TE or TM field by means of a strong attenuation on the crossed polarization. The large polarization- dependent losses of the plasmonic modes can be exploited by the use of metals [2, 3, 4, 5, 6, 7, 8], transition-metal oxides with metal-insulator transitions [9], graphene [10, 11] or transparent conducting oxides [12]. Purely dielec- tric polarizers include designs based on resonant tunneling structures [13], shallowly-etched silicon-on-insulator ridge optical waveguides [14], symmet- ric etching of silicon nanowires [15, 16] or subwavelength gratings [17, 18]. In this work, a new all-dielectric polarizer architecture is proposed. It is based on bent coupled waveguides [19, 20]. These structures, originally 1 proposed for the reduction of the radiation loss in improved-Q resonators, can perform strong differential losses for TE/TM polarizations. We will consider the geometries of previous works [19, 20] based on the assumption of deep etched Si3N4 rectangular waveguide with a SiO2 cladding and a center operation wavelength of λ = 1.55 µm. The width of transmission waveguides is w = 1 µm and all the waveguides have a height of h = 300 nm. The refractive index of silicon nitride is 1.9792 and that of the oxide 1.4501. This can be regarded as a standard geometry for silicon nitride integrated circuits. Figure 1: Basic device geometry (left) and a example of a modified configu- ration (right). The waveguide under analysis will always support at least one TE-like and one TM-like propagating mode with very different group indices, modal profiles and bend losses. Therefore, any optical device implemented in the PIC will exhibit a strong polarization-dependent response. The placement of an input polarizer is a simple solution to this issue. With the specifica- tions detailed above, the bend losses are much larger for TM-polarized fields and devices will typically operate with TE optical signals. The waveguide geometry used leads naturally to a TE-pass polarizer, which is precisely the one of interest for this implementation. Alternative waveguide geometries will provide polarizers passing the verctically-polarized light with the same operation fundamentals. The device geometry is shown in Fig. 1 (left). It consists of two cou- pled bent waveguides. Optical signal transmission takes place in the inte- 2 u=0Ru=πRloffloffswwweu=0wwws1R2s2we,1loff,1we,2R1loff,2loff,3u=π(R1+R2)u=πR1Figure 2: Ratio of the imaginary modal index of the TE mode to that of the TM mode for different parameter sets (s, we) and three values of the radius of curvature R. 3 Figure 3: Coupling loss between the straight and bent waveguide sections. 4 00.10.20.30.40.5s (µ m)0.60.70.80.91Coupling coefficientTE, R=25µmTE, R=20µmTE, R=15µmTM, R=25µmTM, R=25µmTM, R=15µmrior waveguide, whereas the exterior arc is designed to control the radiation losses by adjusting the separation s and the width of the second waveguide we. Since all the waveguide structures involved in the design have a simple rectangular cross-section with the same height, the proposed devices can be fabricated using only basic standard steps. Additional coupling losses arise due to the discontinuities at the input/output connections with the straight waveguides. These are minimized by adjusting the lateral offset lof f [21]. The structure is very flexible and can be modified by using smaller-than- π sections and/or connecting several of these sections as shown in Fig. 1 (right). Table 1: Design examples R (µm) lof f (nm) s (µm) we (nm) IL (dB) ER (dB) 15 20 25 130 90 70 1.9 2.1 2.3 400 500 600 2.53 0.75 0.23 34.86 24.95 18.46 An in-depth analysis of modal properties of curved asymmetric coupled waveguides used in our polarizers has been presented in [20]. The study is based on intensive 3D full-vector numerical computations and has permitted to identify the modes that remain predominantly confined in the transmis- sion (inner) waveguide. Figure 2 displays the imaginary indices ni of these modes, relevant for the calculation of the radiation loss as calculated using the wgms3d software package [22] . The propagation constant of the mode is 2π/λ0(nr + jni). In order to visually determine the optimal operation condi- tions, the ratio of ni for the TE mode over that of ni for the TM mode over the design space (s, we) is plotted. For this waveguide geometry, this ratio is always larger that 1 and only TE-pass configurations are available. For an inverted aspect ratio of the transverse section of the waveguide, a vertically polarized output could be obtained. The optimal design regions (systemat- ically close to the left lower corner of the plots) define broad regions that provide implementations that are robust against fabrication tolerances, since small deviations in the fabrication process will have little effect on the device performance. The presence of multiple local maxima provides added flexi- bility to the design since alternative close-to-optical device configuration can be used in the case of existence of geometry restraints in the fabrication pro- cess. On the other hand, former studies [19] show that these dispersion plots have very little wavelength dependence, providing the basis for broadband designs. Silicon waveguides, with a higher refractive index contrast, permit the realization of comparable radiation losses with smaller curvature radii and, 5 therefore, are expected to permit the implementation or more compact de- vices with similar performances. Fig. 3 displays the coupling loss of TE and TM modes as a function of the lateral offset lof f for three values of the radius of curvature R. As expected, the coupling loss grows with the waveguide curvature. Also, it is significantly larger for TM polarized fields. This is due to the fact that a larger fraction of the total intensity remais outside the waveguide core for this polarization. Our polarizers will benefit from this behavior of the coupling loss, that adds to the differential polarization-dependent attenuation in the curved coupled waveguides. Table 1 summarizes the results of Fig. 2 and Fig. 3 for three polarizers consisting of a single π rad section as shown in Fig. 1 (left). For R = 15 µm we obtain a very compact device with very large extinction ratio (ER=34.86 dB) at the cost of relatively high insertion loss IL=2.53 dB. The insertion loss is largely reduced for R=20 µm, with IL=0.75 dB and still a very good value of ER. For R = 20 µm, the insertion loss is negligible, while the extinction ratio is still high. This radius could already be close to that of the curved sections that are required in the inter-device connections of any PIC. The modification of several of these curved sections, reducing its curvature radius and using the proposed scheme and, therefore, making them polarization selective can be used to exploit the properties of our proposal with minimal changes of an original circuit design. This could provide a route to almost design-transparent implementation of the proposed integrated polarizers. The transmission properties of the designed polarizers have also been studied using a 3D full-vector finite-differences beam propagation method (BPM) [23]. Fig. 4 shows the magnitude of the transverse input optical field consisting of equal-power contributions from the TE and TM polarized modes of the input straight waveguide. The polarizer output can be observed in Fig. 5, after propagation along a πR-long curved coupled waveguide system with R = 20 µm. Fig. 6 displays the evolution of the amplitudes of the transverse horizontally and vertically polarized field components in the horizontal plane crossing at the waveguide center. r is the radial coordinate measured from the waveguide center and u = Rφ, where φ is the azimuthal coordinate. z is the coordinate perpendicular to the plane defined by r and φ. The parameters used are lof f = 90 nm, s = 2.1 µm and we = 500 nm. The results are consistent with the modal analysis and they show a nearly TE-polarized output. Whereas the horizontally polarized field has survived almost intact to the propagation in the curved coupled waveguides, the vertically polarized component at the output is negligible, except for a spike close to the upper- right corner of the transmission waveguide, which has a vanishingly small contribution to the total output power. 6 Figure 4: Input transverse field in the BPM calculations. Eh and Ev corre- spond, respectively, to the horizontally and vertically polarized components of the transverse electric field intensity. 7 Figure 5: Output transverse field in the BPM calculations. Eh and Ev corre- spond, respectively, to the horizontally and vertically polarized components of the transverse electric field intensity. 8 Figure 6: Evolution of the transverse field amplitude in the horizontal plane crossing the center of the waveguides. 9 In summary, a new type of all-dielectric integrated polarizer has been put forward. The operation of the device is based on highly polarization dependent losses in coupled asymmetric curved waveguides. The simple ge- ometry of the design involve only the basic standard fabrication steps at the production stage. Large extinction ratios and small insertion losses can be obtained in small-extension devices. Alternatively, the proposed polarizers can be introduced at curved waveguides at different parts of a photonic inte- grated circuit with a minimal modification of the original design. A detailed analysis of the performance of a device implemented in the silicon nitride platform has been carried out using highly accurate full-vector mode solving techniques and 3D vector beam propagation techniques. For deep etched sili- con nitride 1 µm wide and 300 tall waveguides, a device of size 46.4 µm×23.2 µm provides 25.7 dB of extinction ratio with an insertion loss of 0.75 dB, while similar performances are expected with much more compact silicon devices. This work has been was supported by Junta de Castilla y Le´on, Project No. VA089U16, and the Spanish Ministerio de Economa y Competitividad, Project No. TEC2015-69665-R References [1] Daoxin Dai, Jared Bauters, and John E Bowers. Passive technologies for future large-scale photonic integrated circuits on silicon: polarization handling, light non-reciprocity and loss reduction. Light: Science & Applications, 1(3):e1–e12, March 2012. [2] Jihua Zhang and Eric Cassan adn Xinliang Zhang. Wideband and com- pact te-pass/tm-stop polarizer based on a hybrid plasmonic bragg grat- ing for silicon photonics. Journal of Lightwave Technology, 32(7), April 2014. [3] Guangyuan Li and Anshi Xu. Li - analysis of the te-pass or tm-pass metal-clad polarizer with a resonant buffer layer. Journal of Lightwave Technology, 26(10):1234–1241, May 2008. [4] Bowen Bai, Lu Liu, Ruixuan Chen, and Zhiping Zhou. Low loss compact tm-pass polarizer based on hybrid plasmonic grating. IEEE Photonics Technology Letters, 29(7), April 2017. [5] X. Sun, M. Mjahedi, and J.S. Aitchison. Hybrid plasmonic waveguide- based ultra-low insertion loss transverse electric-pass polarizer. Optics Letters, 41(17):4020–4023, September 2016. 10 [6] X. Sun, M.Z. Alam, S.J. Wagner, J.S. Aitchison, and M. Mojahedi. Ex- perimental demonstration of a hybrid plasmonic transverse electric pass polarizer for a silicon-on-insulator platform. Optics Letters, 37(23):4814– 4816, December 2012. [7] Tien Khee Ng, Mohammed Zahed Mustafa Khan, Ahmad Al-Jabr, and Boo. S Ooi. Analysis of cmos compatible cu-based tm-pass optical po- larizer. IEEE Photonics Technology Letters, 24(9):724–726, May 2012. [8] Yin Xu and Jinbiao Xiao. A compact te-pass polarizer for silicon-based IEEE Photonics Technology Letters, 27(19):20171– slot waveguides. 2074, October 2015. [9] L. S´anchez, S. Lechago, and P. Sanchis. Ultra-compact te and tm pass polarizers based on vanadium dioxide on silicon. Optics Letters, 40(7):1452–1455, April 2015. [10] Ran Hao, Wei Du, Er-Ping Li, and Hong-Sheng Cheng. Graphene as- sisted te/tm-independent polarizer based on machzehnder interferome- ter. IEEE Photonics Technology Letters, 27(10):1112–1115, May 2015. [11] Xiang Yin, Tian Zhang, Lin Chen, and Xun Li. Ultra-compact te-pass polarizer with graphene multilayer embedded in a silicon slot waveguide. Optics Letters, 40(8):1733–1736, April 2015. [12] Yin Xu and Jinbiao Xiao. Design and numerical study of a compact, broadband and low-loss te-pass polarizer using transparent conducting oxides. Optics Express, 24(14):15373–15382, July 2016. [13] Shaimaa I. Azzam and Salah S.A. Obayya. Ultra-compact resonant tunneling-based te-pass and tm-pass polarizers for soi platform. Optics Letters, 40(6):1061–1064, March 2015. [14] Daoxin Dai, Zhi Wang, Nich Julian, and John E. Bowers. Compact broadband polarizer based on shallowly-etched silicon-on-insulator ridge optical waveguides. Optics Express, 18(26):27404–27415, December 2010. [15] Shaimaa I.H. Azzam, Mohamed Farhat O. Hameed, Nihal F.F. Areed, Maher M. Abd-Elrazzak, H.A. El-Mikaty, and Salah S.A. Obayya. Pro- posal of an ultracompact cmos-compatible te- tm-pass polarizer based on soi platform. IEEE Photonics Technology Letters, 26(16):1633–1636, August 2014. 11 [16] Qian Wang and Seng-Tiong Ho. Ultracompact tm-pass silicon nanopho- tonic waveguide polarizer and design. IEEE Photonics Journal, 2(1):49– 56, February 2010. [17] Xiaowei Guan, Pengxin Chen, Sitao Chen, Peipeng Xu, Yaocheng Shi, and Daoxin Dai. Low-loss ultracompact transverse-magnetic-pass po- larizer with a silicon subwavelength grating waveguide. Optics Letters, 39(15):4514–4517, August 2014. [18] Dong Wook Kim, Moon Hyeok Lee, Yudeuk Kim, and Kyong Hon Kim. Ultracompact transverse magnetic mode-pass filter based on one- dimensional photonic crystals with subwavelength structures. Optics Express, 24(19):21560–, September 2016. [19] Pedro Chamorro-Posada. Q-enhanced racetrack microresonators. Optics Communications, 387:70–78, March 2017. [20] Pedro Chamorro-Posada. Radiation in bent asymmetric coupled waveg- uides. arXiv:1802.04012. [21] T. Kitoh, N. Takato, M. Yasu, , and M. Kawachi. Bending loss reduction in silica-based waveguides by using lateral offsets. Journal of Lightwave Technology, 13:555–562, April 1995. [22] M. Krause. Finite-difference mode solver for curved waveguides with angled and curved dielectric interfaces. Journal of Lightwave Technology, 29:691–699, March 2011. [23] W.W. Lui, C.-L. Xu, T. Hirono, K. Yokoyama, and W.-P-Huang. Full- vectorial wave propagation in semiconductor optical bending waveguides and equivalent straight waveguide approximations. Journal of Lightwave Technology, 16:910–914, May 1998. 12
1911.00392	1	1911	2019-11-01T14:12:27	Effect of polymeric additives on ignition, combustion and flame characteristics and soot deposits of crude oil droplets	[ "physics.app-ph" ]	Many oil fires have resulted from the crude oil train derailments in recent years. Given the importance of crude oil shipping by rail to the energy security of the US, it is important to consider various methods that will decrease the likelihood of crude oil catching fire in case of a crude oil derailment. Present study examines the effect of polybutadiene polymer on the combustion properties and soot deposits of Bakken and Pennsylvania crudes. Treating these crudes as multicomponent liquid fuels and polybutadiene as an additive, droplet combustion experiments were conducted with sub-millimeter sized spherical droplets suspended on very fine support fibers. Polybutadiene polymer additive of two different chain lengths has been investigated. Results show that both polymer chain length and origin of crude oil have a significant effect on various combustion properties like combustion rate, ignition delay, total combustion time, and flame stand-off ratio. Polymeric additives also change the soot deposit structure and particle size compared to the base fuel. Present research is envisioned to aid in theoretical combustion modeling of complex multicomponent liquid fuels, as well as generate interest in investigating more polymeric additives for liquid fuels.	physics.app-ph	physics	Effect of polymeric additives on ignition, combustion and flame characteristics and soot deposits of crude oil droplets Gurjap Singh 1, Mehdi Esmaeilpour 2, Albert Ratner 1 1 Department of Mechanical Engineering, The University of Iowa, Iowa City, IA 52242, USA 2 College of Information Technology and Engineering, Marshall University, Huntington, WV 25755, Contact: [email protected] USA Abstract: Many oil fires have resulted from the crude oil train derailments in recent years. Given the importance of crude oil shipping by rail to the energy security of the US, it is important to consider various methods that will decrease the likelihood of crude oil catching fire in case of a crude oil derailment. Present study examines the effect of polybutadiene polymer on the combustion properties and soot deposits of Bakken and Pennsylvania crudes. Treating these crudes as multicomponent liquid fuels and polybutadiene as an additive, droplet combustion experiments were conducted with sub-millimeter sized spherical droplets suspended on very fine support fibers. Polybutadiene polymer additive of two different chain lengths has been investigated. Results show that both polymer chain length and origin of crude oil have a significant effect on various combustion properties like combustion rate, ignition delay, total combustion time, and flame stand-off ratio. Polymeric additives also change the soot deposit structure and particle size compared to the base fuel. Present research is envisioned to aid in theoretical combustion modeling of complex multicomponent liquid fuels, as well as generate interest in investigating more polymeric additives for liquid fuels. Number of words: 189 Keywords: Bakken crude; Pennsylvania crude; polymer combustion; droplet combustion; burning rate; crude oil Nomenclature: 𝑑𝑑0 = initial droplet diameter 𝑑𝑑(𝑡𝑡) = the droplet diameter at time 𝑡𝑡 𝑡𝑡 𝑜𝑜𝑜𝑜 𝜏𝜏 = combustion time 𝑘𝑘 𝑜𝑜𝑜𝑜 𝐾𝐾 = burning rate 𝑀𝑀𝑛𝑛 = polymer chain length Abbreviations: CCD: charge-coupled device CMOS: complementary metal-oxide semiconductor FSR: flame stand-off ratio GC-MS: gas chromatography-mass spectrometry PBD: Polybutadiene PBD5k: Polybutadiene 𝑀𝑀𝑛𝑛=5,000 PBD200k: Polybutadiene 𝑀𝑀𝑛𝑛=200,000 1 1. Introduction: The oil boom in the North Dakota Bakken formation has presented a logistical problem of having to ship the said crude to oil refineries. The proposed Keystone XL pipeline, meant to connect the Bakken oilfields to the already-existing Keystone oil distribution system [1] has been delayed indefinitely. The shortfall in pipeline transportation has been taken up by rail. From 2010 to 2015, crude oil shipments from the Midwest to the rest of the US have steadily increased [2]. Rail infrastructure has become critical to the energy security of the United States, with rail crude transport supplying more than half the feedstock of East Coast refineries [3]. However, this extra shipping load has placed an unprecedented stress on the aging US rail infrastructure. In recent years, many oil train derailments and crashes have occurred, which generally result in devastating oil fires and a loss of life and property [4][5][6]. Such oil fires are of special relevance to the very sweet and light Bakken crude oil, which contains a large amount of easy to vaporize, easy to burn light ends [7]. It is an expensive and time-consuming proposition to upgrade US rail infrastructure to promote safer crude oil transport. This manuscript explores polymeric additives as a stopgap measure to control the burning behavior of crude in case of a crude train derailment and spillage. Previous research has investigated the addition of long chained polymers to diesel and its blends to suppress mist formation and splashing [8]. It has been found that adding long chain polymers to diesel and Jet-A droplets [9] and their surrogate blends [10], [11] slows down their burning rate and increases their ignition delay. Most notably, polymers are already being used in crude oil pipelines as drag reducing agents [12], [13]. Very few systematic crude oil combustion studies are available currently, and those that are available generally explore pool fire properties of crude oils in relation to in-situ burning. Previous work by the authors has explored the droplet combustion properties of crude oils from various US oil production regions (Bakken, Colorado, Pennsylvania, Texas) [14], as well as combustion property modification of Bakken crude oil using nano-additives [15]. It was found that Bakken, Colorado, and Pennsylvania crude oil droplet combustion rates were comparable, but Texas crude oil burned with high microexplosion intensity. Present work similarly treats Pennsylvania and Bakken crudes as multicomponent liquid fuels and aims to establish the modification to their droplet combustion properties and soot residue properties when polybutadiene (PBD) polymer of two chain lengths (5,000 and 200,000) is blended to them in various proportions. A well-established, bench-top method has been used for present work, which has previously used to establish various combustion properties for isolated spherical droplets of liquid fuels such as combustion rate [9], [10], [14] -- [18], ignition delay [14], [15], [17], and total combustion time [14], [15], [17]. It uses only a small amount of sample, promotes spherical symmetry, and yields conveniently usable data for further analysis. The crude oil used in present work was obtained from the Northeast oil production region (Pennsylvania) and the Rocky Mountain oil production region (Bakken) of the United States. Although significant variety exists in crude oil properties, Table 1 presents typical properties of Bakken crude [7] and Pennsylvania crude [19]. Additionally, Appendix A presents gas chromatography-mass spectrometry (GC-MS) data for both crude oils. Table 1. Typical properties of crudes tested Property/Crude Oil Initial Boiling Point [0C] Specific Gravity Bakken crude 21 0.815 Pennsylvania crude 30 0.832 Polybutadiene (PBD) was obtained from Sigma Aldrich. Formed from the polymerization of 1,3-butadiene (𝐻𝐻2𝐶𝐶 = 𝐶𝐶 = 𝐶𝐶𝐻𝐻 −𝐶𝐶𝐻𝐻2) monomer, PBD is a commercially available, long chain polymer with a several applications, such as making tires, golf balls, and toys. One PBD variety that was tested had chain length 𝑀𝑀𝑛𝑛=~5,000 and is referred to as PBD5k in this manuscript. The other variety had chain length 𝑀𝑀𝑛𝑛=~200,000 and is referred to as PBD200k in this work. Both PBD5k and PBD200k are readily combustible, but PBD5k is a very viscous liquid whereas PBD200k is a rubbery solid. 2 2. Experimental Method: The experimental apparatus has previously been previously detailed in the work presented by Singh et al. [14], [17], [15] which in turn was inspired from Avedisian and Callahan [20] and Bae and Avedisian [21]. Figure 1 shows the schematic of the experiment. A short summary is as follows: a sub-millimeter sized droplet of the fuel is generated using a microsyringe and suspended using three 16 µm silicon-carbide (SiC) fibers. These are fixed between six posts as shown and meet in the center. Solenoids control semi-circular hot wires of 1 mm diameter, which are positioned close to the droplet. At an appropriate time (generally 500ms) the solenoids are activated and withdraw the hot wire away. These hot wires are made of 36-gauge Kenthal with a resistance of 4Ω each and are connected to a power supply of 15 V to make them glow red-hot. Combined, they produce energy equivalent to 14.06 J in a typical experiment where they are energized for 500ms. Figure 1. Experimental setup schematic showing the location of the fuel droplet and various components [14] A microcontroller (Arduino Uno) controls the time for which the hot wires are energized and when they retract. In turn, the microcontroller is controlled by a signal from the high-speed camera. The system acts in sync: when the camera starts to acquire data, the microcontroller is activated, and it controls the hot wires and the solenoid in turn. The high-speed camera is a black and white charge-coupled device (CCD) IDT X-StreamVision XS-3 (IDT Vision, Pasadena, CA, USA) camera operated at 1000 frames per second and fitted with a 105 mm lens (Nikon AF Micro-Nikkor-F/2.8, Tokyo, Japan). Back-lighting is provided by a single bright white LED at 3.3 V. The ignition and combustion process is simultaneously recorded through a magnifying concave mirror (4.0" diameter, 9.0" focal length) by a complementary metal-oxide semiconductor (CMOS) Casio EXILIM Pro EX-F1 (Casio, Tokyo, Japan) high-speed camera operating at 600 frames/second. The CCD camera generates 8-bit grayscale image sequence with 948 x 592 pixels. The open-source digital image processing software ImageJ/Fiji [22] -- [24] was used to post-process the images captured by the CCD camera and is used to calculate the combustion rate. The CMOS camera generates 8-bit RGB video sequence with 432 x 192 pixels, which us post-processed using MATLAB ® (MathWorks, Natick, MA, USA) to generate time-series RGB images, which are then used to calculate ignition delay and total combustion time. Post-processing of CCD camera images yields the evolution of diameter 𝑑𝑑 (and therefore area 𝜋𝜋𝑑𝑑2 4⁄ ) with time 𝜏𝜏 for a given droplet. After normalizing area and 𝜏𝜏 with squared diameter of initial droplet 𝑑𝑑02, the evolution of normalized area (𝑑𝑑 𝑑𝑑0⁄ )2 vs modified time 𝜏𝜏 𝑑𝑑02⁄ was determined. Moving average was used to reduce data, which was then further post-processed to determine combustion or burning rates. 3 To set a base reference, pure crude was burned with no PBD added. Polymer was blended into the crude oil in 0.5%, 1%, 2%, 3% and 4% (w/w) concentrations using a standard lab stir plate for 24 hours. No heat was used, and the stir speed was kept consistent across all experiments. Once prepared, the fuel blend was immediately used, and was returned to the stir plate between experiments. This was done to keep the fuel blend homogeneous and to prevent any possible polymer strand agglomeration. At least 6 experiments were performed for each fuel blend, 5 with full LED backlight and 1 in low light conditions. CCD images from the low-light experiment were used to determine the flame stand-off ratio. CCD images from back-lit experiments were used for plotting time-dependent normalized area to determine combustion rate as outlined above. After post processing, the average of the combustion rates from these experiments yielded the combustion rate of that particular blend, with the standard deviation serving as the error. CMOS camera images for all 6 experiments for a given fuel blend, extracted as outlined in above paragraphs, were used to determine average ignition delay and total combustion time for each fuel blend with standard deviation serving as the error. 3. Results and Discussion: Figure 2 a-f shows the CCD camera image sequence of a Bakken crude-2% PBD5k blend droplet as it goes through various combustion regimes. Figure 3 a-e, Figure 4 a-e shows single droplet combustion regimes of various crudes and their blends, representing normalized area evolution (𝑑𝑑(𝑡𝑡)𝑑𝑑0⁄ )2 with modified time (𝜏𝜏 𝑑𝑑0⁄ )2. As reported before in literature [14] [15] the droplet combustion regime of the crude oils can divided into four categories. As the droplet is heated up, it thermally expands and the normalized droplet area is seen to increase, which is termed as Zone I or ignition delay. The end of ignition delay is marked appearance of a flame and the onset of combustion. For some time, a steady combustion regime is observed in Zone II. The droplet heats up even furhter in this zone, which causes the low-boiling crude fractions to preferentially boil out and cause violent microexplosions in Zone III. This mechanism has been detailed previously in the work of Singh et al. [14]. As these are exhausted, the microexplosion intensity decreases and the droplet burns in a steadier combustion regime in Zone IV, for which a meaningful combustion rate can be calculated. For all PBD-crude blends, an extra zone (Zone V) follows at the end of Zone IV where pure polymer or its residue is seen to burn in a very steady combustion regime, and at a markedly different combustion rate compared to Zone IV. Combustion rates for both Zone IV and Zone V have been calculated in this work. Figure 2. Time evolution of a Bakken crude-PBD5k 2% w/w blend droplet. (a) Initial configuration showing spherical fuel droplet deployed on suspending wires and surrounded by heating coils, (b) Droplet in Zone I: beginning of combustion after ignition delay, (c) Droplet in Zone II: steady combustion regime with heating coils retracted, (d) Droplet in Zone III: violent microexplosion combustion regime, (e) Droplet in Zone IV: steady combustion with low intensity microexplosion combustion regime, (f) Droplet in Zone V: polymer combustion regime Because of the highly multicomponent nature of the crudes (see Appendix A) many microexplosions throughout the combustion regime can be seen in crude oil and PBD-crude blends. A microexplosion is characterized by a rapid spike in the normalized droplet area plot, which is followed by a rapid dip. The spike represents vapor build-up inside the droplet, causing the droplet to puff. The dip occurs when the vapor is released from the droplet surface into the flame and causing a violent fragmentation of the droplet, which results in the aforementioned rapid dip in the normalized dropelt area. Adding both PBD5k and PBD200k at low concentrations reduces microexplosions but at higher concentrations the microexplosion intensity increases (Figure 3 a-e, Figure 4 a-e). However, at higher polymer concentrations PBD200k causes more violent microexplosions compared to PBD5k. 4 The following sections discuss combustion properties for crudes and their polymer blends such as combustion or burning rate, ignition delay, total combustion time, and flame stand-off ratio. Comparison of the soot deposits of these fuels, taken using a scanning electron microscope, is also presented. (a) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 ZONE II Steady Combustion Regime ZONE I Ignition Delay ZONE III ZONE IV Violent Microexplosion Combustion Regime Steady Combustion + Low-intensity Microexplosion Combustion Regime Penn pure, d0=0.886mm 0 0.5 1 2 t/d0 1.5 5 (b) 2 ) 0 d / d ( (c) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE III Violent Microexplosion Combustion Regime ZONE IV ZONE V Steady Combustion + Low-intensity Microexplosion Combustion Regime Polymer Combustion Regime Penn 5k PBD 2%, d0=0.911mm 0 0.5 1 2 t/d0 1.5 2 ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE III Violent Microexplosion Combustion Regime ZONE IV ZONE V Steady Combustion + Low-intensity Microexplosion Combustion Regime Polymer Combustio n Regime Penn 5k PBD 4%, d0=0.963mm 0 0.5 1 2 t/d0 1.5 2 6 (d) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 (e) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 ZONE I Ignition Delay ZONE III Violent Microexplosion Combustion Regime ZONE II Steady Combust ion Regime ZONE IV Steady Combustion + Low-intensity Microexplosion Combustion Regime ZONE V Polymer Combust ion Regime Penn 200k PBD 2%, d0=0.941mm 0 0.5 1 1.5 2 2.5 2 t/d0 ZONE III Violent Microexplosion Combustion Regime ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE IV Steady Combustion + Low-intensity Microexplosio n Combustion Regime ZONE V Polymer Combust ion Regime Penn 200k PBD 4%, d0=0.916mm 0 0.5 1 2 t/d0 1.5 2 7 Figure 3. Comparison of various combustion regimes as seen in typical droplet combustion experiments for (a) Pure Pennsylvania crude [14], (b) 2% PBD5k-Penn blend, (c) 4% PBD5k-Penn blend, (d) 2% PBD200k-Penn blend, (e) 4% PBD200k-Penn blend. Note the lack of Zone V in pure Penn crude burning characteristics. Polymer-crude blends at other concentrations follow a similar trend. (a) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE III Violent Microexplosion Combustion Regime ZONE IV Steady Combustion + Low- intensity Microexplosion Combustion Regime Bakken pure, d0=0.886mm 0 0.5 1 1.5 2 2.5 2 t/d0 8 (b) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 (c) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 ZONE I Ignition Delay ZONE II Steady Combusti on Regime ZONE III Violent Microexplosion Combustion Regime ZONE IV Steady Combustion + Low-intensity Microexplosion Combustion Regime ZONE V Polymer Combustion Regime Bakken 5k PBD 2% d0=0.881mm 0 0.5 1 1.5 2 2.5 2 t/d0 ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE III Violent Microexplo sion Combustion Regime ZONE IV ZONE V Steady Combustion + Low-intensity Microexplosion Combustion Regime Polymer Combusti on Regime Bakken 5k PBD 4%, d0=0.880mm 0 0.5 1 1.5 2 2.5 2 t/d0 9 (d) 2 ) 0 d / d ( 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 (e) 3 2.5 2 2 ) 0 d / d ( 1.5 1 0.5 ZONE I Ignition Delay ZONE II Steady Combust ion Regime ZONE III Violent Microexplosi on Combustion Regime ZONE IV Steady Combustion + Low-intensity Microexplosion Combustion Regime ZONE V Polymer Combust ion Regime Bakken 200k PBD 2%, d0=0.905mm 0 0.5 1 2 t/d0 1.5 2 ZONE I Ignition Delay ZONE II Steady Combusti on Regime ZONE III Violent Microexplosion Combustion Regime ZONE IV Steady Combustion + Low-intensity Microexplosion Combustion Regime ZONE V Polymer Combustion Regime Bakken 200k PBD 4%, d0=0.793mm 0 0 0.5 1 2 t/d0 1.5 2 10 Figure 4. Comparison of various combustion regimes as seen in typical droplet combustion experiments for (a) Pure Bakken crude [14], (b) 2% PBD5k-Bakken blend, (c) 4% PBD5k-Bakken blend, (d) 2% PBD200k- Bakken blend, (e) 4% PBD200k-Bakken blend. Note the lack of Zone V in pure Bakken crude burning characteristics. Polymer-crude blends at other concentrations follow a similar trend. 3.1 Burning rate of fuel droplets (1) 𝑑𝑑(𝑡𝑡)2𝑑𝑑02 =1−𝑘𝑘(𝑡𝑡𝑑𝑑02) Zone IV and Zone V combustion rates are substantially different and show very different trends as polymer Combustion characteristics in Zone IV and Zone V follow the classical 𝑑𝑑2 law [25][26]: where 𝑑𝑑0 is the initial droplet diameter (mm), 𝑑𝑑(𝑡𝑡) is the droplet diameter (mm), 𝑡𝑡 denotes the combustion time (sec), and 𝑘𝑘 is the burning rate (mm2/s). concentrations increase. Zone IV combustion rate is referred to as 𝑘𝑘1and Zone V combustion rate is referred to 𝑘𝑘2in this manuscript. 𝑘𝑘1and 𝑘𝑘2for different fuels and fuel blends has been compared in this section, but since both pure Pennsylvania and pure Bakken crudes lack a Zone V, in general only a comparison for 𝑘𝑘1can be made across all to 𝑘𝑘1. Figure 5 shows the 𝑘𝑘1burning rate trend for Pennsylvania crude oil as PBD5k and PBD200k is added, with Figure 6 showing the average change in 𝑘𝑘1. For Pennsylvania crude, PBD5k is seen to be highly effective at reducing burning rate 𝑘𝑘1with large combustion rate decrease noted at all concentrations, with the largest decrease of 26% seen at 3%w/w PBD5k. PDB200k shows a significant reduction of 10% in burning rate 𝑘𝑘1at 1%w/w, but a moderate to large increase in 𝑘𝑘1at all other blends tested with Pennsylvania crude, with the largest increase (27%) fuels and all fuel blends. Therefore, burning or combustion rate in this manuscript, unless otherwise noted, refers noted at 3%w/w. (a) 1 k e t a r g n i n r u b e d u r c a i n a v l y s n n e P ) - s 2 m m ( 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 -1 0 1 2 3 4 5 Polybutadiene Mn 5,000 Concentration (% w/w) 11 (b) 1 k e t a r g n i n r u b e d u r c a i n a v l y s n n e P ) - s 2 m m ( 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 -1 0 1 2 3 4 5 Figure 5. Effect of (a) 5k and (b) 200k chain length PBD on Pennsylvania crude oil combustion rate 𝑘𝑘1. Error bars show the standard deviation of all experiments. Polybutadiene Mn 200,000 Concentration (% w/w) (a) a i n a v l y s n n e P n i e g n a h c e g a r e v A % 1 k e t a r g n i n r u b 0.0 -5.0 -10.0 -15.0 -20.0 -25.0 -30.0 1 4 Polybutadiene Mn 5,000 Concentration (% w/w) 2 3 12 (b) a i n a v l y s n n e P n i e g n a h c e g a r e v A % 1 k e t a r g n i n r u b 40.0 30.0 20.0 10.0 0.0 -10.0 -20.0 1 4 Polybutadiene Mn 200,000 Concentration (% w/w) 3 2 Figure 6. Effect of (a) 5k and (b) 200k chain length PBD on Pennsylvania crude oil average combustion rate 𝑘𝑘1. Figure 7 shows the change in burning rate 𝑘𝑘1of Bakken crude oil droplets as PBD5k and PBD200k are added at different concentrations. Figure 8 shows the average changes in 𝑘𝑘1. Like PBD5k-Pennsylvania crude blends, a decrease in burning rate 𝑘𝑘1is noted for all PBD5k-Bakken concentrations with the largest decrease (12%) at 2%w/w, generally causes an increase in Bakken crude burning rate 𝑘𝑘1, with the largest increase of 140% noted at 3%w/w. except at 3%w/w where an increase of 5% is noted. Again, similar to PBD200k-Pennsylvania blends, PDB200k 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 ) - s 2 m m ( 1 k e t a r g n i n r u b e d u r c n e k k a B -1 (a) 0 4 Polybutadiene Mn 5,000 Concentration (% w/w) 1 2 3 5 13 (b) ) - s 2 m m ( 1 k e t a r g n i n r u b e d u r c n e k k a B 1.9 1.7 1.5 1.3 1.1 0.9 0.7 0.5 -1 0 1 2 3 4 5 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 7. Effect of (a) 5k and (b) 200k chain length PBD on Bakken crude oil combustion rate 𝑘𝑘1. Error bars show the standard deviation of all experiments. (a) n e k k a B n i e g n a h c e g a r e v A % 1 k e t a r g n i n r u b 10.0 5.0 0.0 -5.0 -10.0 -15.0 1 4 Polybutadiene Mn 5,000 Concentration (% w/w) 3 2 14 (b) n e k k a B n i e g n a h c e g a r e v A % 1 k e t a r g n i n r u b 150.0 130.0 110.0 90.0 70.0 50.0 30.0 10.0 -10.0 1 4 Polybutadiene Mn 200,000 Concentration (% w/w) 2 3 Figure 8. Effect of (a) 5k and (b) 200k chain length PBD on Bakken crude oil average combustion rate 𝑘𝑘1. As can be seen, PBD5k is much more effective at decreasing combustion rates for both Pennsylvania and Bakken crudes. In comparison, the much-longer length polymer PBD200k is much more effective in increasing combustion rates in both crudes. Polymer combustion is a complex phenomenon. Subjected to the high heat and radiation of combustion, polymers inside the liquid droplet bulk can be expected to undergo cross-linking, scissoring, and thermal degradation [27]. Thermal degradation of a polymer chain leads to rupture of primary valence bonds, which leads to a lower overall molecular weight and yields smaller components [28]. These smaller components, typically methane and 𝐶𝐶2−4 hydrocarbons [29] frequently burn faster compared to their parent components. As an unsaturated polymer, PBD is more susceptible to a thermo-oxidative degradation (such as in a combustion environment) compared to a saturated polymer [30]. It can be expected that PBD200k, because of its longer chain length and therefore more unsaturation compared to PBD5k chain, is more susceptible to thermo-oxidative degradation. It has also been established before that increasing chain length in polymers leads to an increase in thermal conductivity [31], and increased thermal conductivity has been linked before to increased droplet combustion rates [17]. Also, due to easier C-C bond scission in longer chain polymers compared to short chain polymers, PBD200k is more susceptible to thermal degradation compared to PBD5k [32]. Therefore, many factors such as increased thermal degradation and increased heat conductivity lead to PBD200k increasing overall combustion rates for its blends with either crude oil tested. PBD5k, due to its much shorter chain length, possibly suffers from smaller thermal degradation but is effective at decreasing diffusion of lighter crude ends through the liquid bulk, and decreasing surface evaporation rates from the droplet surface, leading to lower overall combustion rate for its blends with either crude. This is further substantiated in Section 3.4 of this manuscript, where a general decrease in flames stand-off ratio is observed at all PBD5k-Pennsylvania crude blends, which indicate a lowered flame speed resulting from a decrease in diffusion and surface evaporation. As remarked before, combustion regimes of pure crudes lack a Zone V, which is characterized by a sharp change in combustion rate from the preceding Zone IV. It is also seen that Zone V combustion behavior is quite uniform and shows similarities to that of a pure fuel (Figure 3). From this, we can conclude that it is mainly leftover PBD that is burning in Zone V. This is supported by the fact that more PBD is added to a blend, further away from pure crude its burning rate gets (Figure 9, Figure 10). Furthermore, SEM analysis of the combustion residue shows polymeric residue (Section 3.5) for one of the fuel blends tested (Section 3.5). Figure 9 and Figure 10 show the comparison of combustion rate 𝑘𝑘2for various fuel blends, with combustion rate of pure fuel provided for reference. For both PBD5k and PBD200k, it is seen that as the weight % of PBD 15 when heated dynamically such as in a combustion environment. The first stage is depolymerization that yields lighter, volatile products. Material that does not depolymerize cyclizes and crosslinks to form a residue that degrades in the second stage. Increasing the amount of PBD being degraded shifts the reaction towards the first stage [33], increases the combustion rate 𝑘𝑘2 also increases. This can be explained by PBD degrading by two distinct events thereby increasing the 𝑘𝑘2 combustion rates for blends that have more PBD in them. 2 k e t a r g n i n r u b e d u r c a i n a v l y s n n e P ) - s 2 m m ( (a) (b) 2 k e t a r g n i n r u b e d u r c a i n a v l y s n n e P ) - s 2 m m ( 1.6 1.4 1.2 1.0 0.8 0.6 0.4 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 -1 0 1 2 3 4 5 Polybutadiene Mn 5,000 Concentration (% w/w) -1 0 1 2 3 4 5 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 9. Effect of (a) 5k and (b) 200k chain length PBD on Pennsylvania crude oil combustion rate 𝑘𝑘2( Error bars show the standard deviation of all experiments. Combustion rate 𝑘𝑘1for pure Pennsylvania crude is provided for reference ( ). ). 16 (a) (b) 2.2 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 ) - s 2 m m ( 2 k e t a r g n i n r u b e d u r c n e k k a B ) - s 2 m m ( 2 k e t a r g n i n r u b e d u r c n e k k a B 1.9 1.7 1.5 1.3 1.1 0.9 0.7 0.5 -1 0 4 Polybutadiene Mn 5,000 Concentration (% w/w) 1 2 3 5 -1 0 1 2 3 4 5 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 10. Effect of (a) 5k and (b) 200k chain length PBD on Bakken crude oil combustion rate 𝑘𝑘2 ( bars show the standard deviation of all experiments. Combustion rate 𝑘𝑘1for pure Bakken crude is provided for reference ( ). Error ). 3.2 Ignition delay of fuel droplets Many factors influence ignition delay of fuels. There is a physical delay that is associated with the droplet heating to the point of giving off enough vapors to sustain combustion through evaporation, mixing and diffusion of combustible components through the droplet bulk, and a chemical delay [34][35][36][37]. Figure 11 shows the effect of PBD5k and PBD200k on Pennsylvania crude ignition delay. Figure 12 shows the change in average ignition delay of Pennsylvania crude oil when PBD5k and PBD200k are added at different concentrations. PBD5k causes a large decrease in ignition delay for Pennsylvania crude at all blend proportions, with the largest decrease of 26% occurring at 1% w/w. Generally, PBD200k causes an increase in ignition delay to Pennsylvania crude, with the largest increases of 6% occurring at 1% and 3% w/w. PBD5k, a readily combustible liquid, causes a decrease in ignition delay of Pennsylvania crude droplets by dissociating into more combustible smaller components upon heating. Also, possibility of PBD5k dissociation due to chemical action by components in Pennsylvania crude to produce more readily combustible smaller components 17 cannot be ruled out. Because of its larger chain length, a larger ignition delay is associated with PBD200k. However, it does decrease the average vapor pressure of the droplet, causing a net increase in ignition delay for Pennsylvania crude. 800.0 750.0 700.0 650.0 600.0 550.0 500.0 450.0 400.0 ] s m [ y a l e d n o i t i n g i e g a r e v a e d u r c a i n a v l y s n n e P Penn_neat PBD5k_Penn PBD200k_Penn -0.5 0.5 1.5 2.5 3.5 4.5 Polybutadiene Concentration (% w/w) Figure 11. Effect of 5k and 200k chain length PBD on ignition delay of Pennsylvania crude 18 (a) -5.0 n o i t i n g i e g a r e v a e d u r c a i n a v l y s n n e P n i e g n a h C % y a l e d (b) -10.0 -15.0 -20.0 -25.0 -30.0 0.5 1 2 3 4 Polybutadiene Mn 5,000 Concentration (% w/w) n o i t i n g i e g a r e v a e d u r c a i n a v l y s n n e P n i e g n a h C % y a l e d 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 -1.0 0.5 1 2 3 4 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 12. Effect of (a) 5k and (b) 200k chain length PBD on average ignition delay of Pennsylvania crude 19 Both PBD5k and PBD200k cause a large increase in ignition delay for Bakken crude (Figure 13). As can be seen from Figure 14, a maximum of 52% increase in ignition delay is observed for Bakken crude at addition of 1% PBD5k. A maximum of 42% increase can be observed in ignition delay for Bakken crude at addition of 4% PBD200k. Notably, ignition delay of pure Bakken crude is significantly smaller compared to pure Pennsylvania crude, as noted before by Singh et al. [14] because of Bakken crude having more low boiling fractions. Addition of both PBD5k and PBD200k, which are large chain length molecules, decreases vapor pressure in Bakken crude and increases the ignition delay. ] s m [ y a l e d n o i t i n g i e g a r e v a e d u r c n e k k a B 800.0 750.0 700.0 650.0 600.0 550.0 500.0 450.0 400.0 -0.5 0 0.5 Bakken_neat PBD5k_Bakken PDB200k_Bakken 3.5 4 4.5 1.5 1 3 Polybutadiene Concentration (% w/w) 2 2.5 Figure 13. Effect of 5k and 200k chain length PBD on ignition delay of Bakken crude 20 (a) y a l e d n o i t i n g i e g a r e v a e d u r c n e k k a B n i e g n a h C % 60.0 50.0 40.0 30.0 20.0 10.0 0.0 (b) y a l e d n o i t i n g i e g a r e v a e d u r c n e k k a B n i e g n a h C % 45.0 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 0.5 1 2 3 4 Polybutadiene Mn 5,000 Concentration (% w/w) 0.5 1 2 3 4 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 14. Effect of (a) 5k and (b) 200k chain length PBD on average ignition delay of Bakken crude 21 3.3 Total combustion time of fuel droplets Two main factors affect total combustion time. Aside from the obvious droplet combustion rate, microexplosions can cause liquid to escape from the droplet bulk. More intense and more frequent the microexplosions, less the total combustion time. Figure 15 shows the effect of PBD5k and PBD200k on total combustion time of Pennsylvania crude. Figure 16 shows the average change in total combustion time of Pennsylvania crude when PBD5k and PBD200k are added. Generally, PBD5k causes an increase in Pennsylvania crude total combustion time, mainly because combustion rates decline when PBD5k is added to it. The greatest increase of 3.4% is seen at 3% PBD5k. Similarly, an increase in combustion rates as well as increase in microexplosions (the latter is especially pronounced at lager PBD200k concentrations) on addition of PBD200k generally causes total combustion to decline. The greatest average total combustion time decrease of 15% is seen at 4% PBD200k. n o i t s u b m o c l a t o t e g a r e v a e d u r c a i n a v l y s n n e P ) s m ( e m i t 1700.0 1600.0 1500.0 1400.0 1300.0 1200.0 1100.0 Penn_neat PBD5k_Penn PBD200k_Penn -0.5 0 0.5 Figure 15. Effect of 5k and 200k chain length PBD on total combustion time of Pennsylvania crude 1.5 3 1 Polybutadiene Concentration (%w/w) 2.5 2 3.5 4 4.5 22 (a) l a t o t e g a r e v a e d u r c a i n a v l y s n n e P n i e g n a h C % e m i t n o i t s u b m o c (b) l a t o t e g a r e v a e d u r c a i n a v l y s n n e P n i e g n a h C % e m i t n o i t s u b m o c 4.0 3.0 2.0 1.0 0.0 -1.0 -2.0 -3.0 -4.0 -5.0 5.0 0.0 -5.0 -10.0 -15.0 -20.0 0.5 1 2 3 4 Polybutadiene Mn 5,000 Concentration (% w/w) 0.5 1 2 3 4 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 16. Effect of (a) 5k and (b) 200k chain length PBD on average total combustion time of Pennsylvania crude 23 Figure 17 shows the change in total combustion time of Bakken crude at various PBD5k and PBD200k polymer concentrations. Figure 18 shows the average change in total combustion time when PBD5k and PBD200k are added at different proportions. For PBD5k, the general trend is a decrease in total combustion time, despite the decrease in combustion rates. This can be explained by the general increase in microexplosion intensity. The greatest decrease of 8% is seen at 3% PBD5k. Adding PBD200k to Bakken crude greatly decreases total combustion time because of large increase in microexplosion intensity (especially at large PBD200k concentrations) and a large increase in combustion rates. The largest average total combustion time decrease of 28% is noted at 4% PBD200k. ) s m ( e m i t n o i t s u b m o c l a t o t e g a r e v a e d u r c n e k k a B 1600.0 1500.0 1400.0 1300.0 1200.0 1100.0 1000.0 900.0 800.0 Bakken_neat PBD5k_Bakken PBD200k_Bakken -0.5 0 0.5 1.5 1 3 Polybutadiene Concentration (% w/w) 2 2.5 3.5 4 4.5 Figure 17. Effect of 5k and 200k chain length PBD on total combustion time of Bakken crude 24 (a) l a t o t e g a r e v a e d u r c n e k k a B n i e g n a h C % e m i t n o i t s u b m o c -5.0 -5.5 -6.0 -6.5 -7.0 -7.5 -8.0 -8.5 -9.0 -9.5 -10.0 (b) l a t o t e g a r e v a e d u r c n e k k a B n i e g n a h C % e m i t n o i t s u b m o c 5.0 0.0 -5.0 -10.0 -15.0 -20.0 -25.0 -30.0 0.5 1 2 3 4 Polybutadiene Mn 5,000 Concentration (% w/w) 0.5 1 2 3 4 Polybutadiene Mn 200,000 Concentration (% w/w) Figure 18. Effect of (a) 5k and (b) 200k chain length PBD on average total combustion time of Bakken crude 25 3.4 Flame stand-off ratio (FSR) of fuel droplets FSR is defined as the ratio of instantaneous flame diameter 𝑑𝑑𝑓𝑓 to instantaneous droplet diameter 𝑑𝑑𝑑𝑑. It can reveal important physical insight into the processes at work. Two major phenomena control FSR: thermophoretic flux and Stefan flux. Thermophoretic flux is caused by the temperature gradient in the combustion zone and causes the flame to move towards the surface. Stefan flux is caused by the outgassing of the vapor from the droplet surface and causes the flame to move away from the surface. Therefore, a fuel with a smaller vapor pressure would have a smaller FSR compared to a fuel with a larger vapor pressure. Figure 19 shows the image of a 3% PBD5k-Pennsylvania crude blend in its later stages of combustion as captured by the CCD camera. Figure 20 shows the FSR for pure Pennsylvania crude vs its blends with 0.5%, 1%, 2%, 3%, and 4% PBD5k. This blend was chosen because of the significant effect of PBD5k on Pennsylvania crude on reducing its combustion rate as observed in Section 3.1 of this manuscript. Droplet Flame- front Figure 19. Flame structure of a 3% PBD5k-Pennsylvania crude fuel blend droplet undergoing combustion In general, the FSR for all PBD5k blends with Pennsylvania crude is smaller compared to pure Pennsylvania crude. This points to a general reduction in Stefan flux during crude droplet combustion when PBD5k is added. As noted before, PBD5k is a large-chain, viscous compound which when added to a light hydrocarbon mixture such as Pennsylvania crude can lead to a general decrease in vapor pressure and therefore general decrease in Stefan flux. 26 ) - ( o i t a r f f o - d n a t s e m a l F 4.50 4.30 4.10 3.90 3.70 3.50 3.30 3.10 2.90 2.70 2.50 ) - ( o i t a r f f o - d n a t s e m a l F 4.10 3.90 3.70 3.50 3.30 3.10 2.90 2.70 2.50 ) - ( o i t a r f f o - d n a t s e m a l F 4.10 3.90 3.70 3.50 3.30 3.10 2.90 2.70 2.50 0 0 0 Penn pure 0.5% 5k PBD Penn 0.5 Time (s) Penn pure 1% 5k PBD Penn 0.5 Time (s) Penn pure 2% 5k PBD Penn 0.5 Time (s) 1 1 1 (a) (b) (c) 27 (d) (e) ) - ( o i t a r f f o - d n a t s e m a l F 4.10 3.90 3.70 3.50 3.30 3.10 2.90 2.70 2.50 ) - ( o i t a r f f o - d n a t s e m a l F 4.10 3.90 3.70 3.50 3.30 3.10 2.90 2.70 2.50 0 0 Penn pure 3% 5k PBD Penn 0.5 Time (s) Penn pure 4% 5k PBD Penn 0.5 Time (s) 1 1 Figure 20. Comparison of flame stand-off ratios (FSR) for Pennsylvania crude oil at (a) 0.5%, (b) 1%, (c) 2%, (d) 3%, and (e) 4% w/w polymer blends 3.5 Soot residue analysis Soot residue can reveal important information about the combustion process, and determination of particle size can assist in selecting appropriate respiratory personal protection equipment for fire-fighting parties that respond to crude oil fires. Figure 19 shows the flame structure of a fuel blend droplet undergoing combustion, where soot incandescence is visible in the flame. All crudes and crude blends tested in this work left a soot residue on the thin supporting fibers, and scanning electron microscope (SEM) analysis of several pure crudes has been presented before by the authors [14]. The same equipment (Hitachi S-4800 SEM) located at University of Iowa Central Microscopy Research Facility was used to test soot residue samples for this work. Due to limited equipment availability, only soot residue from 1%, 2%, 3%, 4% PBD5k blend with Bakken crude as a representative polymer- crude blend was analyzed. To take an SEM image, the fiber with the residue is mounted on a specialized tape with high electrical conductivity. The soot is gently tapped with a fine-tipped needle to break the soot deposit (Figure 21) so the soot structure at the surface and in the bulk of the soot can be analyzed. The sample is then loaded into the SEM machine and images at various magnifications are taken. 28 Soot SiC fiber Figure 21. SEM images of 1% PBD5k-Bakken blend soot residue, showing soot structure at various levels of magnification. Individual soot particles can be seen It was reported before that pure Bakken crude leaves behind a loose and spongy soot structure [14] with average particle size ~70µm. However, for all PBD5k-crude blends the structure of the soot is observed to be closely packed, with average particle size ~40µm (Figure 22). Additionally, imaging of the soot structure surface reveals globular polymeric structure embedded in the soot residue (Figure 23) in one of the samples. This is unburned polymeric residue from the polymer-only combustion in Zone V as discussed in Section 3 of this manuscript. Ultimately, soot structures and individual soot particle sizes are dependent on the chemical make-up of the fuel or fuel blend. 29 (a) (b) 30 (c) (d) Figure 22. Soot particle size comparison for (a) 1%, (b) 2%, (c) 3%, and (d) 4% PBD5k-Bakken crude blends 31 Polymeric structure Soot structure Figure 23. SEM images of the soot deposit if 1% PBD5k-Bakken blend soot residue, showing globular and stringy polymeric residue structure (shown in magnified images on the right) in the soot structure left behind 4. Conclusions: This work explored the experimental investigation of isolated single droplet burning behavior of two US crude oils sourced from Pennsylvania and North Dakota (Bakken), and the modification of this burning behavior using polybutadiene (PBD) polymer of chain length 5,000 (PBD5k) and 200,000 (PBD200k). Treating the crude oil as a multicomponent liquid fuel, PBD was added at different mass concentrations to make fuel blends. Sub-millimeter droplets of the fuels and fuel blends were burned to completion, and the process was recorded using a CCD high speed camera and a CMOS high speed camera. CCD camera photography with a single LED as a backlight revealed time evolution of droplet area, whereas low-light photography with the same camera revealed the flame stand-off ratio. CMOS camera photography revealed ignition delay and total combustion time. Scanning electron microscopy imaging of the soot deposits left behind by the burned droplets revealed the soot structure and soot particle size. Significant changes in all these combustion and soot properties were observed. These are summarized below: • A decrease in combustion rate is seen for Pennsylvania crude when PBD5k is added, with the largest decrease of 26% occurring at 3%w/w PBD5k. PBD200k generally increases the combustion rate in Pennsylvania crude, with a maximum increase of 27% noted at 3%w/w. A general decrease in combustion rate is noted for Bakken crude when PBD5k is added, with the largest decrease of 12% occurring at 2% w/w. PBD200k is noted to generally increase combustion rate in Bakken crude, with the largest increase of 140% noted at 3%w/w. Both PBD5k and PBD200k can be expected to decrease vapor pressure and inhibit diffusion of lighter components to the surface to decrease combustion rate, but faster thermal degradation and increased thermal conductivity of PBD200k result in an overall increase in combustion rate, thereby making PBD5k as a much more likely additive for crude oil transportation safety. • A decrease in ignition delay is noted for Pennsylvania crude when PBD5k is added, with the largest decrease of 26% occurring at 1% w/w. Generally, addition of PBD200k to Pennsylvania crude causes an increase in ignition delay, with the largest increases of 6% occurring at 1% and 3% w/w. This can be explained by a larger ignition delay associated with PBD200k because of its larger chain length. A large increase in ignition delay for Bakken crude is noted on addition of both PBD5k and PBD200k. A maximum of 52% increase in ignition delay is observed for Bakken crude at addition of 1% PBD5k. A maximum of 42% increase can be observed in ignition delay for Bakken crude at addition of 4% PBD200k. This can be explained by Bakken crude having a lower ignition delay compared to Pennsylvania crude, and the addition of both PBD5k and PBD200k resulting in an overall decrease in ignition delay because of decreased vapor pressure. 32 • PBD5k causes a general increase in Pennsylvania crude total combustion time, with the greatest increase of 3.4% seen at 3% PBD5k. This is caused by the decrease in combustion rate, as noted above. Addition of PBD200k generally causes total combustion to decline with the greatest average total combustion time decrease of 15% seen at 4% PBD200k. This is caused by high intensity microexplosions at high PBD200k concentrations, which cause loss of liquid fuel. PBD5k causes a general decrease in Bakken crude total combustion time, with the greatest decrease of 8% seen at 3% PBD5k. Adding PBD200k to Bakken crude greatly decreases total combustion time with the largest average total combustion time decrease of 28% noted at 4% PBD200k. This can be explained by increased microexplosion intensity in Bakken crude when both PBD5k and PBD200k are added. • Generally, a decrease in flame stand-off ratio (FSR) is noted when PBD5k is added to Pennsylvania crude. This is explained by the decrease in vapor pressure of crude oil with the addition of polymer, as noted above. • SEM imaging of the soot deposits revealed that PBD5k-Bakken blends leave behind a more densely packed structure compared to pure Bakken. Individual average soot particle for PBD5k-Bakken blends was found to decrease to ~40µm from ~70µm in case of pure Bakken crude. Additionally, imaging of the soot structure surface revealed globular polymeric structure embedded in the soot residue. Ultimately, soot structure of combustion residue is determined by the chemical make-up of the fuel. Properties such as burning rate are envisioned to be used for validation of computational and numerical modeling of multicomponent fuel droplet combustion process. Practically, decreased combustion rates achieved for PBD5k blends are expected to be useful for better crude transportation safety, whereas increased combustion rates achieved for PBD200k blends are expected to be useful for achieving better in-situ burning efficiency for crude oil spills. It is expected that present research will also aid in exploration of other polymeric additives to improve crude oil combustion safety. Acknowledgements: This research is funded, in part, by the Mid-America Transportation Center via a grant from the U.S. Department of Transportation's University Transportation Centers Program, and this support is gratefully acknowledged. The USDOT UTC grant number for MATC is: 69A3551747107. The authors would also like to acknowledge use of the University of Iowa Central Microscopy Research Facility, a core resource supported by the Vice President for Research & Economic Development, the Holden Comprehensive Cancer Center and the Carver College of Medicine. We would especially like to thank Dr Jianqiang Shao for his help and patience. We would also like to thank Prof. Lynn M. Teesch and Mr. Vic R. Parcell for their help and support with the GC-MS data. The contents reflect the views of the authors, who are responsible for the facts and the accuracy of the information presented herein and are not necessarily representative of the sponsoring agencies, corporations or persons. References: [1] [2] [3] [4] [5] U. States Department of State Bureau of Oceans and I. Scientific Affairs, "Final Supplemental Environmental Impact Statement for the Keystone XL Project," 2014. US Energy Information Administration, "Rail helps make Midwest a net shipper of crude oil - Today in Energy - U.S. Energy Information Administration (EIA)." [Online]. Available: https://www.eia.gov/todayinenergy/detail.php?id=22432. [Accessed: 06-Jul-2019]. US Energy Information Administration, "Crude by rail accounts for more than half of East Coast refinery supply in February - Today in Energy - U.S. Energy Information Administration (EIA)." [Online]. Available: https://www.eia.gov/todayinenergy/detail.php?id=21092. [Accessed: 06-Jul-2019]. C. Nunez, "Oil Train Derails in Lynchburg, Virginia." [Online]. Available: https://news.nationalgeographic.com/news/energy/2014/04/140430-oil-train-derails-in-lynchburg-virginia/. [Accessed: 06-Jul-2019]. Lisa Riordan Seville; Tracy Connor; Sean Federico-O'Murchu, "Heimdal, North Dakota, Evacuated After Fiery Oil Train Crash." [Online]. Available: https://www.nbcnews.com/news/us-news/heimdal-north- dakota-evacuated-after-fiery-oil-train-crash-n354686. [Accessed: 06-Jul-2019]. 33 [6] [7] [8] T. Hernandez, "Oil train derails near Mosier in Oregon's Columbia River Gorge - oregonlive.com." [Online]. Available: https://www.oregonlive.com/pacific-northwest- news/2016/06/oil_train_derails_near_hood_ri.html. [Accessed: 06-Jul-2019]. J. R. Auers, P. E. Ryan, M. Couture, and D. L. Sutton, "The North Dakota Petroleum Council Study on Bakken Crude Properties Bakken Crude Characterization Task Force Prepared for the By," 2014. Y. Zhang, "A generalized model for hydrocarbon drops spreading on a horizontal smooth solid surface," University of Iowa, 2012. [9] M. Ghamari and A. Ratner, "Combustion characteristics of diesel and Jet-A droplets blended with polymeric additive," Fuel, vol. 178, no. March, pp. 63 -- 70, 2016. [10] M. Ghamari and A. Ratner, "Experimental study of combustion of decane, dodecane and hexadecane with polymeric and nano-particle additives," Bull. Am. Phys. Soc., vol. Volume 60, Number 21, 2015. [11] M. Ghamari, "EXPERIMENTAL STUDY OF COMBUSTION OF POLYMER ADDED N-DECANE AND N-DODECANE DROPLETS," in ASME 2015 International Mechanical Engineering Congress and Exposition, 2017, pp. 1 -- 6. [12] A. Ram, E. Finkelstein, and C. Elata, "Reduction of Friction in Oil Pipelines by Polymer Additives," Ind. Eng. Chem. Process Des. Dev., vol. 6, no. 3, pp. 309 -- 313, Jul. 1967. [13] G. K. Patterson, J. L. Zakin, and J. M. Rodriguez, "DRAG REDUCTION - Polymer Solutions, Soap Solutions, and Solid Particle Suspensions in Pipe Flow," Ind. Eng. Chem., vol. 61, no. 1, pp. 22 -- 30, Jan. 1969. [14] G. Singh, M. Esmaeilpour, and A. Ratner, "Investigation of Combustion Properties and Soot Deposits of Various US Crude Oils," Energies, vol. 12, no. 12, p. 2368, Jun. 2019. [15] G. Singh, M. Esmaeilpour, and A. Ratner, "Effect of carbon-based nanoparticles on the ignition, combustion and flame characteristics of crude oil droplets," Energy (under Rev., 2019. [16] M. Ghamari and A. Ratner, "Combustion characteristics of colloidal droplets of jet fuel and carbon based nanoparticles," Fuel, vol. 188, pp. 182 -- 189, 2017. [17] G. Singh, M. Esmaeilpour, and A. Ratner, "The effect of acetylene black on droplet combustion and flame regime of petrodiesel and soy biodiesel," Fuel, vol. 246, pp. 108 -- 116, Jun. 2019. [18] G. Singh, N. Hentges, D. Johnson, and A. Ratner, "Experimental investigation of combustion behavior of biodiesel-water emulsion (in press)," in International Mechanical Engineering Congress and Exposition, 2019. [19] N. A. C. Smith and E. C. Lane, "Tabulated Analyses of Representative Crude Petroleums of the United States," Washington DC, 1928. [20] C. T. Avedisian and B. J. Callahan, "Experimental study of nonane/hexanol mixture droplet combustion without natural or forced convection," Proc. Combust. Inst., vol. 28, no. 1, pp. 991 -- 997, Jan. 2000. J. . Bae and C. . Avedisian, "Experimental study of the combustion dynamics of jet fuel droplets with additives in the absence of convection," Combust. Flame, vol. 137, no. 1 -- 2, pp. 148 -- 162, Apr. 2004. J. Schindelin et al., "Fiji: an open-source platform for biological-image analysis," Nat. Methods, vol. 9, no. 7, pp. 676 -- 682, Jul. 2012. J. Schindelin, C. T. Rueden, M. C. Hiner, and K. W. Eliceiri, "The ImageJ ecosystem: An open platform for biomedical image analysis," Mol. Reprod. Dev., vol. 82, no. 7 -- 8, pp. 518 -- 529, Jul. 2015. [21] [22] [23] [24] C. A. Schneider, W. S. Rasband, and K. W. Eliceiri, "NIH Image to ImageJ: 25 years of image analysis.," Nat. Methods, vol. 9, no. 7, pp. 671 -- 5, Jul. 2012. 34 [25] G. S. Jackson and C. T. Avedisian, "Combustion of unsupported water-in-n-heptane emulsion droplets in a convection-free environment," Int. J. Heat Mass Transf., vol. 41, no. 16, pp. 2503 -- 2515, Aug. 1998. I. Glassman, R. A. Yetter, and N. Glumac, Combustion. Elsevier, 2014. [26] [27] W. J. Rogers, "The effects of sterilization on medical materials and welded devices," Join. Assem. Med. Mater. Devices, pp. 79 -- 130, Jan. 2013. [28] S. S. Stivala and L. Reich, "Structure vs stability in polymer degradation," Polym. Eng. Sci., vol. 20, no. 10, pp. 654 -- 661, Jul. 1980. [29] F. Chen and J. Qian, "Studies on the thermal degradation of polybutadiene," Fuel Process. Technol., vol. 67, no. 1, pp. 53 -- 60, Jun. 2000. [30] S. Ray and R. P. Cooney, "Thermal Degradation of Polymer and Polymer Composites," Handb. Environ. Degrad. Mater., pp. 213 -- 242, Jan. 2012. [31] X. Wei and T. Luo, "Chain length effect on thermal transport in amorphous polymers and a structure -- thermal conductivity relation," Phys. Chem. Chem. Phys., vol. 21, no. 28, pp. 15523 -- 15530, Jul. 2019. [32] V. D. Knyazev, "Effects of Chain Length on the Rates of C−C Bond Dissociation in Linear Alkanes and Polyethylene†," 2007. [33] D. W. Brazier and N. V. Schwartz, "The effect of heating rate on the thermal degradation of polybutadiene," J. Appl. Polym. Sci., vol. 22, no. 1, pp. 113 -- 124, Jan. 1978. [34] H. Y. Setyawan, M. Zhu, Z. Zhang, and D. Zhang, "Ignition and combustion characteristics of single droplets of a crude glycerol in comparison with pure glycerol, petroleum diesel, biodiesel and ethanol," Energy, vol. 113, pp. 153 -- 159, Oct. 2016. [35] S. S. Sazhin, W. A. Abdelghaffar, E. M. Sazhina, and M. R. Heikal, "Models for droplet transient heating: Effects on droplet evaporation, ignition, and break-up," Int. J. Therm. Sci., vol. 44, no. 7, pp. 610 -- 622, Jul. 2005. [36] D. Zhang and T. F. Wall, "Ignition of coal particles: the influence of experimental technique," Fuel, vol. 73, no. 7, pp. 1114 -- 1119, Jul. 1994. [37] D.-K. Zhang, "Laser-induced ignition of pulverized fuel particles," Combust. Flame, vol. 90, no. 2, pp. 134 -- 142, Aug. 1992. Appendix A. Gas Chromatography -- Mass Spectrometry (GC-MS) data for Pennsylvania and Bakken crude oils GC-MS data for Pennsylvania (Figure A1) and Bakken crude (Figure A2) oils was generated at the High Resolution Mass Spectrometry Facility (HRMSF) located at the University of Iowa Department of Chemistry. Bakken crude oil GC-MS data has been previously reported by Singh et al. [15]. The column used was a 30 m DB- 5MS, 0.25mm diameter and 0.25µm film thickness. The temperature ramp started at 50 ⁰C and held for 1 min. It was then increased at 10 ⁰C/minute until 320 ⁰C and then held for 5 minutes. The numbers on top of the peaks in the chromatogram are retention time, area, and response height. The results reveal that both crudes are rich in low- and medium-boiling components, with Bakken being richer in low-boiling components compared to Pennsylvania crude. 35 Figure A1. Gas Chromatography -- Mass Spectrometry (GC-MS) data for Pennsylvania crude oil. 36 Figure A2. Gas Chromatography -- Mass Spectrometry (GC-MS) data for Bakken crude oil. 37
1810.02101	1	1810	2018-10-04T08:48:09	Incorporation of macroscopic heterogeneity within a porous layer to enhance its acoustic absorptance	[ "physics.app-ph" ]	We seek the response, in particular the spectral absorptance, of a rigidly-backed periodically-(in one horizontal~~ direction) ~inhomogeneous ~layer ~composed ~of ~alternating rigid and macroscopically-homogeneous porous portions, submitted to an airborne acoustic plane body wave. The rigorous theory of this problem is given and the means by which the latter can be numerically solved are outlined. At low frequencies, a suitable approximation derives from one linear equation in one unknown. This approximate solution is shown to be equivalent to that of the problem of the same wave incident on a homogeneous, isotropic layer. The thickness $h$ of this layer is identical to that of the inhomogeneous layer, the effective complex body wave velocity therein is identical to that of the porous portion of the inhomogeneous layer, but the complex effective mass density, whose expression is given in explicit algebraic form, is that of the reference homogeneous macroscopically-porous layer divided by the filling factor (fraction of porous material to the total material in one grating period). This difference of density is the reason why it is possible for the lowest-frequency absorptance peak to be higher than that of a reference layer. Also, it is shown how to augment the height of this peak so that it attains unity (i.e., total absorption) and how to shift it to lower frequencies, as is required in certain applications.	physics.app-ph	physics	Incorporation of macroscopic heterogeneity within a porous layer to enhance its acoustic absorptance Armand Wirgin ∗ June 22, 2021 Abstract direction) layer inhomogeneous We seek the response, in particular the spectral absorptance, of a rigidly-backed periodically- (in one horizontal composed of alternating rigid and macroscopically-homogeneous porous portions, submitted to an airborne acoustic plane body wave. The rigorous theory of this problem is given and the means by which the latter can be numerically solved are outlined. At low frequencies, a suitable approximation derives from one linear equation in one unknown. This approximate solution is shown to be equivalent to that of the problem of the same wave incident on a homogeneous, isotropic layer. The thickness h of this layer is identical to that of the inhomogeneous layer, the effective complex body wave velocity therein is identical to that of the porous portion of the inhomogeneous layer, but the complex effective mass density, whose expression is given in explicit algebraic form, is that of the reference homogeneous macroscopically-porous layer divided by the filling factor (fraction of porous material to the total material in one grating period). This difference of density is the reason why it is possible for the lowest-frequency absorptance peak to be higher than that of a reference layer. Also, it is shown how to augment the height of this peak so that it attains unity (i.e., total absorption) and how to shift it to lower frequencies, as is required in certain applications. Keywords: dynamic response, absorptance, inhomogeneous medium, porous component, grating, effective medium, . Abbreviated title: An effective layer that absorbs like a grating when submitted to sound Corresponding author: Armand Wirgin, e-mail: [email protected] ∗LMA, CNRS, UMR 7031, Aix-Marseille Univ, Centrale Marseille, F-13453 Marseille Cedex 13, France. 1 Contents 1 Introduction 4 2 Response of the macroscopically-inhomogeneous layer to an airborne plane-wave solicitation 2.1 Description of the configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Effective-medium behavior of the macroscopically-homogeneous lossy filler material . 6 6 8 2.2.1 The foam parameters in [37] and the associated effective density and wavespeed 9 2.2.2 The foam parameters in [41] and the associated effective density and wavespeed 11 2.3 The boundary-value problem for the acoustic response of the macroscopically-inhomogeneous layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.4 Field representations via domain decomposition and separation of variables (DD-SOV) 12 2.5 Exact solutions for the unknown coefficients . . . . . . . . . . . . . . . . . . . . . . . 13 2.6 Linear system for the set of unknown coefficients . . . . . . . . . . . . . . . . . . . . 14 m } . . . . . . . . . . . . . 15 2.8 Conservation of flux for the periodic structure . . . . . . . . . . . . . . . . . . . . . . 15 2.9 Numerical comparison of the M = 0, 1, 2 approximations of the grating structure 2.7 Numerical issues concerning the system of equations for {a[1] response to their exact counterpart for the filler material in [37] . . . . . . . . . . . . 17 3 Response of a macroscopically-homogeneous lossy layer to airborne sound 21 3.1 Why study this problem? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 3.2 Description of the configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 3.3 The boundary-value problem of the response of the rigidly-backed layer structure to a plane wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 3.4 Conservation of flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 3.5 Exact solution for the unknown coefficients . . . . . . . . . . . . . . . . . . . . . . . 23 3.6 Origin and meaning of the so-called quarter-wavelength resonances . . . . . . . . . . 24 4 Analytical aspects of the M = 0 approximation of the response of the grating structure 4.1 From the M = 0 approximate solution to an effective medium representation of the 27 response of the grating configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 4.1.1 General considerations on inverse problems such as ours . . . . . . . . . . . . 28 4.2 The methods of solution of the inverse problem adopted herein . . . . . . . . . . . . 29 4.2.1 First solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.2.2 Second solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 4.2.3 Third solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 4.2.4 Further comments on consequences of the M = 0 approximation of grating response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 5 Numerical results for the foam filler of [37]: comparison of the w/d < 1 grating quasi-exact and effective-layer response to the foam-filled layer response 33 2 6 Increasing the absorption at low frequencies treated as an optimization problem 37 6.1 Analytical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 6.2 Numerical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 6.2.1 Decreasing the frequency for optimal absorption by increasing h for the foam of [37] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 6.2.2 Decreasing the frequency for optimal absorption by increasing h for the foam of [41] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 6.3 Persistence of the enhanced absorption effect for other-than-normal incidence . . . . 44 6.4 Numerical results for the M > 0 grating response concerning the effect of increasing w and d for given Φopt to see if optimal absorption is maintained or changed otherwise 45 6.4.1 Φopt = 0.7, h = 0.06 m, and the foam of [37] . . . . . . . . . . . . . . . . . . 45 6.4.2 Φopt = 0.1, h = 0.012 m, and the foam of [41] . . . . . . . . . . . . . . . . . . 49 6.4.3 Comments on the 'red shift' of the total absorption peaks . . . . . . . . . . . 52 . . . 53 6.5 Homothetic increase of h,w and d for Φopt = 0.4, h/d = 1, and the foam of [37] 7 Conclusion 54 3 1 Introduction The absorption of waves (hydrodynamic waves, optical or other electromagnetic waves, acous- tic waves in fluids, elastic waves in solids) is an ongoing major research topic. Four applicative subtopics illustrate this importance: i) harvesting of energy [64], particularly that coming from solar irradiation [89, 45, 22, 93, 71, 35, 47, 88, 96, 68, 73, 82, 92, 10, 34, 103, 26, 66, 54, 56], ii) breaking (i.e., dissipating the energy of) water waves in rising oceans to protect harbors, coastal shores and constructions [58, 30, 75, 94, 76, 103, 2, 104, 65], iii) sound level reduction in increas- ingly noisy open environments and reverberation reduction in enclosures [23, 6, 63, 5, 7, 38, 67, 95, 46, 40, 39, 42, 53, 44, 8, 49], iv) mitigation of earthquake effects in nuclear power plants and cities [36, 91, 19, 69, 1, 1, 100, 98, 99]. For all of these problems, solutions have been explored in which the principal dissipative agent is a natural or man-made film, barrier, layer, or solid object (e.g., plate) or medium (e.g., portion of the underground) of finite volume in (or on) which inclusions or voids are placed and arranged either in random or periodic manner. Naturally, the interaction of waves with these heterogeneous media has given rise to a considerable amount of theoretical/numerical research to try to predict and optimize the outcome of the interaction. The present contribution is of this sort and concentrates on the acoustic wave absorption problem. Deliberate research on the latter topic probably begins in 1907 when Rayleigh [79] elaborates the first mathematical (called by him 'dynamical') theory of the diffraction by periodically-uneven (in one direction) surfaces and interfaces (note that the region between the lowest and highest points of such a structure can be considered to be an inhomogeneous layer, comprised alternatively of the materials in the upper and lower media), called 'gratings', applicable at all frequencies and for all grating periods. After showing that the high frequency (Fresnel theory) approximation of grating response predicts that none of the incident energy is thrown into the zeroth (specular) reflected order for certain rigid and pressure release gratings, he shows that his own theory predicts, for gratings whose grooves (called by him 'corrugations') have any shape, and whatever be the nature (rigid or pressure release) of the grating material, that all of the incident energy is thrown into the specular reflected order when the grating period is inferior to the wavelength of the incident sound. Rayleigh concludes that there is a need for more detailed examination of what actually goes on near the grating surface (his previous finding applying only to the grating orders which are far-field entities) when its period is small relative to, or near, the sound wavelength λ. In another contribution [80], Rayleigh writes: "The above investigation (i.e., that in [79]) is limited to the case where the second medium is impenetrable, so that the whole energy of the incident wave is thrown back in the regularly reflected wave and in the diffracted spectra. It is an interesting question whether the conclusion that corrugations of period less than λ have no effect can be extended so as to apply when there is a wave regularly transmitted. It is evident that the principle of energy does not suffice to decide the question, but it is probable that the answer should be in the negative. If we suppose the corrugations of given period to become very deep and involved, it would seem that the condition of things would at last approach that of a very gradual transition between the media, in which case the reflection tends to vanish". As far as we know, Rayleigh's intuition (now known as the 'moth's eye principle' [89, 93, 54, 20]) was neither given a theoretical justification by him, nor even exploited until the advent of anechoic chambers incorporating walls and ceilings with deeply-corrugated linings covering porous materials. It has long been known (empirically) that many naturally- porous materials and media such as turf, dry sand, ashes, asbestos and especially snow [51], constitute excellent absorbers of audible 4 sound. Porous man-made materials such as carpets, fibre glass, and rubber-like or plastic foams also have this feature and have long been incorporated in living quarters, work spaces and concert halls to reduce the reverberation of, and thus dissipate, audible sounds, be the latter agreeable or disagreeable (noise). Anechoic chambers (see [29, 83, 90, 57, 16, 28, 52] for electromagnetic microwave absorption materials, notably, but not exclusively, destined to anechoic chambers) in- corporate rather thick linings of manufactured porous materials, but this is usually not sufficient to reduce the reverberations, notably in the low frequency range, to acceptable levels, whence the idea of superposing periodic arrays of absorbing objects such as pyramids on the porous lining so as to provoke reduced reflection (and increased absorption, hopefully for all incident angles and over substantial bandwidths) by what may be thought to be the moth's eye effect [54]. Perhaps the first attempt at explaining the acoustic moth's eye effect is due to De Bruijn [24] who carries out a rigorous analysis of the interaction of a plane acoustic body wave with a lamellar grating (periodic in one direction, grooves of rectangular shape), the boundary of which is composed of rigid vertical strips and absorbing (simulated by an impedance boundary condition) horizontal strips. The analysis resides on the Rayleigh plane wave expansion in the air half space (wherein propagates the incident wave) and a modal representation of the pressure field in the grooves. De Bruijn is able to solve for the scattering amplitudes and reflectance (and thus the absorptance via the principle of conservation of energy) even when the grating period and groove depths are of the order of the wavelength and shows that the absorptance depends considerably not only on the angle of incidence but also on the period and dimensions of the grooves. He concludes that the whole matter of absorption afforded by this structure is completely governed by three physical phenomena: 1. an anomaly due to the emergence of a new spectral order at grazing angle (Wood anomaly [60], characterized by sudden changes in the reflectance as a function of wavelength and explained by Rayleigh as occurring when a body plane wave in the plane wave representation changes to an evanescent plane wave); 2. resonances related to the guided complex wave supportable by the grating (surface resonance); 3. resonances related to the depth of the groove (which is a sort of waveguide). This conclusion (the optical analogs of which are [33, 32, 70, 13, 55]), especially the last one, constitutes an important first step in the comprehension of why corrugations can modify the absorption of otherwise flat absorbing surfaces. Bos et al. [14] continue the study of De Bruijn by replacing the 1D grating by a 2D (i.e., peri- odic in two orthogonal directions) grating with box-like grooves. This grating is rigid and placed over, and in contact, with a semi-infinite porous half space absorbing medium whose constitutive properties are described by the semi-empirical Delany and Blazey model [25]. Particularly inter- esting is the result depicted in their fig. 4 concerning the existence, for normal incidence airborne sound, of a total absorption peak (see [61, 48, 62, 18, 86, 74, 78] for the optical analogs thereof) at a frequency slightly higher than 600 Hz which is far above the ∼ 0.25 absorptance obtained at this frequency without the presence of the grating. [43] tackle the problem of reflection, transmission and absorption of airborne sound by a porous layer containing a 1D periodic set of macroscopic circular cylindrical fluid- like inclusions. Their rigorous analysis appeals to the Rayleigh plane wave expansions as well as multipole expansions to account for the presence of the cylinders. They show that high-contrast inclusions in a porous plate induce an increase in the absorption coefficient, mainly associated with a decrease in the hemispherical transmission coefficient for frequencies that are higher than a frequency offset (∼ 20 kHz). They attribute this effect to the excitation of what they call 'modified plate modes', an explanation that is not in contradiction with one of the conclusions of De Bruijn Groby et al. 5 for a different periodic structure. Groby, Lauriks and Vigran [41] examine the acoustic properties of a low resistivity porous layer backed by a rigid plate containing a 1D periodic array of rectangular irregularities, a structure that constitutes a generalization of the one treated by De Bruijn. Numerical results deriving from the rigorous solution, as well as experimental results, show that such a structure can give rise to a total absorption peak at the frequency of the modified mode of the layer. Groby, Dazel et al. [37] study the absorptive properties of a rigidly-backed porous layer contain- ing a periodic set of rigid circular inclusions in response to an airborne acoustic plane body wave. They obtain numerical results from their rigorous theoretical solution which show that this inho- mogeneous layer gives rise to a quasi-total absorption peak below the quarter-wavelength resonance of the corresponding macroscopically-homogeneous porous layer (i.e., the one not containing inclu- sions). This result is explained by the excitation of a complex trapped mode whose characteristics are similar to those of [43]. This article constitutes the primary inspiration of the present investi- gation so that we terminate here our review of the literature on this subject. Naturally, much more research has been carried out since 2011 on the absorption of acoustic waves in macroscopically- homogeneous or inhomogeneous porous media and some of the associated articles are referenced in the first lines of this section. As seen from this literature review, most recent studies of the absorption of airborne sound concern a porous host layer with fluid-filled or rigid inclusions. In the following, the host layer can indifferently be considered as either rigid with porous inclusions or porous with rigid inclusions. We shall show, by means of a very simple theory, derived from, and verified by, a rigorous dynamical theory, that increased absorption (over that of the rigidly-backed porous layer without rigid inclu- sions) can be optimized so as to be total, and, in any case, that the cause of the lowest-frequency enhanced absorption peak is related to something akin to the quarter wavelength pseudo resonance. 2 Response of the macroscopically-inhomogeneous layer to an air- borne plane-wave solicitation 2.1 Description of the configuration Let Oxyz be a cartesian coordinate system with origin at O, A flat-faced layer is located between the planes z = 0 and z = h. The medium underneath (i.e. in z < 0) the layer is air and the medium above (i.e., z > h) the layer is a perfectly-rigid solid. The medium within the layer (henceforth called a grating) is periodic in the x direction and invariant (and infinite) in the y direction. The period of the grating is d and in each period two contiguous blocks, both of of height h, are present, one of width (along x) w filled with a porous medium such as foam, this block being called 'groove' from now on, and the other of width d− w filled with the same rigid solid as in z > h. The grating is thus a macroscopically-inhomogeneous bi-phasic layer whose porous component will be treated as a macroscopically-homogeneous medium. The acoustic wave sources are assumed to be located in the region beneath the layer and to be infinitely-distant from z = 0 so that the solicitation takes the form of a body (plane) pressure wave in the neighborhood of the layer. The incident wavevector ki is assumed to lie in the x − z plane, i.e., ki = (ki z), which fact, together with the independence of the grating geometry and the constitutive properties of the media in presence with respect to y, means that the incident and total pressure wave fields in the various domains do not depend on y. The wavevector z) = (ki x, 0, ki x, ki y, ki 6 ki of the plane wave solicitation is of the form ki = (k[0] sin θi, 0, k[0] cos θi) wherein θi is the angle of incidence (see fig. ??), and k[l] = ω/c[l] ; l = 0, 1, with ω = 2πf the angular frequency, f the frequency, and c[l] the body wave velocity in air (for l = 0) and in the porous material (for l = 1). More often than not, we shall assume θi = 0◦. Consequently, the to-be-considered problem is 2D and can be examined in the sagittal x − z plane. Fig. ?? depicts the problem in the sagittal plane in which: Ω0 is what was previously below Figure 1: Sagittal plane view of the problem. The absorbing structure can be viewed either as a rigid amellar grating whose grooves are entirely filled with a porous absorbing material or as a periodic layer located between two flat planes z = 0 and z = h composed of alternating rigid and porous blocks (rectangles in the figure wherein the dashed line delineates the upper 'boundary' of the layer). The half-space below z = 0 (Ω0) is filled with air (white area in the figure). The half space above z = 0 not including the rectangular grooves is designated by Ω2 and is occupied by a rigid material (dark grey area in the figure). The central layer domain is Ω10 (width w, height h), its left-hand neighbor is the domain Ω1−1 and its right-hand neighbor is the domain Ω11, etc. All the blocks of width w are filled with an absorptive material ('foam'; light gray areas in the figure) whereas all the blocks of width d− w are rigid. The grating, of period d, is solicited by an airborne acoustic plane body wave whose wavevector (lying in the sagittal plane) makes an angle θi with the z axis. z = 0, Ω2 is what was previously above z > h, and Ω1 = ∪n∈ZΩ1n is the layer-like (composite) domain constituted by the periodic assembly of blocks, with Ω1n the n-th block of rectangular cross section (width w and height h). The air medium in z < 0 is assumed to be non-lossy and non-dispersive over the range of frequencies of interest, and its mass density to be real. The (real) longitudinal-wave velocity in ρ[0] , with ρ[0] the mass density and K [0] the isentropic bulk this fluid is the real constant c[0] = q K [0] modulus. As shown in sects. 2.2.1 and 2.2.2, the macroscopically-homogeneous foam medium in each groove behaves like a lossy and dispersive fluid within which the mass density ρ[1](ω) = ρe(ω) and 7 longitudinal wave velocity c[1](ω) =q K [1] ρ[1] = ce(ω) are generally-complex functions of ω. 2.2 Effective-medium behavior of the macroscopically-homogeneous lossy filler material The theory of wave propagation in porous media, considered to be macroscopically-homogeneous, was initially elaborated by Biot [11, 12]. In most of the plastic foams saturated by a light fluid like air, the rigid frame assumption is valid so that an acoustic wave impinging on such a porous sample induces wave propagation only in the fluid phase. Therefore the viscothermal effects taking place in the pore channels are accounted-for by an effective density and an effective bulk modulus of a so-called equivalent fluid [11, 12]. The rigid frame model was extended to macroscopically- inhomogeneous porous media in [50, 4, 17]. We shall apply what has become to be known as the Johnson-Champoux-Allard (JCA) model to account for the absorption of airborne sound in the foam material component (considered here to be macroscopically-homogeneous) of our grating configuration. In the frequency domain (the exp(−iωt temporal factor is implicit) the wave equation in terms of the fluid pressure u inside the equivalent macroscopically-homogeneous or -inhomogeneous fluid is ∇(cid:18) ∇u ρe(ω)(cid:19) + ω2 Ke(ω) u = 0 . (1) Attenuation, viscothermal losses and dispersion are accounted-for in the complex effective density ρe and effective bulk modulus Ke. The effective sound speed and characteristic impedance are ce(ω) =pKe(ω)/ρe(ω) and Ze(ω) = ρe(ω)ce(ω) [4, 27]. The JCA expressions for ρe and Ke are: c = σ′φ/ρf τ∞ , γ the specific heat ratio, P0 the wherein: ωc = σφ/ρf τ∞ is the Biot frequency, ω′ atmospheric pressure, P r the Prandtl number (equal to 0.707 in air at 25◦C), ρf the mass density of the fluid (in the interconnected) pores, φ the open-cell porosity, τ∞ the high-frequency limit of tortuosity, σ the static air flow resistivity, and σ the static thermal resistivity. The correction functions F and G are ′ F (ω) =s1 − iηρf ω(cid:18) 2τ∞ σφΛ(cid:19)2 G(ω) =s1 − iηρf P rω(cid:18) 2τ∞ σ′φΛ′(cid:19)2 , , (4) (5) ′ in which η is the dynamic viscosity of the fluid (here air), Λ Champoux and Allard [17] and Λ the viscous characteristic length of Johnson et al. 'static' thermal resistivity is related to the thermal characteristic length via σ the thermal characteristic length of [50]. The = 8τ∞η/φΛ ′2. ′ 8 ρe(ω) = Ke(ω) = ρf τ∞ φ (cid:16)1 + i ωc ω γP0φ−1 γ − (γ − 1)(cid:16)1 + i ω′ F (ω)(cid:17) , P r (cid:17)−1 , G(ω) c ω (2) (3) Note that the Biot characteristic angular frequency ωc separates the low and high frequency regimes (the viscous flow and inertial flow in the pores), i.e., when ω < ωc , the viscous forces dominate and when ω > ωc, the inertial forces dominate. Note also [50] that ℜce(ω) < limω→∞ ℜce(ω) and [50] cf√τ∞ ω→∞ℜce(ω) = lim , (6) and since τ∞ > 1, the effective phase velocity in the porous medium is less than the phase velocity in air for all finite frequencies. This essential feature of our porous media will be illustrated hereafter in figs. 2 and 3. 2.2.1 The foam parameters in [37] and the associated effective density and wavespeed The absorbing material employed in [37] is called Fireflex 2 which is an open-celled melamine foam. Its parameters are: φ = 0.95 τ∞ = 1.42 Λ = 180 × 10−6 m Λ′ = 360 × 10−6 m σ = 8900 P a sm−2 fc = 781 Hz ρf = 1.213 Kgm−3 P0 = 1.01325 × 105 P a γ = 1.4 η = 1.839 × 10−5 Kgm−3s−1. With the above parameters, the velocity of sound in the pore fluid (here air) is cf =s γP0 ρf = 341.973 ms−1 . (7) The evolution of the JCA effective constitutive parameters with frequency f is given in fig. 2. 9 f ) ρ ( e R , ) ρ ( e R e f ) c ( e R , ) c ( e R e 2.5 2 1.5 1 0.5 0 0 400 300 200 100 0 0 f ) ρ ( m I , ) e ρ ( m I 0.5 1 1.5 f (Hz) 2 2.5 x 104 20 15 10 5 0 0 f ) c ( m I , ) e c ( m I 0.5 1 1.5 f (Hz) 2 2.5 x 104 0 −20 −40 −60 −80 −100 0 0.5 1 1.5 f (Hz) 2 2.5 x 104 0.5 1 1.5 f (Hz) 2 2.5 x 104 Figure 2: Top left-hand panel is relative to ℜρe(f ) (blue curve) and ℜρf (f ) (red curve). Top right-hand panel is relative to ℑρe(f ) (blue curve) and ℑρf (f ) (red curve). Bottom left-hand panel is relative to ℜce(f ) (blue curve) and ℜcf (f ) (red curve). Bottom right-hand panel is relative to ℑce(f ) (blue curve) and ℑcf (f ) (red curve). 10 2.2.2 The foam parameters in [41] and the associated effective density and wavespeed The absorbing material employed in [41] is supposedly another open-celled foam. Its parameters are: φ = 0.96 τ∞ = 1.07 Λ = 273 × 10−6 m Λ′ = 672 × 10−6 m σ = 2843 P a sm−2 fc = 334 Hz ρf = 1.213 Kgm−3 P0 = 1.01325 × 105 P a γ = 1.4 η = 1.839 × 10−5 Kgm−3s−1. The evolution of the JCA effective constitutive parameters with frequency f is given in fig. 3. f ) ρ ( e R , ) ρ ( e R e f ) c ( e R , ) c ( e R e 2.5 2 1.5 1 0.5 0 0 400 300 200 100 0 0 f ) ρ ( m I , ) e ρ ( m I 0.5 1 1.5 f (Hz) 2 2.5 x 104 20 15 10 5 0 0 0 0.5 1 1.5 f (Hz) 2 2.5 x 104 f ) c ( m −50 e I , ) c ( m I 0.5 1 1.5 f (Hz) 2 2.5 x 104 −100 −150 0 0.5 1 1.5 f (Hz) 2 2.5 x 104 Figure 3: Top left-hand panel is relative to ℜρe(f ) (blue curve) and ℜρf (f ) (red curve). Top right-hand panel is relative to ℑρe(f ) (blue curve) and ℑρf (f ) (red curve). Bottom left-hand panel is relative to ℜce(f ) (blue curve) and ℜcf (f ) (red curve). Bottom right-hand panel is relative to ℑce(f ) (blue curve) and ℑcf (f ) (red curve). 2.3 The boundary-value problem for the acoustic response of the macroscopically- inhomogeneous layer The total compressional wavefield in Ωl is the scalar function u[l](x, ω) wherein x = (x, 0, z). The total wavefield in the rigid medium filling Ω2 is nil by definition. The incident wavefield is ui(x, ω) = u[0]+(x, ω) = a[0]+(ω) exp[i(ki xx + ki zz)] , (8) 11 wherein a[0]+(ω) is the spectral amplitude of the solicitation. The plane wave nature of the solicitation and the d-periodicity of B entails the quasi-periodicity of the field, whose expression is the Floquet condition u(x + d, z, ω) = u(x, z, ω) exp(iki xd) ; ∀x ∈ Ω0 + Ω1 . (9) Consequently, as concerns the response in Ω1, it suffices to examine the field in Ω10. The boundary-value problem in the space-frequency domain translates to the following relations (in which the superscripts + and − refer to the upgoing and downgoing waves respectively) satisfied by the total displacement field u[l](x; ω) in Ωl: u[l](x, ω) = u[l]+(x, ω) + u[l]−(x, ω) ; l = 0, 1 , u[l] ,xx(x, ω) + u[l] ,zz(x, ω) + k2u[l](x, ω) = 0 ; x ∈ Ωl ; l = 0, 1 . u[1] ,z (x, h, ω) = 0 ; ∀x ∈ [−w/2, w/2] , 1 ρ[1] 1 ρ[0] u[0] ,z (x, 0, ω) = 0 ; ∀x ∈ [−d/2, w/2] ∪ [w/2, d/2] , 1 ρ[1] u[1] ,x (±w/2, z, ω) = 0 ; ∀z ∈ [0, h] , u[0](x, 0, ω) − u[1](x, 0, ω) = 0 ; ∀x ∈ [−w/2, w/2] , u[0] ,z (x, 0, ω) − u[1] ,z (x, 0, ω) = 0 ; ∀x ∈ [−w/2, w/2] , 1 ρ[1] 1 ρ[0] (10) (11) (12) (13) (14) (15) (16) wherein u,ζ (u,ζζ) denotes the first (second) partial derivative of u with respect to ζ. Eq. (11) is the space-frequency wave equation for pressure, (12)-(14) the expression of vanishing velocity potential at a boundary facing a rigid medium, (15) the expression of continuity of pressure across the junction between the Ω0 and the central block, and (16) the expression of continuity of velocity potential across this junction. Since Ω0 is of half-infinite extent, the pressure field therein must obey the radiation condition u[0]−(x, ω) ∼ outgoing waves ; kxk → ∞ . (17) 2.4 Field representations via domain decomposition and separation of variables (DD-SOV) As the preceding descriptions emphasize, it is natural to decompose R2 into three domains: Ωl ; l = 0, 1, 2, with the understanding that in Ω1 it is only necessary to consider what happens in the central groove subdomain. Applying the Separation-of-Variables (SOV) technique, The Floquet condition, and the radia- tion condition gives rise, in the lower domain, to the field representation: u[0]±(x, ω) =Xn∈Z a[0]± n (ω) exp[i(k[0] xnx ± k[0] znz)] , (18) 12 wherein: k[0] xn = ki x + 2nπ d , k[0] zn =rk2 −(cid:16)k[0] xn(cid:17)2 and, on account of (8), ; ℜk[0] zn ≥ 0 , ℑk[0] zn ≥ 0 ω > 0 , a[0]+ n (ω) = a[0]+(ω) δn0 , (19) (20) (21) with δn0 the Kronecker delta symbol. In the central groove, SOV, together with the rigid boundary conditions (12), (14), lead to u[1]±(x, ω) = ∞ Xm=0 in which a[1] m (ω) cos[k[1] xm(x + w/2)] exp[±k[1] zm(z − h)] , k[1] xm = mπ w , k[1 zm =qk2 −(cid:0)k[1] xm(cid:1)2 ; ℜk[1] zm ≥ 0 , ℑk[1] zm ≥ 0 ω > 0 . 2.5 Exact solutions for the unknown coefficients (22) (23) (24) Eqs. (13) and (16) entail 1 ρ[0] Z d/2 −d/2 ,z (x, 0, ω) exp(−ik[0] u[0] xj x) dx d = 1 ρ[1] Z w/2 −w/2 ,z (x, 0, ω) exp(−ik[0] u[01 xj x) dx d ; ∀j = 0,±1,±2, .... , (25) which, on account of the SOV field representations and the identity Z d/2 −d/2 exphi(cid:16)k[0] xn − k[0] xj(cid:17) xi dx d = δnj , (δnj is the Kronecker delta) yields a[0]− j = a[0]+ j − w 2id ρ[0] ρ[1] 1 k[0] zj ∞ Xm=0 wherein m k[1] a[1] zm sin(cid:16)k[1] zmh(cid:17) E− jm ; ∀j = 0,±1,±2, .... , (26) (27) (28) E± jm =Z w/2 −w/2 exp(cid:16)±ik[0] with sinc(ζ) = sin ζ ζ xm(x + w/2)i dx w/2 = xj x(cid:17) coshk[1] imnsinch(cid:16)±k[0] xj + k[1] xm(cid:17) w/2i + (−1)msinch(cid:16)±k[0] xj − k[1] xm(cid:17) w/2io , and sinc(0)=1. 13 Eq. (15) entails Z w/2 −w/2 u[[0](x, 0, ω) cos [kxl([1](x + w/2)] dx w/2 = Z w/2 −w/2 u[[1](x, 0, ω) cos [kxl([1](x + w/2)] dx w/2 ; ∀l = 0, 1, 2, .... , (29) which, on account of the SOV field representations, and the identity Z w/2 −w/2 coshk[1] xm(x + w/2)i coshk[1] xl (x + w/2)i = 2 ǫl δlm , with ǫl the Neumann symbol (=1 for l = 0 and =2 for l > 0), enables us to find a[1] l =  ǫl zl h(cid:17) 2 cos(cid:16)k[1] ∞ Xn−∞(cid:16)a[0]+   n + a[0]− n (cid:17) E+ nl ; ∀l = 0, 1, 2, .... to determine the two sets of unknowns {a[0]− We thus have at our disposal two coupled expressions (i.e., (27) and (31) which make it possible n }. Note that the number of members of each of these sets is infinite which is the fundamental source of complexity of the problem at hand and the principal reason why one should strive to simplify the theoretical analysis. This will be done in a later section. }, {a[1] n (30) (31) (32) (33) (34) (35) 2.6 Linear system for the set of unknown coefficients Inserting (27) into(31) yields, after the summation interchange, the system of linear equations: wherein ∞ Xm=0 XlmYm = Zl ; ∀l = 0, 1, 2, .... , Ym = a[1] m , Zl = a[0]+ǫlE+ 0l , Xlm = δlm cos(cid:16)k[1] zmh(cid:17) + w 2id ρ[0] ρ[1] ǫl 2 k[1] zm sin(cid:16)k[1] zmh(cid:17) Σlm , Σlm = ∞ Xn=−∞ 1 k[0] zn E+ nlE− nm . Once the Ym = a[1] m are determined they can be inserted into (27) to determine the a[0]− , i.e., j a[0]− j = a[0]+ j − w 2id ρ[0] ρ[1] 1 k[0] zj ∞ Xm=0 Ymk[1] zm sin(cid:16)k[1] zmh(cid:17) E− jm ; ∀j = 0,±1,±2, .... , Until now everything has been rigorous provided the equations in the statement of the boundary- value problem are accepted as the true expression of what is involved in the acoustic wave response of our grating and certain summation interchanges are valid. In order to actually solve for the sets m } (each of whose populations is considered to be infinite at this stage) we must } and {a[1] {a[0]− n resort either to numerics or to approximations. 14 2.7 Numerical issues concerning the system of equations for {a[1] m} We strive to obtain numerically the set {a[1] set is found, it is introduced into (27) to obtain the set {a[0]− m } from the linear system of equations (32). Once this }. When all these coefficients (we mean those whose values depart significantly from zero) are found, they enable the computation of the acoustic wave response (i.e., the displacement field) in all the subdomains of the configuration via (8), (10), (18), (22). n Concerning the resolution of the infinite system of linear equations (32), the procedure is basi- cally to replace it by the finite system of linear equations X (M ) lm Y (M ) m = Zl ; l = 0, 1, 2, ...M , (36) M Xm=0 in which X (M ) to increase M so as to generate the sequence of numerical solutions {Y (0) lm signifies that the series in Xlm is limited to the terms n = 0,±1, ...,±M , and m },....until the values of the first few members of of these sets stabilize and the remaining members become very small (this is the so-called 'reduction method' [81] of resolution of an infinite system of linear equations). m }, {Y (1) m , Y (2) Note that to each Y (M ) m = a[1](M ) m is associated a[0]−(M ) m via (35), i.e., a[0]−(M ) j = a[0]+ j − w 2id ρ[0] ρ[1] 1 k[0] zj M Xm=0 Y (M ) m k[1] zm sin(cid:16)k[1] zmh(cid:17) E− jm ; ∀j = 0,±1,±2, .... ± M , (37) The so-obtained numerical solutions (it being implicit that Y (M ) m = a[1](M ) = 0 ; m > M and a[0]−(M ) ; j > M ), which for all practical purposes can be considered as 'exact' for sufficiently-large j M (of the order of 25 for the range of frequencies and grating parameters considered herein) and which are in agreement with numerical results obtained by a finite element method [36], constitute the reference by which we shall measure the accuracy of the approximate solutions of later sections. 2.8 Conservation of flux for the periodic structure We again refer to fig. 1 wherein we now focus on the integration domain Ω00 ⊂ Ω0 bounded by the dashed lines in the lower part of the figure. Since the boundary-value problem is the same as previously, we again refer to its governing equations given in sect. 2.3. Let ν designate the outward-pointing unit normal to a domain Ω whose closed boundary is ∂Ω, and assume that the total pressure field within Ω is u(x, ω) obeying the Helmholtz equation (∆ + k2)u = 0. Then, applying Green's second identity leads to ℑZ∂Ω u∗ ν · ∇u dγ + ℑ[k2]ZΩ kuk2d = 0 , (38) wherein ∗ designates the complex conjugate operator, dγ the differential element of arc length, and d the differential element of area. Since, by definition, it was assumed that air is lossless, ℑk[0] = 0, so that (39) ℑZ∂Ω00 u[0]∗ν · ∇u[0] dγ = 0 , 15 and, since the field obeys the Floquet condition, − ℑZ d/2 −d/2 u[0]∗(x,−H, ω)u[0] ,z (x,−H, ω) dx + ℑZ d/2 −d/2 u[0]∗(x, 0, ω)u[0] ,z (x, 0, ω) dx = 0 . (40) Due to the boundary and transmission conditions on z = 0, we obtain − ℑZ d/2 −d/2 u[0]∗(x,−H, ω)u[0] ,z (x,−H, ω) dx + ℑ" ρ[0] ρ[1] Z w/2 −w/2 u[1]∗(x, 0, ω)u[1] ,z (x, 0, ω) dx# = 0 . (41) By employment of the plane-wave field representations in Ω00 we find that the first term does not depend on H, and, in fact: − k[0] z0 dka[0]+k2 − ℜXn∈Z from which it follows that in which: k[0] zndka[0]−k2 + ℑ" ρ[0] ρ[1] Z w/2 −w/2 u[1]∗(x, 0, ω)u[1] ,z (x, 0, ω) dx# = 0 , ρ(ω) + α(ω) = 1 . 2 , k[0] zn k[0] z0 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[0]− n a[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ρ(ω) = ℜXn∈Z ka[0]+k2 Z w/2 1 −w/2 u[1]∗(x, 0, ω)u[1] ,z (x, 0, ω) dx k[0] z0 d# , ρ[1]! α(ω) = ℑ" ρ[0] By introducing the SOV field representation relative to u[1] into (45) we finally get (42) (43) (44) (45) (46) α(ω) = ℑ  ρ[0] ρ[1] w d ∞ Xm=0 k[1] zm k[0] z0 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[1] m a[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2 (cid:16)cos(k[1] zmh)(cid:17)∗ sin(k[1] zmh)  . Eq. (44) shows us that ρ(ω) depends only on the 'reflected' field in the half space Ω0 so that it is legitimate to associate it with what in optics [59] is termed the 'spectral reflectance'. We prefer to term it the (normalized) 'reflected flux'. Eq. (46) shows us that α(ω) depends only on the field in the foam-filled grooves (the foam playing the role of absorbing medium), so that it is legitimate to associate it with what in optics is termed the 'spectral absorptance'. We prefer to call it the (normalized) 'absorbed flux'. It follows that (43) is the expression of the conservation of (normalized) flux, the left-hand side of this equation representing the (normalized) output flux and the right hand side the (normalized) input flux. The acoustic design problem, if such be our preoccupation, is to maximize, via the presence of the grating, α(ω) over the widest possible low-frequency bandwidth; (43) shows us that this is possible only by reducing ρ as close as possible to zero at which point α attains its maximal value of unity. More specifically, we shall show that the absorbed flux of the foam-loaded rigid grating structure can be larger, over a certain low-frequency bandwidth) than the absorbed flux of a layer entirely filled with foam and backed by a rigid material when the foams and thicknesses of both structures are identical. 16 ρ(M )(ω) = ℜ α(M )(ω) = ℑ  ρ[0] ρ[1] w d M Xm=0 k[1] zm k[0] 2 , a[0]−(M ) n a[0]+ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) zmh)(cid:17)∗ (cid:16)cos(k[1] M k[0] zn k[0] z0 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xn=−M z0 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[0]+ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[1](M ) m 2 sin(k[1] zmh)  . (47) (48) (49) As we shall employ further on the M < ∞ approximations of the field amplitudes, we must also define the finite-M approximations of the reflected flux (ρ(M )) and absorbed flux (α(M )). Quite naturally, these are: These approximations (if they are satisfactory) should satisfy the conservation relation ρ(M )(ω) + α(M )(ω) = 1 . 2.9 Numerical comparison of the M = 0, 1, 2 approximations of the grating struc- ture response to their exact counterpart for the filler material in [37] We are now in a position to compare the noteworthy features (amplitudes of the fields in the air and in the porous filler, reflected flux, absorbed flux and output flux) of the M = 0, 1, 2 approximations of the grating response to the exact (actually the M = 25 numerical approximation of this) response. In figs. 4-7, the reference solutions (full blue or red curves) are for M = 25 which are compared to the approximate (dashed curves) M = 0 solutions (left-hand panels), M = 1 solutions (middle panels) and M = 2 solutions (right-hand panels). The upper row of panels are for the moduli of the pressure field amplitudes in the air half space, the middle row of panels are for the moduli of the pressure field amplitudes in the layer region and the lower row of panels are for the reflected (red curves), absorbed (blue curves) and output (black curves) fluxes. Note that the output flux is taken as the sum of the computed reflected flux and computed absorbed flux and should be equal to 1 in order for flux to be conserved. 17 1 0.5 s p m a r i a r e d r o − h t 0 s e x u l f r e d r o − h t 0 1 0.5 s p m a r i a r e d r o − t s 1 s e x u l f r e d r o − t s 1 s p m a r e y a l r e d r o − h t o 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 3 2 1 1 s p m a r e y a l r e d r o − t s 1 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 s p m a r e y a l r e d r o − d n 2 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 1 0.5 s p m a r i a r e d r o − d n 2 s e x u l f r e d r o − d n 2 Figure 4: The reference solutions (full blue or red curves) are for M = 25. The dashed (blue or red) curves in the left-hand, middle and right-hand panels are for M = 0, 1, 2 respectively. The upper row of panels depict the moduli of the amplitudes (ka[0]−(M ) k) in the air-filled half space, the middle row of panels are for the moduli of the pressure field amplitudes in the layer region (ka[1](M ) k) and the lower row of panels are for the reflected (red curves, ρ(M )), absorbed (blue curves, α(M )) and output (black curves, ρ(M ) + α(M )) fluxes. Case h = 0.02 m, w = 0.015 m, d = 0.02 m, θi = 0◦. 0 0 18 1 0.5 s p m a r i a r e d r o − h t 0 s p m a r e y a l r e d r o − h t 0 0 3 2 1 1 0.5 0 1 0.5 s e x u l f r e d r o − h t 0 s p m a r i a r e d d r o − h t 0 s e x u l f r e d r o − h 0 t 1 0.5 s p m a r i a r e d r o − t s 1 1 0.5 s p m a r i a r e d r o − t s 1 s e x u l f r e d r o − t s 1 1 0.5 s p m a r i a r e d r o − d n 2 s e x u l f r e d r o − d n 2 s e x u l f r e d r o − d n 2 1 0.5 s p m a r i a r e d r o − d n 2 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 s e x u l f r e d r o − t s 1 0.5 1.5 1 f (Hz) 2 x 104 s p m a r e y a l r e d r o − t s 1 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 s p m a r e y a l r e d r o − d n 2 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 Figure 5: Same as fig. 4 except that θi = 20◦. s p m a r e y a l r e d r o − h 0 t 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 s p m a r e y a l r e d r o − t s 1 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 s p m a r e y a l r e d r o − d n 2 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 4 2 0 1 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 Figure 6: Same as fig. 4 except that θi = 40◦. 19 1 0.5 s p m a r i a r e d r o − h 0 t s p m a r e y a l r e d r o − h 0 t 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 2 1 0 1 s p m a r e y a l r e d r o − t s 1 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 2 1 0 1 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 s e x u l f r e d r o − h 0 t 1 0.5 s p m a r i a r e d r o − t s 1 s e x u l f r e d r o − t s 1 1 0.5 s p m a r i a r e d r o − d n 2 s p m a r e y a l r e d r o − d n 2 0.5 1 1.5 2 x 104 0.5 1 1.5 2 x 104 0 2 1 0 1 0.5 0 0.5 1.5 1 f (Hz) 2 x 104 s e x u l f r e d r o − d n 2 Figure 7: Same as fig. 4 except that θi = 60◦. 20 The figures show that the M = 0 solution is a good approximation of the response functions at low frequencies provided the absolute value of the incident angle does not exceed 40◦. Moreover, all the (i.e., M = 0, 1, 2) approximate solutions, as well as the M = 25 reference solutions satisfy (numerically) the conservation of flux relation. Consequently, it seems reasonable to adopt the M = 0 solution as a suitable representation of the low-frequency, near normal-incidence, response of the grating. This is what is done in the sequel. 3 Response of a macroscopically-homogeneous lossy layer to air- borne sound 3.1 Why study this problem? The reason why we should be interested in this problem is because the configuration of a rigidly- backed layer entirely filled with foam is ubiquitous in the noise-reduction applications and thus serves as a reference to test the effectiveness of replacing the layer by a grating of alternating rigid and foam materials. Also, further on, we shall shown that the low-frequency, small incident angle, response of the inhomogeneous layer reduces to that of avhomogeneous layer. 3.2 Description of the configuration The first task is to establish the quantities that provide a measure of the noise-reduction effectiveness of the introduction of the foam layer between the air and rigid half spaces. We shall also show that these quantities are related by a conservation law. Figure 8: Same as fig. 1 except that the former macroscopically periodically-heterogeneous layer of thickness h (i.e., the grating whose grooves are filled with foam of density ρ[1] and wavespeed c[1]) is now a macroscopically-homogenous layer of thickness H filled with another foam of density R[1] and wavespeed C [1]. Moverover, the lower integration domain Ω00 is as previously. Fig. 8 depicts the problem in the sagittal plane in which: Ω0 is the half-space domain occupied 21 by a light fluid such as air, Ω1 the domain of the layer occupied by another macroscopically- homogeneous fluid (i.e., the approximation resulting from the JCA model of foam) which is lossy and dispersive, and Ω2 the half-space above the layer occupied by an acoustically-rigid material. The densities of the lower medium and layer are R[0] and R[1] respectively, with R[0] positive real and R[1] complex (real and imaginary parts positive). The compressional-wave velocities in the lower medium and layer are C [0] and C [1] respectively, with C [0] positive real and C [1] complex (real part positive, imaginary part negative). The light fluid-borne plane-wave solicitation is as previously and given by U i(x, ω) = U [0]+(x, ω) = A[0]+(ω) exp[i(K i xx + K i zz)] , (50) wherein A[0]+(ω) is the spectral amplitude of the solicitation, K i and K [l] = ω/C [l]. The total wavefield U (x, ω) in Ωl is designated by U [l](x, ω). x = K [0] sin θi, K i z = K [0] cos θi, 3.3 The boundary-value problem of the response of the rigidly-backed layer structure to a plane wave The boundary-value problem in the space-frequency domain translates to the following relations: U [l](x, ω) = U [l]+(x, ω) + U [l]−(x, ω) ; l = 0, 1 , U [l] ,xx(x, ω) + U [l] ,zz(x, ω) + (K [l])2U [l](x, ω) = 0 ; x ∈ Ωl ; l = 0, 1 . 1 R[1] U [1] ,z (x, H, ω) = 0 ; ∀x ∈ R , 1 R[0] U [0](x, 0, ω) − U [1](x, 0, ω) = 0 ; ∀x ∈ R, (x, 0, ω) − U [0]−(x, ω) ∼ outgoing waves ; x → ∞ . U [1] ,z (x, 0, ω) = 0 ; ∀x ∈ R . 1 R[1] Due to the translational symmetry along x, the field obeys a sort of Floquet condition U [l](x + d, z, ω) = U [l](x, z, ω) exp(iK i xd) ; l = 0, 1 , this relation being true for all d. 3.4 Conservation of flux (51) (52) (53) (54) (55) (56) (57) We proceed as in sect. 2.8 to obtain, by means of the boundary condition, the transmission conditions, and the 'Floquet' condition − ℑZ d/2 −d/2 U [0]∗(x,−H, ω)U [0],z(x,−H, ω) dx + ℑ" R[0] R[1] Z d/2 −d/2 U [1]∗(x, 0, ω)U [1] ,z (x, 0, ω) dx# = 0 . (58) To make this relation more explicit, we apply the DD-SOV technique, and the radiation condition to obtain, in the lower domain, the field representation: U [0]±(x, ω) = A[0]±(ω) exp[i(K [0] x x ± K [0] z z)] , (59) 22 wherein: K [0] x = K i x , In the layer, the SOV, together with the boundary condition (53), lead to K [0] ; ℜK [0] z =r(cid:0)K [0](cid:1)2 −(cid:16)K [0] x (cid:17)2 U [1](x, ω) = A[1](ω) exphiK [1] x xi coshK [1] z (z − H)i , z ≥ 0 , ℑK [0] z ≥ 0 ω > 0 . K [1] z =q(cid:0)K [1](cid:1)2 −(cid:0)K [1] z ≥ 0 , ℑK [1] z ≥ 0 ω > 0 . Employing these relations in (58) leads, in the same manner as previously, to the conservation of flux relation (60) (61) (62) (63) (64) (65) (66) 2 E(ω) . (67) in which in which K [1] x = K [0] x = K i x , x (cid:1)2 ; ℜK [1] R(ω) + A(ω) = 1 , , 2 A[0]− A[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) R(ω) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) z !(cid:16)cos(K [1] z H)(cid:17)∗ 2 R[0]K [1] z R[1]K [0] A(ω) = ℑ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)  A[1] A[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) sin(K [1] z H)  =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) A[1] A[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) These expressions show, as one would expect, that the reflected flux R and absorbed flux A do not depend on d (the width of the integration domain) nor on H (the height of the integration domain). Furthermore, although R is a straightforward function of the reflected field amplitude via kA[0]−k2, A is not a straightforward function of the amplitude of the field in the layer via kA[1]k2 because this latter function is multiplied by E. To go further into this matter, we must solve for these two amplitudes. 3.5 Exact solution for the unknown coefficients The introduction of the field representations into (54)-(55) yields the two equations A[0]+ + A[0]− = A[1] cos(cid:16)−K [1] z H(cid:17) , i R[0] K [0] z (cid:16)A[0]+ − A[0]−(cid:17) = − 1 R[1] K [1] z A[1] sin(cid:16)−K [1] z H(cid:17) , the exact solution of which is: A[1] = A[0]+  2 cos(cid:16)K [1] z H(cid:17) + R[0]K [1] iR[1]K [0] z z   sin(cid:16)K [1] z H(cid:17) = A[0]+"N [1] D # . 23 (68) (69) (70) A[0]− = A[0]+  z H(cid:17) , S = sin(cid:16)K [1] wherein C = cos(cid:16)K [1] cos(cid:16)K [1] cos(cid:16)K [1] z H(cid:17) − R[0]K [1] z H(cid:17) + R[0]K [1] iR[1]K [0] z sin(cid:16)K [1] z H(cid:17) sin(cid:16)K [1] z H(cid:17)   D # , = A[0]+"N [0] (71) iR[1]K [0] z z z z H(cid:17) , G = R[0]K [1] z R[1]K [0] z , D = C−iGS , N [1] = 2 , N [0] = C +iGS . (72) It follows that: R(ω) = kCk2 + kGSk2 − 2ℑ(GSC ∗) kCk2 + kGSk2 + 2ℑ(GSC ∗) , A(ω) = 4ℑ(GSC ∗) kCk2 + kGSk2 + 2ℑ(GSC ∗) , (73) which shows that (65) is satisfied, i.e., flux is conserved, as is necessary. Eq. (73) also shows that there can be no absorption unless G and/or SC ∗ are complex which means that non-vanishing absorption requires that R[1] and/or C [1] be complex. Choosing the values of these complex parameters to obtain maximal absorption is a multiparameter optimization problem that will be considered further on. Also, we shall then give numerical examples of the way the absorbed flux varies with the thickness and constitutive parameters of the foam layer as well as with the frequency. 3.6 Origin and meaning of the so-called quarter-wavelength resonances The so-called quarter-wavelength resonances (QWR) are often mentioned (without giving their theoretical basis) in connection with explanations of the oscillatory behavior in general, and the maxima in particular, of the response of a rigidly-backed (usually homogeneous) layer submitted to airborne sound [37, 49]. At present, we address three questions relative to the QWR: (i) where do they come from?, (ii) are they really resonances?, and (iii) what does the theory underlying the QWR actually tell us about the response of the layer? In elastic wave problems, particularly those related to studies of ground shaking during earth- quakes in a configuration in which the ground overlies a relatively-soft soil layer, whose boundaries are planar and mutually-parallel, and which is underlain by a very hard rock basement, the notion of soil layer (natural) frequency was developed as early as 1930 [84] and employed successfully by scores of geophysicists to furnish simple explanations of many features (frequencies of occurrence and heights of the peaks of the transfer functions) of recorded seismograms on the earth's surface. The notion of soil frequency was even employed (and renamed the 1D resonance frequency) to furnish a rough explanation of the seismic response of surface layers with curved upper or lower boundaries (hills or basins), e.g. [3, 9, 97]. It is probable that similar notions appeared even earlier than 1930 in connection with optical problems involving a dielectric layer over a near-perfect con- ductor and perhaps in similar water wave contexts. Here, we adopt the acoustic wave formalism for a homogeneous layer, with flat-plane parallel boundaries, backed above by an infinitely-rigid half-space and below by a light fluid such as air in which propagates a plane body acoustic wave. The Neumann boundary condition on the upper (z = H) face of the layer is the same as the one which prevails on the ground overlying the geophysical flat-faced layer so that we shall proceed as in the geophysical problem by focusing our attention (there on the ground) on the field on the upper face (here in contact with the rigid backing and termed the 'top') of the layer. 24 We showed previously that the frequency domain pressure in the layer is: U [1](x, z, f ) = A[1] cos[K [1] z (z − H)] exp(iKxx) , wherein Kx = K [0] x = K [1] x and with G = R[0]K [1] z /R[1]K [0] z . A[1] = A[0]+(cid:18) 2 C − iGS(cid:19) , (74) (75) Recall that it was (and continues to be) assumed that both R[0] and C [0] are real and non- dispersive. The case (e.g., such as in a macroscopically-homogeneous porous layer) in which either or both R[1] and C [1] are complex and dispersive demands a very involved analysis that is out of the scope of the present discussion, so that we shall make the drastically-simplified assumption that both R[1] and C [1] are real and non-dispersive. We define the modulus squared of the top transfer function as so that on account of our drastic assumptions 2 T (f ) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) u[1](x, H, f ) A[0]+(f ) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) A[1](f ) A[0]+(f )(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) T (f ) = 4 C 2 + G2S2 . , (76) (77) By equating to zero the partial derivative of this expression with respect to f and, on account of the fact that G is now independent of f , we obtain (1 − G2)CS = 0 , (78) which means that T is extremal when either C = 0 or S = 0. To see which of these roots correspond to maximal T , we use (79) T (f )(cid:12)(cid:12)(cid:12)S=0 = 4 , T (f )(cid:12)(cid:12)(cid:12)C=0 = 4 G2 , so that the identification of the frequencies at which T is maximal relies on whether G > 1 or G < 1. It is straightforward to show that G = R[0]C [0]η[0] R[1]C [1]η[1] , (80) wherein η[l] =r1 −(cid:16) C[0] sin θi C[l] (cid:17)2 . For small θi and reasonable velocity contrasts, η[l] ≈ 1, so that R[0]C [0] R[1]C [1] . G ≈ (81) If we refer to the porous material of fig. 2, neglect the dispersion and imaginary parts of the effective wavespeed and density, then a plausible choice of the real parts is: R[l] = 2 Kg m−3 and C [l] = 250 ms−1, considering, as previously, that R[0] = 1.2 Kg m−3 and C [0] = 342 ms−1, so that G ≈ 0.821 which fact shows that C = 0 corresponds to a maximum of T and S = 0 to a minimum 25 of T . Moreover, C = 0 implies that K [1] This means that the response function T is maximal for layer thicknesses z H = (2m + 1)π/2, or K [1]H ≈ (2m + 1)π/2 for l = 0, 1, .... Hm ≈ C [1] 4f (2m + 1) = Λ[1] 4 (2m + 1) ; m = 0, 1, .... . or for frequencies fm ≈ C [1] 4H (2m + 1) ; m = 0, 1, .... . (82) (83) Thus, from the acoustic point of view, the top response function T is maximal, but not generally infinite, for layer thickness that are odd multiples of the quarter wavelength Λ[1], a finding we could qualify as 'quarter wavelength maxima' (QWM) or 'quarter wavelength pseudo resonance' (QWPR), rather than QWR. However, it should be noted that these maxima tend to infinity as G tends to zero, which means that QWM tends to QWR as G → 0. On the other hand, from the elastic wave point of view, the top response function T is maximal, for frequencies fm that are odd multiples of C[1] 4H is named the 'soil (layer) natural frequency' or simply the 'soil frequency'. Once again, as G → 0, the maximum of response tends to infinity which fact is reminiscent of a resonance so that f0 is also named the 'soil layer resonant frequency' or '1D resonance frequency' by geophysicists. 4H and f0 = C[1] The adjective 'natural' in connection with a frequency suggests that some sort of mode is involved as concerns the field within the layer. Recall that the boundary condition on the top z = H was U,z(x, H, f ) = 0. Eq. (75) shows that so that U [1](x, 0, f ) = A[1]C exp(iKxx) , U [1](x, 0, fm) = 0 , (84) (85) which means that at the frequencies fm, the layer behaves like a cavity whose upper face is the locus of a Neumann (rigid) boundary condition and whose lower face is the locus of a Dirichlet (pressure release) boundary condition. Such a situation suggests that the field in the layer is a mode of the cavity at the frequencies fm and therefore it is legitimate to term the latter the 'natural (or eigen) frequencies' of the cavity modes. Another consequence of (83) is that the first maximum of top response shifts to lower frequencies as H increases or C [1] decreases, and a final consequence of this formula is that the fm do not depend on G. In fact G only controls the height of the maxima of the response function. All these features turn out to hold remarkably well for rigidly-backed layers (see fig. 9 in sect. 5) even when the drastic assumption of real, non-dispersive effective wavespeed and density is relaxed as is necessary when considering a foam material with characteristics such as is depicted in fig. 2. Moreover, since the top T is related to kA[1]k2, as is the absorbed flux in the layer, many of the above-mentioned features of T apply to, and help to understand, the layer absorption too. As we shall see further on in sect. 5, these features hold rather well at low frequencies even when the layer is replaced by a grating of alternating (in the x direction) blocks of rigid material and foam material, this being the real reason why the QWR (more properly the QWM) plays a useful role in the understanding of how such a structure responds (notably as concerns absorption) to an airborne acoustic plane wave. 26 whence However, so that Moreover Y (0) 0 = a[1](0) 0 00 = 2sinc(k[0] E± a[1](0) 0 = a[0]+  a[0]−(0) 0 which reduces to = a[0]+  cos(cid:16)k[1] x0w/2) , Σ(0) 00 = E+ 00 ρ[0] ρ[1] k[1] z0 sin(cid:16)k[1] z0 h(cid:17) Σ(0) 00   . z0 h(cid:17) + w 4id E+ 00E− 00 = 1 k[0] z0 [sinc(k[0] x0w/2)]2 , 4 k[0] z0 cos(cid:0)k[1] ρ[0] ρ[1] id z0 h(cid:1) + w 0 − a[1](0) 0 = a[0]+ 2sinc(cid:0)k[0] x0w/2(cid:1) z0(cid:16)sinc(cid:0)k[0] k[1] z0 k[0] w 2id ρ[0] ρ[1] k[1] z0 k[0] z0 .   sin(cid:0)k[1] z0 h(cid:1) 00 , x0w/2(cid:1)(cid:17)2 sin(cid:0)k[1] z0 h(cid:1)E+ x0w/2(cid:1)(cid:17)2 x0w/2(cid:1)(cid:17)2 (86) (87) (88) (89) (90) (91) 4 Analytical aspects of the M = 0 approximation of the response of the grating structure A consequence of (36) is Y (0) 0 = Z0 X (0) 00 , a[0]−(0) 0 = a[0]+  id cos(cid:0)k[1] cos(cid:0)k[1] z0 h(cid:1) − w z0 h(cid:1) + w id ρ[0] ρ[1] ρ[0] ρ[1] k[1] z0 k[0] k[1] z0 k[0] z0(cid:16)sinc(cid:0)k[0] z0(cid:16)sinc(cid:0)k[0] sin(cid:16)k[1] z0 h(cid:17) sin(cid:0)k[1] z0 h(cid:1) .   These results call for three comments. The first is that there is a striking resemblance of these amplitudes with the layer configuration amplitudes in sect. 3.5. This similarity will be exploited further on. The second comment is just to recall that we can compute the M = 0 approximation of reflected and absorbed fluxes via (47)-(48) associated with these amplitudes. The third comment has to do with the possibility (or impossibility) of surface-wave resonances (SWR, not to be con- founded with the QWR) showing up in the response functions. SWR, typically those associated with the excitation of homogeneous layer modes (typically of the Love variety [31]), occur (i.e., at a set of frequencies) for which the denominator in the response amplitudes are equal (in the absence of losses in the media in presence) or very nearly equal (in the presence of slightly-lossy media) to zero, therefore leading to infinite or very large response. For this to occur, while assuming that the medium in the grating groove is lossless (or very slightly lossy), would require that the second term (i.e., the one including sin) in the denominators of (89) and (91) be real and negative in relation to the first (i.e., the cos) term, but this is impossible because the factor w/id multiplying sin is imaginary. It follows that the M = 0 approximation of the uneven boundary response cannot account for SWR behavior, which fact was already observed in the numerical results presented in sect. 2.9. We shall return to this issue in sect. 4.1. 27 4.1 From the M = 0 approximate solution to an effective medium representation of the response of the grating configuration Let Then S = sin(cid:0)k[1] z0 h(cid:1) sinc(cid:0)k[0] x0w/2(cid:1) = a[0]+  = a[0]+  C − w C + w id C + w id id , C = . 2 cos(cid:0)k[1] z0 h(cid:1) sinc(cid:0)k[0] x0w/2(cid:1)   x0w/2(cid:1)(cid:17)2 S(cid:17) x0w/2(cid:1)(cid:17)2 S(cid:17)   x0w/2(cid:1)(cid:17)2 S(cid:17) z0(cid:16)sinc(cid:0)k[0] z0(cid:16)sinc(cid:0)k[0] z0(cid:16)sinc(cid:0)k[0] k[1] z0 k[0] k[1] z0 k[0] (92) (93) (94) . . a[1](0) 0 a[0]−(0) 0 ρ[0] ρ[1] k[1] z0 k[0] ρ[0] ρ[1] ρ[0] ρ[1] The comparison of (93)-(94) with (70)-(71) shows that the zeroth-order approximate solution of the grating problem is structurally-similar to the exact solution of the homogeneous layer problem. This suggests that there exists a relation of the homogeneous layer parameters H,C [1], R[1] to the grating parameters h, c[1], ρ[1]. To establish this relation is equivalent to solving an inverse problem, and since it is well-known that the solution of inverse problems are generally not unique, we can expect to be able to find many possible relations of H,C [1], R[1] to h, c[1], ρ[1]. 4.1.1 General considerations on inverse problems such as ours There exists a host of manners of formulating the inverse parameter retrieval problem. Basically, one tries to minimize some function of the discrepancy between the 'data' involving so-called 'true parameters' and a hypothetical model that one thinks is able to generate the data, this model involving the so-called 'trial parameters'. Note that this minimization problem has to be solved for many frequencies (i.e., at least those in the bandwidth of the acoustic solicitation) since the data is frequency-dependent even if the true parameters do not depend on frequency. form the set P = {H, C [1], R[1]}. The response function data is a[1](0) In the present case, the true parameters form the set p = {h, c[1], ρ[1]} and the trial parameters or both of these. The data is more traditionally a measurable quantity such as the pressure at one or several points within or outside of the grating, these pressures being, of course directly related to a[1](0) 0 , but it could be a[0]−(0) and a[0]−(0) 0 0 0 . Let a response function associated with the true parameters be f(p, ω) and its counterpart associated with the trial parameters be F(P, ω). Since we reserve the possibility of employing several response functions to deal with our inverse problems, we attach a subscript j to f and F, with j ranging from 1 to Nd, the latter quantity being the number of response functions we deal with. Often, Nd is taken equal to the number of unknown parameters in P . If we are very sure of the veracity of our trial model then we can attempt to solve for P via the system of equations Fj(P(ω), ω) − fj(p(ω), ω) = 0 j = 1, 2, ..., Nd . (95) Note that we have explicitly introduced the frequency-dependence of the two parameter sets, since this dependence reflects the reality of physical problems such as the one we are faced with. This, 28 and the previously-mentioned remark, mean that the set of (generally non-linear) equations (95) must be solved for all the frequencies within the bandwidth of the source. If we are less sure of the veracity of the trial model, or unable to solve the set of nonlinear equations explicitly, then a common strategy is to to search for P in the following optimization problem manner P(ω) = arg min P∈S Nd Xj=1 kFj(P(ω), ω) − fj(p(ω), ω)k2 , (96) which means that, for each ω, P is varied over the multidimensional (dimension L equal to the number of to-be-retrieved parameters) search domain, the cost function PNd j=1 kFj(P(ω), ω) − fj(p(ω), ω)k2 is computed for each trial P, and the so-called optimal P, denoted by P is cho- sen to be the one giving rise to the minimal cost. The so-called effective or optimal model is then Fj(ω) = Fj( P(ω), ω). This procedure, like the one associated with solving a system of nonlinear equations, usually does not give rise to a unique solution. Moreover, the minimization procedure may be unstable, which means that the slightest modification of F or f can lead to entirely-different P. It can be advantageous for the optimization to be carried out for a number of unknowns in P inferior to L provided, of course, that one has a decent idea of the values to be assigned to the other parameters in P. This advantage stems from the facts: i) as concerns the procedure (96), that the non-uniqueness and instability decrease with the dimensionality of the search space, and ii) as concerns (95), that it might be possible to solve explicitly the system of equations if there are less unknowns, therefore affording physical insights that are otherwise difficult to obtain from numerical solutions. 4.2 The methods of solution of the inverse problem adopted herein Hereafter, we treat the inverse problem via (95) in which Nd = 2, f1(ω) = a[1](0) 0 f2(ω) = a[0]−(0) 0 (ω) = a[0]+  c+ w id ρ[0] ρ[1] k k 0 (ω) = A[0]+  c− w id ρ[0] ρ[1] F1(ω) = A[1] (ω) = a[0]+  c+ w id ρ[0] ρ[1] (ω) = A[0]+  ,   s(cid:17) , s(cid:17) s(cid:17)   K K [1] z0 [0] [1] z [0] z 2 C+ R[0] iR[1] 2sinc(kx0w/2) [1] z0 [0] x0w/2(cid:1)(cid:17)2 z0(cid:16)sinc(cid:0)k[0] S  , x0w/2(cid:1)(cid:17)2 z0(cid:16)sinc(cid:0)k[0] x0w/2(cid:1)(cid:17)2 z0(cid:16)sinc(cid:0)k[0]   C+ R[0] iR[1] C− R[0] R[1] [1] z [0] z0 [1] z [0] z [1] z0 [0] k k k k K K K K S S , F2(ω) = A[0]− 0 wherein C = cos(K [1] z H) , C = sin(K [1] z H) , c = cos(k[1] z0 h) , s = sin(k[1] z0 h) . (97) (98) 29 and we solve for only two of the trial parameters H,C [1], R[1] of the homogeneous layer model at a time (this is similar to the NRW technique of parameter retrieval [77] popular in metamaterial research). Thus, we search these two unknowns from A[1](ω) − a[1](0) A[0]−(ω) − a[0]−(0) 0 0 (ω) = 0 , (ω) = 0 . (99) It turns out to be difficult to solve this system for two of the parameters H,C [1], R[1] in explicit manner without making the assumption (justified, in that we are essentially interested in the low frequency context) that the frequency and incident angle are such that which implies sinc(kx0w/2) ≈ 1 (but not necessarily normal incidence), and therefore k[0] x0w/2 << 1 , (100) (101) f1(ω) = a[1](0) 0 f2(ω) = a[0]−(0) 0 (ω) ≈ a[1] (ω) ≈ a[0]− 0 0 (ω) = a[0]+  (ω) = a[0]+  2 ρ[0] ρ[1] k k c+ w id c− w id c+ w id ρ[0] ρ[1] ρ[0] ρ[1] s [1] z0 [0] z0 k [1] z0 [0] z0 [1] z0 [0] z0 k k k ,     s s , It seems natural to: i) excite the two (true and trial) configurations with the same wave (i.e., the angles of incidence are the same, as assumed from the outset, and the spectral amplitudes of the incident plane waves are the same (i.e., A[0]+ = a[0]+), and ii) assume that the media in which this plane wave propagates to be the same (i.e., air). Consequently, R[0] = ρ[0] and C [0] = c[0] (which implies K [0] z = k[0] z0 ), so that Therefore (99) is of the form F1(ω)) = a[0]+  F2(ω) = a[0]+    −    −  [1] z [0] z0 [1] z [0] z0 [1] z [0] z0 S S S K K K k k k C− ρ[0] iR[1] C+ ρ[0] iR[1] 2 C+ ρ[0] iR[1]     2 C+ ρ[0] iR[1] C− ρ[0] R[1] C+ ρ[0] iR[1] K k [1] z [0] z0 K k [1] z [0] z0 [1] z [0] z0 K k S S S 2 ρ[0] ρ[1] k k [1] z0 [0] z0 s c+ w id c− w id ρ[0] ρ[1] c+ w id ρ[0] ρ[1] k k k k [1] z0 [0] z0 [1] z0 [0] z0 s 30         , . = 0, . = 0 (102) (103) 2 C+ ρ[0] iR[1] K k [1] z [0] z0   implies   S C− ρ[0] iR[1] C+ ρ[0] iR[1] K k K k [1] z [0] z0 [1] z [0] z0 S S   and   2 ρ[0] ρ[1] k k [1] z0 [0] z0 s c+ w id   implies so that the two equations in (103) are equivalent and it suffices to choose one We showed previously that     [1] z0 [0] z0 [1] z0 [0] z0 of them, i.e.,   c− w id c+ w id ρ[0] ρ[1] ρ[0] ρ[1] s s k k k k   1 C + ρ[0] iR[1] K [1] z k[0] z0 S   −  1 c + w id ρ[0] ρ[1] k[1] z0 k[0] z0 s   = 0 , (104) which should enable the retrieval of only one of the parameters H,C [1], R[1], assuming, of course, 'plausible' values for the other two. The translation of this 'equivalence' is a series of relations between the parameters of the layer and their counterparts of the grating. We enumerate hereafter several possible solutions (for a single parameter) of (104). 4.2.1 First solution Eq. (104) is satisfied provided: 1a) C [1] = c[1] ⇒ K [1] z = k[1] z0 , 1b) H = h ⇒ C = c , S = s , 1c) R[1] = ρ[1] . w d (105) Note that 1a)-1b) are the 'plausible' guesses and 1c) their consequence, this retrieval being explicit, of simple nature, and unique. Also, note that the optimal parameters C [1], R[1] of the layer problem depend on the frequency in exactly the same manner as those of the grating problem. Finally, note that the model of amplitude response in the effective layer is Associated with this we also have the effective model of response in the air-filled lower half space A[1](ω) = a[0]+  2 C + ρ[0] i R[1] K [1] z k[0] z0 S   as well as the effective reflected and absorbed fluxes C − ρ[0] i R[1] C + ρ[0] i R[1] A[0]−(ω) = a[0]+  A(ω) = ℑ  = ρ(ω) , 2 K [1] z k[0] z0 K [1] z k[0] z0 S S   2 w a[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[1] 0 d (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 31 R(ω) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) a[0]− 0 a[0]+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) = a[0]+  = a[0]+  s s w d 2 c + ρ[0] i ρ[1] k[1] z0 k[0] z0 k[1] z0 k[0] z0     z0!(cid:16)cos(k[1] z0 h(cid:17)∗ k[1] z0 k[0] z0 c − ρ[0] i ρ[1] c + ρ[0] i ρ[1] w d w d s ρ[0]k[1] z0 ρ[1]k[0] = a[1] 0 (ω) . (106) = a[0]− 0 (ω) , (107) sin(k[1] z0 h)  = α(ω) , (108) which satisfy the conservation of flux relation R(ω) + A(ω) = ρ(ω) + α(ω) = 1 , 4.2.2 Second solution Eq. (104) is satisfied provided: 2a) H = h , 2b) R[1] = ρ[1] w d 2c) K [1] z solution (6= k[1] z0 h(cid:17) + w ρ[0]k[1] z0 ρ[1]k[0] z0 id z0 ) of (cid:20)cos(cid:16)K [1] sin(cid:16)k[1] sin(cid:16)K [1] z0 h(cid:17)(cid:21) = 0 and C [1] = ω/q(K [1] z h(cid:17) + w ρ[0]K [1] z ρ[1]K [0] z id z h(cid:17)(cid:21)− (cid:20)cos(cid:16)k[1] z )2 + (k[0] x0)2 . (109) (110) Note that 2a)-2b) are the 'plausible' guesses and 2c) their consequence. Note also that now the retrieval of C [1] requires solving a nonlinear equation for each frequency and the solution of this equation is not unique. Finally, note that we did not put tildes on the retrieved parameters because we did not make this choice of optimality. 4.2.3 Third solution Eq. (104) is satisfied provided: 3a) C [1] = c[1], 3b) R[1] = ρ[1] w d 3c) H solution (6= h) of (cid:20)cos(cid:16)k[1] (cid:20)cos(cid:16)k[1] sin(cid:16)k[1] z0 h(cid:17) + w ρ[0]k[1] z0 ρ[1]k[0] z0 z0 H(cid:17) + w z0 h(cid:17)(cid:21) = 0 . id id ρ[0]k[1] z0 ρ[1]k[0] z0 sin(cid:16)k[1] z0 H(cid:17)(cid:21)− (111) Note that 3a)-3b) are the 'plausible' guesses and 3c) their consequence. Note also that now the retrieval of H requires solving a nonlinear equation for each frequency and the solution of this equation is not unique. Finally, note that we did not put tildes on the retrieved parameters because we did not make this choice of optimality. 4.2.4 Further comments on consequences of the M = 0 approximation of grating response If, for a reason to be evoked further on, one chooses one of the three solutions as a means of identifying some or all of the effective medium parameters (i.e., those denoted by upper-case letters), then he should be aware of the fact that these choices all derive from a M = 0 approximation of the uneven boundary response, which, as shown previously in sect. 4.2, is only valid at very low frequencies and/or near-normal incidence, cannot account, by any means, for surface wave resonant behavior of the uneven boundary. To deal with SWR behavior, one must solve an inverse problem by matching 'true data' deriving from the M = 1 approximation to the layer model thereof, as is done in [98]. 32 From here on, we shall make the first choice of effective model, i.e., the rigidly-backed macroscopically-homogeneous layer whose constitutive and geometric parameters are C [1] = c[1], H=h, and R[1] = ρ[1] . w d 5 Numerical results for the foam filler of [37]: comparison of the w/d < 1 grating quasi-exact and effective-layer response to the foam-filled layer response (ω), At this point, it seems useful to accomplish two tasks: 1) show to what extent all the functions (such as the absorbed flux α(ω)) deriving from a[1] (itself being a result of assuming (102) in a[1](0) a[1](0) 0 2) show to what extent the grating response, notably the approximation of the absorbed flux a[1] 0 (ω)), differs from the rigidly-backed macroscopically-homogeneous layer faced by air response with constitutive and geometric parameters C [1] = c[1], R[1] = ρ[1] and H=h, notably the absorbed flux by this layer resulting from ¯A[1](ω). 0 (ω) (ω)) compare with their counterparts deriving from 0 ¯A(ω) = 4ℑ(gsc∗) kck2 + kgsk2 + 2ℑ(gsc∗) , (112) wherein g = ρ[0] ρ[1] k[1] z0 k[0] z0 . 0 (ω), the red curves from a[1] In figs. 9-12 the blue curves result from a[1](0) 0 (ω) and the black curves from ¯A[1](ω). In the upper panels, the full curves denote real parts and the dashed curves imaginary parts of the amplitude functions. The left-hand panels are relative to functions in the lower air half space whereas the right-hand panels are relative to functions in the layer or grating. More specifically: (a) the upper left-hand panel is relative to the reflected amplitudes and the upper right-hand to the amplitudes in the layer/grating, (b) the lower left-hand panel is relative to the reflected fluxes and the lower right-hand panel to the absorbed fluxes (full curves) and output fluxes (dashed curves). 33 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t u p t u o d n a d e b r o s b a 4 2 0 −2 −4 0 1 0.8 0.6 0.4 0.2 2000 0 0 2000 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 0 Figure 9: The blue curves refer to the M = 0 approximate responses for the grating=periodically- inhomogeneous layer. The red curves refer to the effective-layer responses of the grating. The black curves refer to the responses of the macroscopically-homogeneous foam-filled layer. Upper left- hand panel: ℜa[0]−(0) (ω) (blue full), ℑa[0]−(0) (ω) (red dashed referring to the approximate results for the grating), ℜ ¯A[0]−(ω) (black full), ℑ ¯A[0]−(ω) (black dashed). Upper right-hand panel: ℜa[1](0) 0 (ω) (red full), ℑa[1] 0 (ω) (red dashed), ℜ ¯A[1](ω) (black full), ℑ ¯A[1](ω) (black dashed). Lower left-hand panel: ρ(0)(ω) (blue full), ρ(ω) (red full) ¯R(ω) (black full). Lower right-hand panel: α(0)(ω) (blue full), α(ω) (red full) ¯A(ω) (black full), output fluxes (dashed). h = 0.02 m, w = 0.015 m, d = 0.02 m, θi = 0◦, M = 0, and the foam parameters taken (see sect. 2.2.1) from [37]. (ω) (blue dashed), ℜa[0]− (ω) (blue full), ℑa[1](0) 0 0 (ω) (red full), ℑa[0]− (ω) (blue dashed), ℜa[1] 0 0 0 34 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) g n i t a r g / r e y a l n i s e d u t i l p m a 4 2 0 −2 −4 0 2000 8000 10000 1 0.8 0.6 0.4 0.2 s e x u l f t u p t u o & d e b r o s b a 8000 10000 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 Figure 10: Same as fig. 9 except that θi = 20◦. 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) g n i t a r g / r e y a l n i s e d u t i l p m a 4 2 0 −2 −4 0 2000 8000 10000 1 0.8 0.6 0.4 0.2 s e x u l f t u p t u o & d e b r o s b a 8000 10000 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 Figure 11: Same as fig. 9 except that θi = 40◦. 35 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) g n i t a r g / r e y a l n i s e d u t i l p m a 4 2 0 −2 −4 0 2000 8000 10000 1 0.8 0.6 0.4 0.2 s e x u l f t u p t u o & d e b r o s b a 8000 10000 0 0 2000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 Figure 12: Same as fig. 9 except that θi = 60◦. 36 It is seen from these figures that: 1. In the low frequency region depicted in the figures, the effective model responses (red curves) appear to be quite close to the M = 0 grating responses (the latter having been shown numerically in figs. 4-7 to be adequate approximations at low frequencies and small incident angles) for small θi (up to ∼ 40◦) as one would expect from the condition (102); this authorizes the use of the effective layer model instead of the grating model to predict the low frequency, small incident angle response of the grating. 2. The black curves in the upper right-hand panels of figs. 9-10 constitute numerical verifications, for a model of foam with dispersive, complex density and wavespeed, of the predictions of the analysis of sect. 3.6, even though these predictions were based on the assumption of real, non- dispersive density and wavespeed. In particular, this analysis predicts quite well: (i) the frequency f0 of the lowest-frequency response peak in the layer provided the angle of incidence is near-normal, (ii) the fact that the real part of the amplitude A[1] in the layer is nil at this frequency and (iii) the fact that the behavior of A[1] is not really resonant near this peak even though the quarter wavelength formula is commonly-thought to be associated with a resonant process. 3. The black curves in the upper and lower right-hand panels of figs. 9-10 show that the frequency f0 of the pseudo-resonance in A[1] does not coincide with the maximum of absorbed flux, this being due to the fact that the latter depends not only on k A[1] A[0]+k2 but also on a slowly-varying (with respect to ω) factor E(ω) as per (67). 4. The comparison, relative to the absorption in the lower right-hand panels of figs. 9-10, of the red-blue curves (grating and effective layer) with the black curves (reference layer absorbed flux designated by ¯A) shows that: (i) not only are the peaks of the red-blue curves shifted to lower frequencies with respect to the black curves, (ii) but also the red-blue peaks are higher (although narrower) than the black peaks, this meaning that the replacement of the reference layer by the grating enables the sought-for result of increasing the absorption and producing this increase at lower frequencies. We have not attempted to analytically prove this result (which has also been observed in publications such as [37]), but rather verified numerically its apparently-systematic nature. 6 Increasing the absorption at low frequencies treated as an opti- mization problem 6.1 Analytical approach The QWR analysis and numerical examples (not shown here) indicate that: (i) the thickness h of the layer/grating particularly influences the positions of the response peaks, with the number of these peaks in a frequency interval increasing with h, and (ii) the transverse filling factor (of foam relative to all the material in one period d, i.e., Φ = w/d) principally influences the height of the absorption peaks. This suggests, for a given thickness h and given foam characteristics, that the optimization (i.e., to attain maximal absorption) be carried out for varying Φ, or, in other words, that the partial derivative of the normalized absorbed flux function with respect to Φ be equal to zero. Since it was shown that α(ω) accounts quite well for the exact low-frequency response, it is legitimate to adopt α(ω) (instead of the quasi-exact grating absorbed flux function which exists only numerically) as a proxy for α(ω). The interest of doing so is, of course, that we thus dispose of 37 a simple, explicit, algebraic expression for α(ω) which can easily be differentiated. Moreover, it was underlined that, by definition, A(ω) = α(ω), so that, by associating A(ω) with the first method of solution of the inverse problem, varying Φ amounts to varying the density of the effective layer, since R[1] = ρ[1]/Φ , (113) thus affording some physical insight into the absorption optimization problem. We found previously (73) that, for a rigidly-backed homogeneous layer of thickness H submitted to an airborne acoustic plane wave: A(ω) = 4ℑ(GSC ∗) kCk2 + kGSk2 + 2ℑ(GSC ∗) , (114) so that, on account of (113) (i.e., the replacement of the homogeneous layer by the effective layer), A(ω) = 4Φℑ(g S C ∗) k Ck2 + Φkg Sk2 + 2Φℑ(g S C ∗) = 4Φℑ(gsc∗) kck2 + Φkgsk2 + 2Φℑ(gsc∗) = α(ω) , (115) in which g = ρ[0] ρ[1] k[1] z0 k[0] z0 . The partial derivative of A(ω) with respect to Φ translates to A,Φ(ω) = −Φ2kgsk2 + kck2 = 0 , from which we deduce (since Φ must be positive) (116) (117) (118) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ρ[1]k[0] z0 ρ[0]k[1] z0(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Previous numerical results show that the characteristics of the foam, ρ[1] and c[1] are slowly-varying Φopt(ω) = kck kgsk (cid:13)(cid:13)(cid:13) z0 h(cid:17)(cid:13)(cid:13)(cid:13) cot(cid:16)k[1] functions of frequency except at very low frequencies, so that the factor(cid:13)(cid:13)(cid:13)(cid:13) frequencies) on account of the factor cot(cid:16)k[1] . function of ω, which means that Φopt(ω) is a quasi-periodic function of ω (other than at very low a more precise picture of the spectral characteristics of Φopt. z0 h(cid:17). We shall resort to numerics further on to obtain Another feature of (118) is that kck can exceed kgsk in which case Φopt(ω) exceeds 1, which is nonsense since the filling fraction is defined by w/d and the width w of the foam-loaded grooves cannot exceed the period d. This means that that values of h and/or ω for which the optimal filling fraction exceeds 1 must be excluded from the optimization analysis, as will be explained in the commentary of the numerical results. ρ[1]k[0] z0 ρ[0]k[1] z0(cid:13)(cid:13)(cid:13)(cid:13) is a slowly-varying Of great interest is to find out how A(cid:12)(cid:12)(cid:12)Φ=Φopt and how it compares with the absorbed flux in the reference configuration of a rigidly-backed homogeneous foam layer (with characteristics H=h, ρ[1] and c[1]), submitted to an airborne plane varies with the various structural parameters 38 acoustic wave (R =[0]= ρ[0], C [0] = c[0] and A[0]+ = a[0]+) previously-designated by ¯A. It is easily shown that (119) = 2 For the rigidly-backed homogeneous (i.e., not the effective) layer, (114) entails A(ω)(cid:12)(cid:12)(cid:12)Φ=Φopt . 1 + kgsck ℑ(gsc∗) ¯A(ω) = 4ℑ(gsc∗) kck2 + kgsk2 + 2ℑ(gsc∗) = 2 (cid:16) kck2+kgsk2 ℑ(gsc∗) 2 (cid:17) 1 + . (120) But any two complex numbers Z1 and Z2 satisfy the inequality kZ1k2+kZ1k2 2 ≥ kZ1Z2k, so that ¯A(ω) ≤ A(ω)(cid:12)(cid:12)(cid:12)Φ=Φopt ; ∀ω ≥ 0 , (121) which means that the optimal absorption in the rigidly-backed grating structure with alternating foam and rigid regions is larger than or equal to, at all frequencies, the absorption in the rigidly- backed homogeneous foam (the same as that of the grating) layer of thickness equal to that of the grating. This, of course, is the desired feature of using the grating instead of the homogeneous layer. A last remark concerning (119): this expression shows that the absorption can be total (i.e. A = 1 when kgsck = ℑ(gsc∗) . (122) Finding the h, ω pairs for which this total absorption can occur requires solving the non-linear equation (122), a task that is not addressed herein. 6.2 Numerical approach 6.2.1 Decreasing the frequency for optimal absorption by increasing h for the foam of [37] In figs. 13-15 we plot Φopt(f ) (top panels) and A(f )(cid:12)(cid:12)(cid:12)Φ=Φopt as well as ¯A(f ) (bottom panels). 39 102 101 100 t p o Φ 10−1 0 s e x u l f d e b r o s b a r e y a l f e r d n a d e z m i i t p O 1 0.8 0.6 0.4 0.2 0 0 X: 2085 Y: 0.4013 1000 2000 3000 4000 5000 f (Hz) 6000 7000 8000 9000 10000 1000 2000 3000 4000 5000 f (Hz) 6000 7000 8000 9000 10000 Figure 13: Φopt(f ) (blue curve in the top panel in which the green line denotes the physical limit (blue curve in the bottom panel) and ¯A(f ) (red curve in the bottom panel). Case h = 0.03 m, θi = 0◦ for the constitutive parameters of the foam of sect. 2.2.1. of Φ, i.e., Φ = 1), A(f )(cid:12)(cid:12)(cid:12)Φ=Φopt 104 102 100 t p o Φ 10−2 0 s e x u l f d e b r o s b a r e y a l f e r d n a d e z m i i t p O 1 0.8 0.6 0.4 0.2 0 0 X: 1019 Y: 0.7043 500 1000 f (Hz) 1500 2000 2500 500 1000 f (Hz) 1500 2000 2500 Figure 14: Same as fig. 13 except that h = 0.06 m. 40 104 102 100 t p o Φ 10−2 0 s e x u l f d e b r o s b a r e y a l f e r d n a d e z m i i t p O 1 0.8 0.6 0.4 0.2 0 0 X: 723.2 Y: 0.4006 200 400 600 800 1000 f (Hz) 1200 1400 1600 1800 2000 200 400 600 800 1000 f (Hz) 1200 1400 1600 1800 2000 Figure 15: Same as fig. 13 except that h = 0.1 m. 41 These results call for the following comments: 1. Although the absorption is optimal for the indicated h and ω, it is not necessarily total (i.e. equal to 1). 2. Although the absorption at Φopt is optimal for the indicated h and ω, it is not generally-optimal for other h and ω. 3. As observed in the figures, the absorption can be total (i.e., equal to 1) for certain combinations of Φopt, ω, h. 4. Φopt > 1 is meaningless since the width w of the foam regions of the grating cannot exceed the period d. This means that the blue curve in a given bottom panel is meaningful only in the frequency intervals for which the blue curve is below the cyan line in the corresponding upper panel. 5. As predicted theoretically, the absorption of the optimized grating (actually its proxy) is superior, at all allowable frequencies, to that of the reference rigidly-backed homogeneous layer (both of thickness h) filled with the same foam as that of the grating. does not depend on 6. The position (in terms of frequency) of optimal absorption A(f )(cid:12)(cid:12)(cid:12)Φ=Φopt either w relative to f nor on d relative to f , i.e., it depends only on their ratio w/d. This is a prediction of a quasi-static model of the grating response and does not necessarily hold for high frequencies (relative to w and d). We shall see further on what the effect is of changing w and d while maintaining constant their ratio Φopt = w/d. 7. The positions (in terms of frequency) of the maxima of the peaks of A(f )(cid:12)(cid:12)(cid:12)Φ=Φopt are shifted to the left (i.e., to lower frequencies) with respect to the maxima of the peaks of ¯A(f ). This is the second desired feature of employing the grating, stressed, and also observed, in (fig. 3 of) [37], but we have been unable to mathematically prove its systematic nature and establish its amount, this being so because of the dispersive nature of ρ[1] and c[1]. 6.2.2 Decreasing the frequency for optimal absorption by increasing h for the foam of [41] The results in [41] seem to apply to a rigidly-backed reference layer that is relatively thin (the authors of this publication write that the thickness is 8 mm, but this figure seems to us to be questionable). Thus, we felt it to be useful to see what happens for relatively-thin foam fillers or layers. This is done in figs. 16-17. 42 104 102 100 10−2 0 1 0.8 0.6 0.4 0.2 0 0 X: 9852 Y: 0.08071 5000 f (Hz) 10000 15000 5000 f (Hz) 10000 15000 t p o Φ s e x u l f d e b r o s b a r e y a l f e r d n a d e z m i i t p O Figure 16: Φopt(f ) (blue curve in the top panel in which the green line denotes the physical limit (blue curve in the bottom panel) and ¯A(f ) (red curve in the bottom panel). Case h = 0.008 m, θi = 0◦ for the constitutive parameters of the foam of sect. 2.2.2. of Φ, i.e., Φ = 1), A(f )(cid:12)(cid:12)(cid:12)Φ=Φopt t p o Φ 104 102 100 10−2 0 s e x u l f d e b r o s b a r e y a l f e r d n a d e z m i i t p O X: 6513 Y: 0.1002 1000 2000 3000 4000 5000 f (Hz) 6000 7000 8000 9000 10000 1 0.8 0.6 0.4 0.2 0 0 1000 2000 3000 4000 5000 f (Hz) 6000 7000 8000 9000 10000 Figure 17: Same as fig. 16 except that h = 0.012 m, θi = 0◦. 43 We thus see that increasing h results in the shift of the the total absorption peak to lower frequencies. Otherwise, these results call for the same comments as in the previous section. 6.3 Persistence of the enhanced absorption effect for other-than-normal inci- dence Here we start with the optimization parameters of fig. 13, i.e., Φ = 0.4, h = 0.03 m, in a low- frequency regime, which were obtained for normal incidence θi = 0◦, and inquire as to whether the enhanced (actually total) absorption obtained for normal incidence persists for other angles of incidence as well. This question is of importance in sound absorbiton as well as electromagnetic wave energy harvesting applications. x u l f d e b r o s b a 0 = θ i x u l f d e b r o s b a 0 4 = θ i 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 1000 1000 2000 3000 f (Hz) 2000 3000 f (Hz) x u l f d e b r o s b a 0 2 = θ i x u l f d e b r o s b a 0 6 = θ i 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 4000 5000 4000 5000 1000 1000 2000 3000 f (Hz) 2000 3000 f (Hz) 4000 5000 4000 5000 Figure 18: In all four panels: α(1)(ω) (blue full), α(ω) (red full) ¯A(ω) (black full). The upper left-hand, upper right-hand, lower left-hand and lower right-hand panels apply to θi = 0, 20, 40, 60◦ respectively. The foam parameters are those of sect. 2.2.1, h = 0.03 m, w = 0.004 m, d = 0.01 m. 44 The answer provided by this figure is clearly affirmative for angles of incidence (in absolute value, and for this choice of grating parameters) that do not exceed 40◦. 6.4 Numerical results for the M > 0 grating response concerning the effect of increasing w and d for given Φopt to see if optimal absorption is maintained or changed otherwise Φopt was obtained from an analysis employing a proxy of the grating which is an outcome of a low frequency (i.e., M = 0) approximation. In fact, the frequency does not intervene explicitly in the expression for Φopt nor in that of the corresponding absorbed flux other than through the frequency-dependent properties of the filler material. Thus, it is not at all obvious that the desired increase of absorption (over that of the reference rigidly-backed layer) will be maintained if the frequency is increased beyond static or quasi-static conditions. The purpose of the graphs in this section is therefore to show how, for a given Φopt and h, the absorbed flux of the proxy evolves with increasing w and d while maintaining w/d constant at the value Φopt. Since the parameters w and d do not enter in the response of the reference rigidly- backed layer, this response (black curves in the following figures) does not change with w, d. The response of the proxy does depend on w/d but this fraction is constant as a function of f so that the proxy response (red curves) does not change either with w, d. But, of course, the grating response (blue curves) is expected to change with w, d. 6.4.1 Φopt = 0.7, h = 0.06 m, and the foam of [37] Figs. 19-22 apply to the foam of sect. 2.2.1. 45 1 0.5 0 r i a n i s e d u t i l p m a −0.5 0 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t t u p u o & d e b r o s b a 2 1 0 −1 −2 0 1 0.8 0.6 0.4 0.2 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 0 Figure 19: Upper left-hand panel: ℜa[0]−(M ) (ω) (red full), ℑa[0]− (ω) (red dashed), ℜ ¯A[0]−(ω) (black full), ℑ ¯A[0]−(ω) (black dashed). Upper right- hand panel: ℜa[1](M ) (ω) (blue full), ℑa[1](M ) 0 (ω) (red dashed), ℜ ¯A[1](ω) (black full), ℑ ¯A[1](ω) (black dashed). Lower left-hand panel: ρ(M )(ω) (blue full), ρ(ω) (red full) ¯R(ω) (black full). Lower right-hand panel: α(M )(ω) (blue full), α(ω) (red full) ¯A(ω) (black full), output fluxes (dashed). h = 0.06 m, w = 0.028 m, d = 0.04 m, θi = 0◦, M = 3, the constitutive parameters of the foam are those of sect. 2.2.1. (ω) (blue dashed), ℜa[0]− 0 (ω) (red full), ℑa[1] (ω) (blue full), ℑa[0]−(M ) (ω) (blue dashed), ℜa[1] 0 0 0 0 0 46 1 0.5 0 r i a n i s e d u t i l p m a −0.5 0 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 g n i t a r g / r e y a l n i s e d u t i l p m a 1000 f (Hz) 2000 3000 t s e x u l f t u p u o & d e b r o s b a 1000 f (Hz) 2000 3000 2 1 0 −1 −2 0 1 0.8 0.6 0.4 0.2 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 Figure 20: Same as fig. 19 except that w = 0.14 m, d = 0.2 m. 1 0.5 0 r i a n i s e d u t i l p m a −0.5 0 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t u p t u o & d e b r o s b a 2 1 0 −1 −2 0 1 0.8 0.6 0.4 0.2 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 Figure 21: Same as fig. 19 except that w = 0.21 m, d = 0.3 m. 47 1 0.5 0 r i a n i s e d u t i l p m a −0.5 0 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t t u p u o & d e b r o s b a 2 1 0 −1 −2 0 1 0.8 0.6 0.4 0.2 0 0 1000 f (Hz) 2000 3000 1000 f (Hz) 2000 3000 Figure 22: Same as fig. 19 except that w = 0.28 m, d = 0.4 m. 48 6.4.2 Φopt = 0.1, h = 0.012 m, and the foam of [41] Figs. 23-26 apply to the foam of sect. 2.2.2. 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t t u p u o & d e b r o s b a 15 10 5 0 −5 −10 0 2000 1 0.8 0.6 0.4 0.2 0 0 2000 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 0 0 Figure 23: Upper left-hand panel: ℜa[0]−(M ) (ω) (red full), ℑa[0]− (ω) (red dashed), ℜ ¯A[0]−(ω) (black full), ℑ ¯A[0]−(ω) (black dashed). Upper right- hand panel: ℜa[1](M ) (ω) (blue full), ℑa[1](M ) 0 (ω) (red dashed), ℜ ¯A[1](ω) (black full), ℑ ¯A[1](ω) (black dashed). Lower left-hand panel: ρ(M )(ω) (blue full), ρ(ω) (red full) ¯R(ω) (black full). Lower right-hand panel: α(M )(ω) (blue full), α(ω) (red full) ¯A(ω) (black full), output fluxes (dashed). h = 0.012 m, w = 0.001 m, d = 0.01 m, θi = 0◦, M = 6, the constitutive parameters of the foam are those of sect. 2.2.2. (ω) (blue dashed), ℜa[0]− 0 (ω) (red full), ℑa[1] (ω) (blue full), ℑa[0]−(M ) (ω) (blue dashed), ℜa[1] 0 0 0 0 49 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 g n i t a r g / r e y a l n i s e d u t i l p m a t s e x u l f t u p u o & d e b r o s b a 15 10 5 0 −5 −10 0 2000 1 0.8 0.6 0.4 0.2 0 0 2000 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) Figure 24: Same as fig. 23 except that w = 0.005 m, d = 0.05 m. 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t u p t u o & d e b r o s b a 15 10 5 0 −5 −10 0 2000 1 0.8 0.6 0.4 0.2 0 0 2000 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) Figure 25: Same as fig. 23 except that w = 0.008 m, d = 0.08 m. 50 1 0.5 0 −0.5 r i a n i s e d u t i l p m a −1 0 2000 1 0.8 0.6 0.4 0.2 s e x u l f d e t c e l f e r 0 0 2000 g n i t a r g / r e y a l n i s e d u t i l p m a s e x u l f t t u p u o & d e b r o s b a 20 10 0 −10 0 2000 1 0.8 0.6 0.4 0.2 0 0 2000 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) 8000 10000 8000 10000 4000 6000 f (Hz) 4000 6000 f (Hz) Figure 26: Same as fig. 23 except that w = 0.010 m, d = 0.10m. 51 Note that the maximum absorption is nowhere near the 500 Hz area, even for the largest w and d, and nowhere near the ∼ 700 Hz peak in fig. 4 of [41]. 6.4.3 Comments on the 'red shift' of the total absorption peaks 1. The shift to lower frequencies (i.e., the 'red shift') of the total absorption peak is obtained only for α(M >0) (i.e., neither for α which depends only on the ratio w/d, nor for ¯A which does not even depend on w/d) by increasing both w and d while maintaining their ratio w/d at the chosen value Φopt. 2. The bandwidth of substantial absorption near this peak decreases with increasing w and d. 3. The difference between the grating absorption and the reference layer absorption at the location of the maximum absorption increases with increasing w and d essentially because the absorption of the reference configuration decreases with frequency in the region of maximal grating absorption. 4. Substantial very-low frequency absorption cannot be obtained other than in a relatively-small bandwidth and with relatively thick gratings. 5. Nothing close to the results in fig. 4 of [41] have been obtained for our grating with the foam material of [41] . 6. The red shift (and Wood anomalies [60], marked by kinks or near-discontinuous behavior in response functions) can probably be explained using the M = 1 approximation of grating response as is done in [98]. 52 6.5 Homothetic increase of h,w and d for Φopt = 0.4, h/d = 1, and the foam of [37] In the previous figure, h was kept constant so that the condition of optimality was maintained. Now (fig. 27) we augment h while augmenting w and d in the same proportions, so that we are thus increasingly departing from the condition of optimality. x u l f d e b r o s b a x u l f d e b r o s b a 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 x u l f d e b r o s b a x u l f d e b r o s b a 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 0.5 1 f (Hz) 1.5 2 x 104 0.5 1 f (Hz) 1.5 2 x 104 0.5 1 f (Hz) 1.5 2 x 104 0.5 1 f (Hz) 1.5 2 x 104 Figure 27: Upper left-hand panel: hand panel: {0.03 m, 0.0075 m, 0.03m}. Lower left-hand panel: {h, w, d} = {0.04 m, 0.0100 m, 0.04m}. {h, w, d} = {0.01 m, 0.0025 m, 0.01m}. Upper right- {h, w, d} = {h, w, d} = {0.02 m, 0.0050 m, 0.02m}. Lower left-hand panel: 53 This prediction is confirmed in the figure. Moreover, for the largest h we can even arrive at a situation in which the height of the lowest frequency peak is lower than that the height of the lowest frequency reference layer peak, which means that for this choice of parameters, making the layer inhomogeneous is of no use. 7 Conclusion As mentioned in the Introduction, the present investigation was inspired by what we learned in [37], and, in particular by the numerical result exhibited in fig. 3 of this paper concerning the gain in low-frequency absorption enabled by replacing a rigidly-backed foam-filled layer of thickness h by the same layer containing a periodic (along the x axis) distribution of rigid, circular cylinders. Owing to the fact that such a structure is not easy to fabricate, we wondered whether a material interchange of the foam and cylinder components might enable a similar gain of absorption (over that of the reference rigidly-backed foam-filled layer), and if so, would it be possible to explain this gain in a simple manner. To do this, we chose to replace the generic circular cylinder by a rectangular cylinder whose height equals h , thus enabling the field representations to be much simpler than for a circular cylinder. Our first numerical trials led to a numerical result depicted in fig. 28 concerning the absorption of a grating composed of a periodic distribution of foam-filled rectangular cylinder grooves. This grating configuration is rather close to the one in fig. 3 of [37], notably by the fact that the thickness of both gratings are the same, i.e., h = 0.02 m, but, of course, the host medium in the layer is now rigid (instead of being a foam) and the medium in the grooves is now a foam (instead of being rigid). Despite this difference, something quite similar, notably as concerns the position and height of the r i a n i s e d u t i l p m a 1 0.5 0 −0.5 −1 s e x u l f d e t c e l f e r 1 0.8 0.6 0.4 0.2 0 0.5 1 f (Hz) 1.5 2 x 104 0.5 1 f (Hz) 1.5 2 x 104 g n i t a r g / r e y a l n i s e d u t i l p m a 6 4 2 0 −2 −4 s e x u l f t u p t u o & d e b r o s b a 1 0.8 0.6 0.4 0.2 0 0.5 1 f (Hz) 1.5 2 x 104 0.5 1 f (Hz) 1.5 2 x 104 Figure 28: Same as fig. 19 except that h = 0.02 m, w = 0.005 m, d = 0.02 m. reference layer and grating lowest frequency peaks, to the results in fig. 3 of [37] is observed for 54 our grating with the same foam material as in [37]. The gain of absorption in our lowest-frequency peak (over that of the lowest-frequency reference layer peak) is obtained for a transverse area of foam material of 10−4 m2 per period whereas the comparable absorption gain of [37] is obtained for a transverse area of foam material of 2.233 × 10−4 m2 per period which means that the same absorption gain is obtained in our configuration with less than twice the amount of foam material employed in [37], even though the two gratings have the same thickness. Moreover, our grating appears to be structurally more sound and easier to fabricate than the grating of rigid circular cylindrical inclusions within a foam layer. The main advantage of our grating is that it enables a very simple rigorous analysis of its re- sponse to an acoustic wave as well as an even-simpler, mathematically-explicit, approximation of its low-frequency response. The expression of this low-frequency (amplitude A[1](ω)) response in the heterogeneous layer region of the rigidly-backed grating configuration turns out to be mathemati- cally identical to the corresponding response in the same region of a rigidly-backed homogeneous (effective) foam layer configuration of the same thickness as that of the grating provided that the (effective) density R[1] of the layer is taken to be the density ρ[1] of the foam filler of the grating grooves divided by the areal filling fraction w/d, wherein w is the generic groove width and d the grating period. We showed, by comparison with the rigorous solution, that A[1](ω) is a quite ac- curate approximation of the rigorous amplitude a[1] 0 (ω) in the low-frequency region of interest and for incident angles that do not exceed ∼ 40◦. Moreover we showed that the approximation of the absorbed flux A is expressed as the product of a term involving k A[1](ω)k2 (that depends on w/d) with a trigonometric term E(ω) that does not depend on w/d but rather on the thickness h. It is not easy to explain theoretically (for a given h) the position and height of the lowest-frequency absorption peak produced by our grating because of the dispersive nature of the foam material, but a numerical study has enabled to establish the principal tendencies. Above all, the expression for A[1](ω) can differentiated with respect to Φ = w/d in order to find the optimal Φ = Φopt (i.e., in the sense of maximizing the absorption), and from this even find the frequency at which total absorption can be expected to occur. Subsequently, we showed theoretically that the absorption produced by the so-optimized grating is always equal to or greater than that of the homogeneous reference layer (also of thickness h) filled with the same foam material as that of the filler material in the grooves of the grating. As mentioned in the Introduction, it can also useful to lower the frequency of occurrence of the lowest-frequency absorption peak (such as in the design of anechoic chambers). This can be done by increasing h, but usually results in a lowering of height of this peak. We thus decided to find out if another strategy might lead to the desired result. With the approximate response as a starting point, and a Φopt obtained from this response as an invariant (i.e., maintaining w/d constant), we chose to increase w and d at the same rate and, by so doing, depart from the quasi- static situation which previously enabled the simple approximation of response. Thus, we were obliged to resort to a numerical exploitation of our rigorous solution which revealed the interesting fact that increasing w and d, while maintaining w/d at the value Φopt can enable the lowering of the frequency of occurrence of the first low-frequency absorption peak without affecting its height (which is maintained at the total absorption level). Unfortunately, this is obtained at the expense of narrowing the bandwidth of large absorption around this peak. Among the perspectives of this work we may cite; 1. The M = 1 approximation of our 2D grating response will probably enable the explanation (in the manner of [98]) of Wood's anomalies and the red shift of the total absorption peak; 55 2. It should be rather easy to extend, as in [14], this type of rigorous and approximate analysis to a 2D grating composed of a periodic (in the x and y directions) array of foam filled box-shaped troughs, the interesting question being whether this structure will enable even larger absorption at even lower frequencies. 3. It should be rather easy to extend (in the manner of [6]) this type of rigorous and approximate analysis to a 1D grating with structured (e.g., multi-stepped) grooves [71], the interesting question again being whether this may enhance the sought-for effects; 5. It might be possible to enhance the absorption at very low frequencies (< 500 Hz) by appealing to better filler materials [7, 8], and the present analysis could be applied to such configurations without fundamental changes ((i.e., only the effective parameters ρe(ω) and ce(ω) change from their values given in sects. 2.2.1 and 2.2.2). References [1] Achaoui Y., Ungureanu B., Enoch S., Brul´e S. and Guenneau S., Seismic waves damping with arrays of inertial resonators, Extreme Mech. Lett., 8, 30-37 (2016). [2] Ahmed H., Schlenkhoff A., Rousta R. and Abdelaziz R., Experimental and semi-analytical investigation of wave interaction with double vertical slotted walls, Int. J. Environ. Ecolog. Engrg., 8(8), 1447-1455 (2014). [3] Aki, K. and K. L. Larner K.L., Surface motion of a layered medium having an irregular interface due to incident plane SH waves, J. Geophys. Res. 75, 933-954 (1970). [4] Allard J.F. and Atalla N., Propagation of Sound in Porous Media: Modeling Sound Absorbing Materials, Wiley, Chichester, Ch. 5, 73-107 (2009). [5] Ando Y., Concert Hall Acoustics, Springer, Berlin (1985). [6] Ando Y. and Kato K., Calculations on the sound reflection from periodically uneven surfaces of arbitrary profile, Acta Acust. Acustica, 35(5), 321-329 (1976). [7] Arenas J.P. and Crocker M.J., Recent trends in porous sound-absorbing materials, Sound and Vibr., July, 12-17 (2010). [8] Attenborough K., Microstructures for lowering the quarter wavelength resonance frequency of a hard-backed rigid-porous layer, Appl. Acoust., 130, 188-194 (2018). [9] Bard P.-Y. and Bouchon M., The two-dimensional resonance of sediment-filled valleys, Bull. Seism. Soc. Am., Vol. 75(2), 519-541 (1985). [10] Bermel P., Ghebrebrhan M., Chan W., Yeng Y.X., Araghchini M., Hamam R., Marton C.H., Jensen K.F., Soljacic M., Joannopoulos J.D., Johnson S.G. and Celanovic I., Design and global optimization of high-effciency thermophotovoltaic systems, Opt. Express, 18(103), A314-A334 (2010). [11] Biot M.A., Theory of propagation of elastic waves in a fluid-saturated porous solid. I. Low- Frequency Range, J. Acoust. Soc. Am., 28(2) , 168-178 (1956). 56 [12] Biot M.A., Theory of propagation of elastic waves in a fluid-saturated porous solid. II. High- Frequency Range, J. Acoust. Soc. Am., 28(2) , 179-191 (1956). [13] Bonod N., Tayeb G., Maystre D., Enoch S., and Popov E., Total absorption of light by lamellar metallic gratings, Opt. Express, 16(20), 15431 (2008). [14] Bos L., Ducourneau J., Planeau V. and Lami S.S., Acoustic reflection coefficient of periodically uneven facings in industrial rooms, Forum Acusticum, 2437-2442 (2005). [15] Carretero-Palacios S., Murray A.R.J., Martn-Moreno L. and Hibbins A.P., Broadband and broadangle extraordinary acoustic transmission through subwavelength apertures surrounded by fluids, New J. Phys., 16, 083044 (2014). [16] Catalkaya I. and Kent S., Analysis of multiple wedges electromagnetic wave absorbers, Prog. In Electromag. Res. M, 26, 19 (2012). [17] Champoux Y. and Allard J.F., Dynamic tortuosity and bulk midulus in air-saturated porous media, J. Appl. Phys., 70, 1975-1979 (1991). [18] Chang K.C., Shah V. and Tamir T., Scattering and guiding of waves by dielectric gratings with arbitrary profiles, J. Opt. Soc. Am., 70(7), 804-813 (1980). [19] Chen Y.Y., Barnhart M.V., Chen J.K., Hu G.K., Sun C.T. and Huang G.L., Dissipative elastic metamaterials for broadband wave mitigation at subwavelength scale, Composite Structs., 136, 358-371 (2015). [20] Contractor R., D'Atuanno G. and Menyuk C., Ultra-broadband, polarization-independent, wide-angle absorption in impedance-matched metamaterials with anti-reflective moth-eye sur- faces, Opt. Express, 26(18), 24031 (2018)/ [21] Cui Y., Fung K.H., Xu J., He S., and Fang N.X., Multiband plasmonic absorber based on transverse phase resonances, Optics Express, 20(16), 17552 (2012). [22] Cuomo J., Ziegler J. and Woodhall J., A new concept for solar energy thermal conversion, Appl. Phys. Lett. 26, 557-559 (1975). [23] De Bruijn A., The sound absorption of an absorbing periodically uneven surface of rectangular profile, Acta Acust. Acustica, 18(3), 123-131 (1967). [24] De Bruijn A., Anomalous effects in the sound absorption of periodically uneven surfaces, Acustica, 24, 75-84 (1971). [25] Delany M.E. and Bazley E.N., Acoustical properties of fibrous absorbent materials, Appl. Acoust. 3, 105-116 (1970). [26] De Lima Monteiro D.W., Honorato F.P., De Oliveira Costa R.F. and Salles L.P., Surface texturing with hemispherical cavities to improve effciency in silicon solar cells, Int. J. Pho- toenergy, 743608 (2012). 57 [27] De Ryck L., Groby J.-P., Leclaire P., Lauriks W., Wirgin A., Depollier C. and Fellah Z.E.A., Acoustic wave propagation in a macroscopically inhomogenous porous medium saturated by a fluid, Appl. Phys. Lett., 90, 18901 (2007). [28] Diaz-Rubio A., Torrent D., Carbonell J., and Sanchez-Dehesa J., Extraordinary absorption by a thin dielectric slab backed with a metasurface, Phys. Rev. B, 89, 245123 (2014). [29] Emerson W.H., Electromagnetic wave absorbers and anechoic chambers through the years, IEEE Trans. Antenn. Propagat., AP-21, 484490 (1973). [30] Evans D.V. and Porter R., Trapped modes about multiple cylinders in a channel, J. Fluid Mech., 339, 331-356 (1997). [31] Ewing W.M., Jardetzky W.S. and Press F., Elastic Waves in Layered Media, McGraw-Hill, New York (1957). [32] Garcia-Vidal F.J. and Martin-Moreno L., Transmission and focusing of light in one- dimensional periodically nanostructured metals, Phys. Rev. B, 66, 155412 (2002). [33] Garcia-Vidal F.J., Sanchez-Dehesa J., Dechelette A., Bustarret E., Lopez-Rios T., Fournier T., and Pannetier B. , Localized surface plasmons in lamellar metallic gratings, J. Lightwwave Tech., 17(11) (1999). [34] Ghebrebrhan M., Bermel P., Yeng Y.X., Celanovic S.I., Soljacic M. and Joannopoulos J.D., Tailoring thermal emission via Q matching of photonic crystal resonances, Phys. Rev. A, 83, 033810 (2011). [35] Golomb M., Diffraction gratings and solar selective thin film absorbers: An experimental study, Opt. Comm., 27(2), 177-180 (1978). [36] Groby J.-P., Mod´elisation de la propagation des ondes elastiques g´en´er´ees par un s´eisme proche ou ´eloign´e `a lint´erieur dune ville, PhD thesis, Universit´e de la M´editerran´ee, Marseille (2005). [37] Groby J.-P., Dazel O., Duclos A., Boeckx L. and Kelders L., Enhancing the absorption co- efficient of a backesd rigid frame porous layer by embedding circular periodic inclusions, J. Acoust. Soc. Am., 130(6), 3771-3780 (2011). [38] Groby J.-P., Duclos A., Dazel O., Boeckx L. and Lauriks W., Absorption of a rigid frame porous layer with periodic circular inclusions backed by a periodic grating, J. Acoust. Soc. Am. 129 (5), 3035-3046 (2011). [39] Groby J.-P., Huang W., Lardeau A. and Aur´egan Y., The use of slow waves to design simple sound absorbing materials, J. Appl. Phys., 117, 124903 (2015). [40] Groby J.-P. , Lagarrigue C., Brouard B. , Dazel O., and Tournat V., Enhancing the absorption properties of acoustic porous plates by periodically embedding Helmholtz resonators, J. Acoust. Soc. Am., 137, 273 (2015). 58 [41] Groby J.-P., Lauriks W. and Vigran T.E.,Total absorption peak by use of a rigid frame porous layer backed by a rigid multi-irregularities grating, J. Acoust. Soc. Am., 127(5), 2865-2873 (2010). [42] Groby J.-P., Pommier R. and Aur´egan, Use of slow sound to design perfect and broadband passive sound absorbing materials, J. Acoust. Soc. Am., 139(4), 1660-1671 (2016). [43] Groby J.-P., Wirgin A., De Ryck L., Lauriks W., Gilbert R.P. and Xu Y.S., Acoustic response of a rigid-frame porous medium plate with a periodic set of inclusions, J. Acoust. Soc. Am., 126, 685-693 (2009). [44] Hibbins A.P., Rance H.J., Sambles J.R., Starkey T.A., Gareth P., Ward G.P. and Smith J.D., Absorbing and manipulating sound with metamaterials, Proc. ICSV24, London (2017). [45] Hollands K.G.T., Directional selectivity, emittance, and absorptance properties of vee corru- gated specular surfaces, Solar Energy, 7(3), 108-116 (1963). [46] Hong H. and Lau S.-K., Effects of inclusion shapes within rigid porous materials on acoustic performance, 164th Meeting of the Acoust. Soc. Am., Kansas City, (2012). [47] Horwitz C.M., Solar-selective globular metal films, J. Opt. Soc. Am. 68, 1032-1038 (1978). [48] Hutley M.C. and Maystre D., The total absorption of light by a diffraction grating, Optics Comm., 19(3), 431-436 (1976). [49] Jimenez N., Romero-Garca V. and Groby J.-P., Perfect absorption of aound by rigidly-backed high-porous materials, Acta Acust. Acustica, 104, 396-409 (2018). [50] Johnson D.L., Koplik J. and Dashen R., Theory of dynamic permeability and toruosity in fluid-saturated porous media, J. Fluid Mech., 176, 379-402 (1987). [51] Kaye G.W.C. and Evans E.J., The sound-absorbing properties of some common out-door materials, Proc. Phys. Soc., 52(3), 371 (1940). [52] Kollatou T.M., Dimitriadis A.I., Assimonis S.D., Kantartzis N.V. and Antonopoulos C.S., Multi-band, highly absorbing, microwave metamaterial structures, Appl. Phys. A, 115(2), 555-561 (2014). [53] Lagarrigue C., Groby J.-P., Dazel O. and Tournat V., Design of metaporous supercells by genetic algorithm for absorption optimization on a wide frequency band, Appl. Acoust., 102, 4954 (2016). [54] Lan W., Cui Y., Yang Q., Lo M.-F., Lee C.-S. and Zhu F., Broadband light absorption en- hancement in moths eye nanostructured organic solar cells, AIP Advances, 5, 057164 (2015). [55] Le Perchec J., Qu´emerais P., Barbara A. and Lopez-Rios T., Why metallic surfaces with grooves a few nanometers deep and wide may strongly absorb visible light, Phys. Rev. Lett., 100(6), 066408 (2008). 59 [56] Liu X., Gao J., Yang H., Liu H., Wang X. and Zhao J.., Silicon-based multilayer gratings with a designable narrowband absorption in the short-wave infrared, Opt. Express, 24(22), 25103 (2016). [57] Lockyear M.J., Hibbins A.P., Sambles J.R., Hobson P.A. and Lawrence C.R., Thin resonant structures for angle and polarization independent microwave absorption, Appl. Phys. Lett., 94(4), 041913 (2009). [58] Losada I.J., Losada M.A., and Baquerizo A., An analytical method to evaluate the efficiency of porous screens as wave dampers, Appl. Ocean Res., 15, 207-215 (1993). [59] Lynch D.W., Optical constants and their measurement, J. Physique Coll. supp. 2.5, 38(11), C5-21 -- C530 (1977). [60] Maradudin A.A. , Simonsen I., Polanco J. and Fitzgerald R.M., Rayleigh and Wood anomalies in the diffraction of light from a perfectly conducting reflection grating, J. Optics, 18(2), 024004 (2016). [61] Maystre D. and Petit R., Brewster incidence for metallic gratings, Optics Comm., 17(2), 196-200 (1976). [62] Maystre D. and Nevi`ere M., Quantitative theoretical study of the plasmon anomalies of diffrac- tion gratings, J. Opt., 8(3), 165-171 (1977). [63] Mechel P., Sound fields at periodic absorbers, J. Sound Vibr., 136(3), 1990, 379-412 (1990). [64] Mikoshiba K., Manimala T.J. and Sun C.T., Energy harvesting using an array of multifunc- tional resonators, J. Intell. Mater. Syst. Struct., 24. 168-179 (2013). [65] Nakamura T. , Kohno T., Morita Y., and Morishita S., Development of a pile-supported breakwater with dissipative front walls, in Breakwaters, coastal structures and coastlines, Allsop H.W.N (ed.), ICE, London (2016). [66] Nam Y., Yeng Y.X., Lenert A., Bermel P., Celanovic I., Soljacic M. and Wang E.N., So- lar thermophotovoltaic energy conversion systems with two-dimensional tantalum photonic crystal absorbers and emitters, Solar Ener. Mater. Solar Cells, 122, 287296 (2014). [67] Nennig B., Renou Y., Groby J.-P. and Aur´egan Y., A mode matching approach for modeling two dimensional porous grating with infinitely rigid or soft inclusions, J. Acoust. Soc. Am., 131(5), 3841-3852 (2012). [68] Niklasson G.A. Granqvist C.G., Surfaces for selective absorption of solar energy: an anno- tated bibliography 1955-1981, J. Mater. Sci., 18(12), 3475-3534(1983). [69] Palermo A., Krodel S., Marzani A. and Daraio C., Engineered metabarrier as shield from seismic surface waves, Scientific Repts. 6, 39356 (2016). [70] Pendry J.B. , Martin-Moreno L. and Garcia-Vidal F.J., Mimicking surface plasmons with structured surfaces, Science, Aug 6, 305(5685), 847-8 (2004). 60 [71] Peng S.T., Tamir T. and Bertoni H.L., Theory of periodic dielectric waveguides, IEEE Trans., MTT-23, 123-133 (1975). [72] Popov E., Total absorption of light in metallic gratings : a comparative analysis of spectral dependence for shallow and deep grooves, J. Mod. Opt., 36(5), 669-674 (1989). [73] Popov E., Maystre D., McPhedran R.C., Nevi´ere M., Hutley M.C. and Derrick G.H., Total absorption of unpolarized light by crossed gratings, Optics Express, 16(9), 6146 (2008). [74] Popov E. and Tsonev L., Comment on 'Resonant electric field enhancement in the vicinity of a bare metallic grating exposed to s-polarized light by A.A. Maradudin and A. Wirgin': Anomalous light absorption by lamellar metallic gratings, Surf. Sci., 271(3), 1992, L378-L382 (1992). [75] Porter R. and Evans D. V., Rayleigh-Bloch surface waves along periodic gratings and their connection with trapped modes in waveguides, J. Fluid Mech., 386, 233-258 (1999). [76] Porter R. and Evans D. V., Embedded Rayleigh-Bloch surface waves along periodic rectangular arrays, Wave Motion, 43, 29-50 (2005). [77] Qi J., Kettunen H., Wall´en H. and Sihvola A., Different homogenization methods based on scattering parameters of dielectric-composite slabs, Radio Sci., 46, RS0E08, doi:10.1029/2010RS004622 (2011). [78] Rahmani B., Bagheri A., Khavasi A. and Mehrany K., Effective medium theory for graphene covered metallic gratings, J. Opt., 18, 105005 (2016). [79] Rayleigh L., On the dynamical theory of gratings, , Prec. Roy. Soe. London A 79, 339-416 (1907). [80] Rayleigh, L., The Theory of Sound, Dover, N.Y., 2, 96 (1945). [81] Riesz F., Les Syst`emes dEquations Lin´eaires `a une Infnit´e dInconnues, Gauthier-Villars, Paris (1913). [82] R. Santbergen R. and Van Zolingen R.J.C., The absorption factor of crystalline silicon PV cells: A numerical and experimental study, Solar Ener. Mater. Solar Cells, 92(4), 432-444 (2008). [83] Saville P., Review of Radar Absorbing Materials, Technical Memorandum, DRDC Atlantic, TM 2005-003, (2005). [84] Sezawa, K. and Kanai K., Possibility of free oscillations of strata excited by seismic waves, Bull. Earthq. Res. Inst., 8, 1-11 (1930); 10, 1-18 (1932); 10, 273-298 (1932). [85] Shah V. and Tamir T., Brewster phenomena in lossy structures, Optics Comm., 23(1) 113-117 (1977). [86] Shah V. and Tamir T., Anomalous absorption by multi-layered media, Optics Comm., 37(6), 383-387 (1981). 61 [87] Sihvola A., Mixing models for heterogeneous and granular Media, in Advances in Electromag- netics of Complex Media and Metamaterials, Zouhdi S., Sihvola A. and Arsalane M. (eds.), Kluwer, Amsterdam (2002). [88] Stephens R.B. and Cody G.D., Inhomogeneous surfaces as selective solar absorbers, Proc. 2nd Ann. Conf. Absorber Surfaces for Solar Receivers, SERI, 125-138 (1979). [89] Tabor H., Solar collectors, selective surfaces,and heat engines, Proc. Natl. Acad. Sci. Israel, 47, 1271-1278 (1961). [90] Tao H., Bingham C., Strikwerda A., Pilon D., Shrekenhamer D., Landy N., Fan K., Zhang X., Padilla W. and Averitt R., Highly flexible wide angle of incidence terahertz metamaterial absorber: Design, fabrication, and characterization, Phys. Rev. B 78(24), 241103 (2008). [91] Tan K.T., Huang H.H. and Sun C.T., Optimizing the band gap of effective mass negativity in acoustic metamaterials, Appl. Phys. Lett., 101(24) (2012). [92] Teperik T.V. , Garcia de Abajo F.J., Borisov A.G., Abdelsalam M., Bartlett P.N., Sugawara Y. and Baumberg J.J. Omnidirectional absorption in nanostructured metal surfaces, Nature Photonics, doi:10/1038/nphoton.2008.76 (2008). [93] Thornton B.S., Limit of the moths eye principle and other impedance-matching corrugations for solar-absorber design, J. Opt. Soc. Am., 65(3), 267-270 (1975). [94] Utsunomiya T. and Eatock Talyor R., Trapped modes around a row of circular cylinders in a channel, J. Fluid Mech., 386, 259-279 (1999). [95] Wang J., Leistner P. and Li X., Prediction of sound absorption of a periodic groove structure with rectangular profile, Appl. Acoust., 73, 960968 (2012) [96] Wirgin A., Sur la r´eponse spectrale de certaines surfaces textur´ees, J. Phys. Coll. C1, 42, C1-57 -- C1-84 (1981). [97] Wirgin A., Resonant response of a soft semi-circular cylindrical basin to an SH seismic wave, Bull. Seism. Soc. Am., 85, 285-299 (1995). [98] Wirgin A., Effective medium description of the resonant elastic-wave response at the periodically-uneven boundary of a half-space, arXiv:1805.09823v1 [physics.comp-ph] (2018). [99] Wirgin A., Layer model description of the SH seismic response in anisotropically-idealized cities, Waves Rand. Complex Media, doi.org/10.1080/17455030.2018.1517950 (2018). [100] Witarto W., Wang S.J., Yang C.Y., Nie X., Mo Y.L., Chang K.C., Tang Y. and Kassawara R., Seismic isolation of small modular reactors using metamaterials, AIP Advances, 8, 045307 (2018). [101] Xue Y., Wang C., Sun Y., Wang W., Wu Y. and Ning Y., Preparation and spectral properties of solar selective absorbing M oSi2 − Al2O3 coating, Physica Status Solidi, 211(7), 1519-1524, (2014). 62 [102] Yeng Y.X., Ghebrebrhan M., Bermel P., Chan W.R., Joannopoulos J.D., Soljacic M. and Celanovic I., Enabling high-temperature nanophotonics for energy applications, Proc. Natl. Acad. Sci. U.S.A., 109(7), 2280-2285 (2012). [103] Yu J. and Mei C.C., Do longshore bars shelter the shore?, J. Fluid Mech., 404(10), 251-268 (2000) [104] Zhu D.T. and Xie Y.F., Hydrodynamic characteristics of offshore and pile breakwaters, Ocean Engrg., 104, 257-265 (2015). 63
1902.00254	1	1902	2019-02-01T09:57:43	Monatomic phase change memory	[ "physics.app-ph" ]	Phase change memory has been developed into a mature technology capable of storing information in a fast and non-volatile way, with potential for neuromorphic computing applications. However, its future impact in electronics depends crucially on how the materials at the core of this technology adapt to the requirements arising from continued scaling towards higher device densities. A common strategy to finetune the properties of phase change memory materials, reaching reasonable thermal stability in optical data storage, relies on mixing precise amounts of different dopants, resulting often in quaternary or even more complicated compounds. Here we show how the simplest material imaginable, a single element (in this case, antimony), can become a valid alternative when confined in extremely small volumes. This compositional simplification eliminates problems related to unwanted deviations from the optimized stoichiometry in the switching volume, which become increasingly pressing when devices are aggressively miniaturized. Removing compositional optimization issues may allow one to capitalize on nanosize effects in information storage.	physics.app-ph	physics	This manuscript was published in: Nature Materials, volume 17, pages 681 -- 685 (2018) Link to published article: https://www.nature.com/articles/s41563-018-0110-9 DOI: 10.1038/s41563-018-0110-9 SharedIt link to access a view-only version of the article for free: https://rdcu.be/bhNaG This file also includes the supplementary information that accompanied the manuscript. Title: Monatomic phase change memory Authors: Martin Salinga1,2, Benedikt Kersting1,2, Ider Ronneberger1,2, Vara Prasad Jonnalagadda1, Xuan Thang Vu2, Manuel Le Gallo1, Iason Giannopoulos1, Oana Cojocaru-Mirédin2, Riccardo Mazzarello2, Abu Sebastian1 1 IBM Research-Zurich, CH-8803 Rüschlikon, Switzerland 2 RWTH Aachen University, D-52074 Aachen, Germany Phase Change Memory (PCM) has been developed into a mature technology capable of storing information in a fast and non-volatile way1-3, with potential for neuromorphic computing applications4-6. However, its future impact in electronics depends crucially on how the materials at the core of this technology adapt to the requirements arising from continued scaling towards higher device densities. A common strategy to fine-tune the properties of PCM materials, reaching reasonable thermal stability in optical data storage, relies on mixing precise amounts of different dopants, resulting often in quaternary or even more complicated compounds6-8. Here we show how the simplest material imaginable, a single element (in this case antimony), can become a valid alternative when confined in extremely small volumes. This compositional simplification eliminates problems related to unwanted deviations from the optimized stoichiometry in the switching volume, which become increasingly pressing when devices are aggressively miniaturized9,10. Removing compositional optimization issues may allow capitalizing on nanosize effects in information storage. 1/17 Continued miniaturization of PCM devices is not only required to achieve memory chips with higher data density and neuromorphic hardware capable of processing larger amounts of information, but also to increase the power efficiency per operation. PCM is based on the ability to switch so-called phase change materials between crystalline states with high electrical conductivity and meta-stable amorphous states with low electrical conductivity. The electrical pulses thermally inducing the fast transitions between those states can be of less energy when less material needs to be heated up11,12. In the endeavour to scale further and further one is reaching volumes in which it starts to make a significant difference to the composition in the switchable regime, whether there are a few more atoms of the right kind in the ensemble or not. And even if we manage to achieve the precise stoichiometry in each of the billions of PCM elements across a silicon wafer, it is known that under the regularly applied strong electric fields (> 0.1 V/nm) and high temperatures (> 1000 K) the different elements move in and out of the region of interest -- by field-assisted motion or simply by phase segregation -- limiting the device cyclability and lifetime9,10. This also results in significant increase in stochasticity associated with the operation of these devices. Another key challenge with miniaturization is that the interface effects become increasingly important. Various studies have shown how the crystallization kinetics of phase change materials, when scaled into few nanometer length scales, can change significantly depending on the material with which it is in contact13-16. Thus, compositions carefully trimmed for showing ideal characteristics in larger volumes must be expected to lose their favourable behaviour in aggressively scaled devices. In this work, we take a drastically different approach. Instead of tweaking the properties of phase change materials by compositional variation, we decide to use the naturally most homogeneous and at the same time simplest material imaginable: a pure element. Amorphous Sb has been found to conduct electrical charge in a semiconducting way with orders of magnitude higher electrical resistivity than crystalline antimony17 -- a contrast even sufficient for assigning multiple resistance levels per memory cell18. However, to-date amorphous Sb could only be created by careful deposition of very thin films or 2/17 at reduced temperatures17,19. The creation of a glass by quenching from the melt, the essential process in switching a PCM, has never been accomplished for pure Sb due to its extreme proneness to crystallization. Recently, however, great progress has been made in forming glasses even out of materials that for the longest time had been deemed impossible to stabilize in a glassy state: most notably, small volumes of monatomic metallic glasses were formed by attaining ultrafast cooling from the melt20-22. Building on those findings, in our present study, we first used ab-initio molecular dynamics (AIMD) simulations to study the effect of quenching molten antimony down to room temperature at different cooling rates (more information in methods section). The results show that the stability of Sb against crystallization at room temperature is significantly dependent on the rate at which it was cooled from the melt (Fig. 1). The demonstrated qualitative trend can be expected to hold for much longer times even for more accessible quenching rates. (The transferability of those results to timescales of more practical interest is discussed in the methods section.) Therefore, our AIMD simulations can be taken as encouragement in the sense that if we manage to quench antimony rapidly enough, we can succeed in creating a glass of pure antimony with sufficient stability against crystallization even around room temperature. Accordingly, we performed experiments using electrically contacted structures of nanometric dimensions (depicted in Fig. 2a) in order to be able to effectively dissipate the heat from the small volume of molten Sb through its interfaces into the surrounding matter (the device fabrication is described in the methods section). Trapezoidal voltage pulses were applied to the devices at various base temperatures (see section on electrical testing at cryogenic temperatures in the methods section for details). Control of the trailing edges of such pulses in the range of few nanoseconds proved to be instrumental (see supplementary information section 'Critical cooling rate for glass formation'). This 3/17 way, we were able to reliably melt-quench the antimony into a semiconducting state reflected by an increase of the device resistance of more than 2.5 orders of magnitude compared to the fully crystalline state (Fig. 2b), in line with what has been reported in literature for thin films of amorphous Sb deposited on a cold substrate17. This contrast in resistance is not only crucial for applications in the field of electronic information processing such as those mentioned above, in a fundamental study like ours it allows for immediate insights into how successful an amorphization attempt has been. The melt- quenched amorphous state can, for example, be observed to exhibit a characteristic temperature dependence of its electrical transport (Fig. 2b). Another attribute of amorphous phase change materials is a temporal increase in electrical resistivity ascribed to structural relaxation of the material towards an energetically more favourable ideal glass state23,24. At constant ambient temperature the resistance typically exhibits a temporal dependence characterized by 𝑅(𝑡)=𝑅(𝑡&)∙())+, where 𝑅(𝑡&) is the resistance measured at time 𝑡&. We measured drift coefficients 𝜈 for the melt-quenched amorphous Sb (Fig. 2c) in the range of 0.10 ± 0.02, which is remarkably similar to that reported for conventional, multi- elemental phase change materials such as Ge2Sb2Te5 25. Also, we observe the typical threshold switching phenomenon when measuring the dynamic current response to a voltage sweep (see supplementary information section 'Observation of threshold switching'). Following this examination of evidence of the realization of amorphous Sb, we now turn towards the influence of quench rates through a systematic modification of the trailing edges of the applied electrical pulses (Fig. 2d). At a given peak-power of a trapezoidal voltage pulse applied to a fully crystalline device under test, the resulting high device resistance decreases quite linearly with increasing trailing edge duration of the pulse. While the peak-power determines how much antimony is molten just before the voltage is decreased, the trailing edge controls how far the crystal structure can grow back into the previously molten volume. The more time is spent at elevated temperatures where crystal growth is extremely fast, the smaller is the volume of Sb that survives in a disordered state when ambient temperature is reached again. A trailing edge of only 10 nanoseconds is enough to form a fully 4/17 crystalline conduction path through the device even in the most ideal of all investigated conditions (lowest ambient temperature). With lower programming powers and thus smaller molten volumes to start the quenching process from, the trailing edges must be even sharper to leave an amorphous plug that can significantly obstruct the charge transport through the device. Further reduction of the programming power must ultimately result in a situation where the molten material does not even cover the whole cross-section of the device, and hence a glass that completely blocks the conduction path cannot be created. These scenarios define the borders of the amorphization window, i.e. the range of programming powers and trailing edges allowing a successful creation of an amorphous mark that is detectable as a pronounced elevation of the device resistance above its crystalline level (Fig. 3 and supplementary information section 'Definition of an amorphization window'). Note that, even if the heating from the electrical pulse could be turned off infinitely quickly, the dissipation of heat from the molten antimony into its surrounding keeps quenching rates finite. In our test structures the thermal boundary conditions of the antimony are strongly dependent on the thickness of the underlying SiO2 acting as a heat barrier towards the vast thermal bath of the silicon wafer. With a 200 nm thin sheet of SiO2 electrically insulating the antimony from the silicon substrate it is barely possible to amorphize any fraction of the Sb, even at an ambient temperature of only 100 K (Fig. 3a). Reducing the thickness of this dielectric layer from 200 nm to 100 nm and further to 40 nm results in substantially faster cooling rates. At the same time, the maximum temperatures reached across the antimony line are reduced by the more effective heat flux into the substrate and so is the size of the molten volume. This reduction, however, is clearly more than counterbalanced by a crucial shortening of the 'thermal trailing edge', which gives room for longer electrical trailing edges (Fig. 3a). After having enlarged the amorphization window by a modification of the thermal environment of the antimony, the devices with a 40 nm thin SiO2 heat barrier are suited to experiments at higher ambient 5/17 temperatures. A higher base temperature has two major impacts on the ability to amorphize, similar to the influence of the heat barrier: on the one hand the same electrical heating power will melt a larger volume of antimony; on the other hand the process of cooling from the melt will be slowed down due to the smaller temperature gradients resulting in a longer time spent in the temperature regime of fast crystallization. Again, as might be expected for a material with a very strong tendency to rapidly crystallize, expanded recrystallization dominates over a somewhat larger molten volume to begin the quenching with. Overall, the amorphization window shrinks for higher ambient temperature, but it remains clearly open up to 250 K (Fig. 3b). Furthermore, in the measurement series performed at 100 K, 150 K, 200 K and 250 K there is no indication for a sudden end of the feasibility to amorphize Sb even above room temperature. However, when turning towards higher ambient temperatures, quenching the melt rapidly enough to pass through the high-temperature regime without substantial crystallization is only one of the challenges posed. At higher base temperatures the stability of a successfully created amorphous state weakens considerably. In order to quantify this stability, we observe how the device resistance evolves and determine the time that has passed since the electrical melting pulse until the device resistance is only a factor of two above its crystalline value. This crystallization time appears to depend on temperature in an Arrhenius way, at least over the four orders of magnitude in time covered by our experiments (Fig. 4 and supplementary information section 'Arrhenius behaviour of crystallization time'). For the devices with 5 nm thick antimony investigated until now, the 'retention time' above room temperature is in the range of few seconds and below, still far from the stabilities one is used to from established phase change materials. However, narrowing the confinement of a glass between interfaces with neighbouring materials can be an effective way to stabilize it by restricting its structural dynamics26-29. Therefore, we investigated how the robustness of antimony against crystallization can be influenced by scaling down its thickness. Indeed, the stability is boosted by more than 100 K in base 6/17 temperature or by many orders of magnitude in time respectively, when reducing the thickness of Sb from 10 nm to 3 nm (Fig. 4). The retention time we have achieved already meets the requirements for some computing-in-memory tasks30 and is orders of magnitude higher than that of dynamic random access memory (see supplementary information section 'Retention times in different memory applications'). For enhanced stability of the amorphous state, besides a further well-controlled reduction of the thickness, systematic investigations of alternative neighbouring materials with particular focus on their atomic-scale roughness31 and rigidity will be instrumental in enhancing amorphous stability. Ultimately, when moving towards next-generation memory devices with three-dimensional nanoscale confinement, also the effects of mechanical stress must be expected to play an important role for the stability of the amorphous states. Quantitative analyses in theory, in simulation and in experiments studying these dependencies under careful control of all mentioned aspects are lacking. Our work demonstrates what impact such research could have for realizing reliable phase-change based devices integrated with highest spatial density. With this work we advocate a paradigm shift for the research on phase change materials for information processing. Instead of following the wisdom of the past and proposing ever new mixtures with questionable chances of being achievable and maintainable in ultra-scaled structures, we turn towards the most radical simplification on the material side. As a consequence, in this context, discussions about how a particular composition might be necessary for achieving improved phase-change functionality become obsolete. In contrast, quantitative knowledge of effects related to nanoscale confinement emerges as a matter of highest importance. 7/17 FIGURES FIG. 1: Increase of quenching rate hinders crystallization. An Ab-initio Molecular Dynamics model of 360 Sb atoms is quenched from the melt with varied rates (3 K/ps purple, 9.5 K/ps blue, 30 K/ps green, 300 K/ps orange, abrupt quenching in a single simulation step red) reaching its target temperature of 300 K at time t=0 ps (see inset). A drop in potential energy (top panel) below around -150.5 eV per atom accompanied by an increase of the number of crystalline-like particles Nc (bottom panel) marks crystallization. Details about the definition of Nc is given in the supplementary information section 'AIMD simulations of quenching rate variations'. The simulation with the slowest quenching rate (purple) was aborted before reaching the target temperature as crystallization started already during the quenching period. 8/17 FIG. 2: Creation of melt-quenched amorphous Sb in electrical switching experiments. (a) Device geometry with dimensions given in nm. (b) Device resistance in a fully crystalline device (blue x) and directly after successful amorphization at ambient temperature of 100 K with electrical pulse power of 957 µW (red x). Resistance series (red +) taken during stepwise temperature decrease after annealing at 225 K reveals thermal activation of the electrical transport with an apparent activation energy of 0.065 eV between 225 K and 175 K (dotted black line). (c) Temporal increase of device resistance after melt-quenching at 100 K characteristic for amorphous phase change materials. Dashed line is a fit of the power law conventionally used to describe such a resistance drift. (d) Influence of trailing edge of applied electrical pulses on resulting device resistance at a base temperature of 100 K. Different shades of red represent pulse powers between 779 and 1051 µW. Variation of pulse parameters are illustrated in the inset to panel (d). The error bars denote the standard deviation determined from five identical electrical excitations. Each amorphization (b-d) was induced by a trapezoidal voltage pulse with 50 ns plateau duration. The trailing edge in (b) and (c) was 3 ns. 9/17 FIG. 3: Controlling the amorphization window. (a) Increasing heat dissipation into the thermal environment of the PCM device by thinning the heat barrier enlarges the amorphization window enough to keep it open even when (b) the ambient temperature is increased towards room temperature. Each data point, in (a) and (b), is based on a series of trailing edge experiments as plotted in Fig. 2d. Filled symbols represent the longest trailing edge at a certain pulse power that still results in a device resistance higher than the crystalline state (forming an upper bound of the amorphization window); empty symbols mark the minimum programming power below which no significant resistance increase was achievable with a certain trailing edge (forming the left border of the amorphization window). More accurately, the minimum programming power lies between the highest programming power that did not yet increase the device resistance above its crystalline starting point on the one hand side and the lowest programming power that did significantly raise the device resistance on the other hand side. Those empty symbols are positioned in the middle between those two power values. The error bars reach from the first to the latter power value. The base temperature in (a) is 100 K. The heat barrier thickness in (b) is 40 nm. The thickness of Sb is 5 nm in (a) and (b). Because the amorphization window does not change considerably going from 100 K to 150 K (b), it is coloured only once (purple). The successfully created amorphous states are stable for much longer than the few seconds it takes to measure their resistance after the melting pulse. 10/17 FIG. 4: Improving robustness against crystallization by narrowing the confinement of the elemental glass. While reducing the device thickness from 10 nm (dark green asterisk) to 5 nm (full green x) the increase in absolute crystallization times is already very strong, further confinement from 5 nm to 3 nm (light green +) even results in a steeper temperature dependence of the crystallization time. To obtain an average value for the crystallization time at a given temperature, devices were melt-quenched (and recrystallized) at least five times. The error bars correspond to the standard deviation of log10(crystallization time). 11/17 References Intel Corporation. Intel® Optane™ Technology. Available at: https://www.intel.com/content/www/us/en/architecture-and-technology/intel-optane- technology.html. (Accessed: 16 August 2017) Choe, J. Intel 3D XPoint Memory Die Removed from Intel OptaneTM PCM (Phase Change Memory). http://www.techinsights.com/ (2017). Available at: http://www.techinsights.com/about-techinsights/overview/blog/intel-3D-xpoint-memory-die- removed-from-intel-optane-pcm/. (Accessed: 16 August 2017) Burr, G. W. et al. Recent Progress in Phase-Change Memory Technology. IEEE Journal on Emerging and Selected Topics in Circuits and Systems 6, 146 -- 162 (2016) Burr, G. W. et al. Neuromorphic computing using non-volatile memory. Advances in Physics: X 2, 89 -- 124 (2016). Burr, G. W. et al. Experimental Demonstration and Tolerancing of a Large-Scale Neural Network (165 000 Synapses) Using Phase-Change Memory as the Synaptic Weight Element. IEEE Trans. Electron Devices 62, 3498 -- 3507 (2015). Tuma, T., Pantazi, A., Le Gallo, M., Sebastian, A. & Eleftheriou, E. Stochastic phase-change neurons. Nature Nanotechnology 11, 1 -- 8 (2016). 7. Wuttig, M. & Yamada, N. Phase-change materials for rewriteable data storage. Nature Materials 6, 824 -- 832 (2007). 8. Matsunaga, T. et al. From local structure to nanosecond recrystallization dynamics in 1. 2. 3. 4. 5. 6. 9. 10. 11. 12. 13. 14. 15. 16. 19. 20. AgInSbTe phase-change materials. Nature Materials 10, 129 -- 134 (2011). Debunne, A. et al. Evidence of Crystallization -- Induced Segregation in the Phase Change Material Te-Rich GST. J. Electrochem. Soc. 158, H965 -- H972 (2011). Xie, Y. et al. Self-Healing of a Confined Phase Change Memory Device with a Metallic Surfactant Layer. Adv. Mater. 30, 1705587 (2018). Xiong, F., Liao, A. D., Estrada, D. & Pop, E. Low-Power Switching of Phase-Change Materials with Carbon Nanotube Electrodes. Science 332, 568 -- 570 (2011). Salinga, M. & Wuttig, M. Phase-Change Memories on a Diet. Science 332, 543 -- 544 (2011). Raoux, S., Jordan-Sweet, J. L. & Kellock, A. J. Crystallization properties of ultrathin phase change films. J. Appl. Phys. 103, (2008). Raoux, S., Cheng, H.-Y., Jordan-Sweet, J. L., Munoz, B. & Hitzbleck, M. Influence of interfaces and doping on the crystallization temperature of Ge-Sb. Appl. Phys. Lett. 94, (2009). Simpson, R. E. et al. Toward the Ultimate Limit of Phase Change in Ge2Sb2Te5. Nano Lett. 10, 414 -- 419 (2010). Chen, B., Brink, ten, G. H., Palasantzas, G. & Kooi, B. J. Size-dependent and tunable crystallization of GeSbTe phase-change nanoparticles. Scientific Reports 6, 265 (2016). 17. Hauser, J. J. Hopping conductivity in amorphous antimony. Phys. Rev. B 9, 2623 -- 2626 (1974). 18. Koelmans, W. W. et al. Projected phase-change memory devices. Nature Communications 6, 8181 (2015). Krebs, D. et al. Threshold field of phase change memory materials measured using phase change bridge devices. Appl. Phys. Lett. 95, (2009). Zhong, L., Wang, J., Sheng, H., Zhang, Z. & Mao, S. X. Formation of monatomic metallic glasses through ultrafast liquid quenching. Nature 512, 177 -- + (2014). Schroers, J. Glasses made from pure metals. Nature 512, 142 -- 143 (2014). 21. 22. Greer, A. L. New horizons for glass formation and stability. Nature Materials 14, 542 -- 546 (2015). 23. Raty, J. Y. et al. Aging mechanisms in amorphous phase-change materials. Nature Communications 6, 1 -- 8 (2015). 12/17 24. 25. Zipoli, F., Krebs, D. & Curioni, A. Structural origin of resistance drift in amorphous GeTe. Phys. Rev. B 93, (2016). Boniardi, M. et al. Statistics of Resistance Drift Due to Structural Relaxation in Phase-Change Memory Arrays. IEEE Trans. Electron Devices 57, 2690 -- 2696 (2010). 26. Watanabe, K., Kawasaki, T. & Tanaka, H. Structural origin of enhanced slow dynamics near a wall in glass-forming systems. Nature Materials 10, 512 -- 520 (2011). Ellison, C. J. & Torkelson, J. M. The distribution of glass-transition temperatures in nanoscopically confined glass formers. Nature Materials 2, 695 -- 700 (2003). Priestley, R. D., Ellison, C. J., Broadbelt, L. J. & Torkelson, J. M. Structural Relaxation of Polymer Glasses at Surfaces, Interfaces, and In Between. Science 309, 456 -- 459 (2005). Sohn, S. et al. Nanoscale size effects in crystallization of metallic glass nanorods. Nature Communications 6, 1 -- 6 (2015). Sebastian, A. et al. Temporal correlation detection using computational phase-change memory. Nature Communications 1 -- 10 (2017). doi:10.1038/s41467-017-01481-9 Scheidler, P., Kob, W. & Binder, K. The Relaxation Dynamics of a Supercooled Liquid Confined by Rough Walls †. J. Phys. Chem. B 108, 6673 -- 6686 (2004). 27. 28. 29. 30. 31. 13/17 Supplementary Information is available in the online version of the paper. Acknowledgements The research leading to these results has received funding from the People Programme (Marie Curie Actions) of the European Union's Seventh Framework Programme FP7/2007- 2013/ under REA grant agreement No 610781, from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement Numbers 640003 and 682675), and from Deutsche Forschungsgemeinschaft (DFG) through the collaborative research centre Nanoswitches (SFB 917). We also acknowledge the computational resources provided by JARA-HPC from RWTH Aachen University under Projects No. JARA0150 and JARA0176. Finally, we thank Evangelos Eleftheriou and Wabe W. Koelmans at IBM Research Zurich for his support of this work. Author Contributions B.K., A.S. and M.S. conceived and designed the experiments; V.P.J. and X.T.V. fabricated the Sb-based devices with support from I.G.; O.C.-M. analysed the integrity of the deposited Sb via Atom-Probe Tomography; B.K. performed the experiments supported by M.L. and advised by M.S. and A.S.; B.K. and M.S. analysed the data; I.R. performed the computer simulations and analysed the data, with help from R.M.; M.S. wrote the manuscript with input from all the authors. Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Readers are welcome to comment on the online version of the paper. Correspondence and requests for materials should be addressed to M.S. ([email protected]) and A.S. ([email protected]). 14/17 Methods AIMD: Different from a very recent publication simulating supercooled liquid Sb32, for our Ab initio Molecular Dynamics (AIMD) simulations based on Density Functional Theory (DFT) we employed the second generation Car-Parrinello scheme33 which is implemented in the Quickstep code of the CP2K simulation package34 (a more detailed discussion of our AIMD simulations is given in the supplementary information sections 'AIMD methodology' and 'AIMD simulations of quenching rate variations'). Effects due to the finite size of the used simulation box (360 atoms) can lead to a significantly reduced crystallization time (see the effect of increasing the system size to 540 and even 720 atoms in the supplementary information section 'Finite size effects in AIMD simulations', in particular Fig. S4). Also, the speed of the crystallization process is very sensitive to the atomic density, which is quite difficult to determine accurately for experimentally realistic volumes. As an example: allowing the Sb atoms to fill a volume that is enlarged by only 7 % reduces the stress by a factor of 5, leading to a 10 times higher stability against crystallization at 450 K (a detailed discussion is provided in the supplementary information section 'AIMD simulations with different densities'). Device fabrication: The devices were fabricated on a silicon substrate with a thermally grown SiO2 top layer of varying thickness (40, 100, 200 nm) acting as thermal and electrical insulation. The Sb with a thickness of 3, 5 or 10 nm and a 5 nm thick capping layer of (ZnS)80(SiO2)20 were deposited by sputtering with an average rate of <0.1 nm/s. The purity of the deposited Sb was assured to be >99.9 % using Atom-Probe Tomography. The lateral device shape depicted in Fig. 2a was then realized using using e-beam lithography with hydrogen silsesquioxane (HSQ) resist and ion milling. The patterned structure was immediately passivated with additional ~18 nm of sputtered SiO2. In order to electrically contact the Sb layer, holes were ion-milled into the capping layers in another e-beam lithography step, ensuring that 15/17 the etching process is stopped once it reached the SiO2 heat barrier underneath the Sb. Then, a third e- beam lithography step with lift-off was used to shape a sputter-deposited layer of tungsten into lateral electrical leads connected to the phase change material. Also, for better switching stability, a titanium resistor (ranging between 2 to 4 kOhm) was added in series with the Sb device using lift-off. The whole chip was then encapsulated with a layer of 80-nm-thick sputtered SiO2. Finally, e-beam lithography and reactive-ion etching were employed to locally open the encapsulation before adding gold probe pads (200 nm, sputter-deposited, and shaped via optical lithography and lift-off). Electrical testing at cryogenic temperatures: The electrical measurements were performed in a liquid-nitrogen-cooled cryogenic probing station (JANIS ST-500-2-UHT) operating between 77 to 400 K. The temperature was controlled using two heaters with powers of 50 and 25 W, calibrated Lakeshore Si DT-670B-CU-HT diodes with an accuracy of <0.5 K at four positions in the chamber and a Lakeshore 336 Automatic Temperature Controller. A radiation shield is fixed above the sample mount and thermally connected to the nitrogen out flux isolates. The pressure inside the chamber is reduced below 1023 mbar to avoid heat exchange via convection and water condensation at low temperatures. A high- frequency Cascade Microtech Dual-Z probe is used to contact the devices inside the cryogenic probe station. It is thermally connected via cooling braids to the bulk metallic sample holder keeping sample and probe at the same temperature. A Keithley 2400 Source Measure Unit was used for DC measurements of the device resistance (at a constant voltage of 0.1 V). An Agilent 81150A Pulse Function Arbitrary Generator sent the voltage pulses to the device under test and a Tektronix oscilloscope (TDS3054B/DPO5104B) recorded applied voltage and transmitted current signals. Mechanical relays (OMRON G6Z-1F-A) were used to switch between AC and DC configuration. All switching voltage pulses had a trapezoidal shape with a plateau length of 50 ns and equal leading and trailing edge. The pulse power experienced by the Sb device is calculated based on the plateau values of the time-resolved current- and voltage traces recorded by the oscilloscope 16/17 under consideration of the series resistor. In order to exclude gradual, irreversible changes in the tested device being responsible for the observed effect of the varying trailing edges (Fig. 2d), the first measurement series (with the shortest trailing edge) was always reproduced at the end of all other experiments with longer trailing edges. Data availability: The data that support the findings of this study are available from the corresponding author upon reasonable request. References cited only in the Methods section: 32. 33. 34. Ropo, M., Akola, J. & Jones, R. O. Crystallization of supercooled liquid antimony: A density functional study. Phys. Rev. B 96, 161 -- 8 (2017). Kühne, T. D., Krack, M., Mohamed, F. R. & Parrinello, M. Efficient and Accurate Car-Parrinello-like Approach to Born-Oppenheimer Molecular Dynamics. Phys. Rev. Lett. 98, 066401 (2007). VandeVondele, J. et al. Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach. Computer Physics Communications 167, 103 -- 128 (2005). 17/17 Supplementary information Title of the main article: Monatomic phase change memory Authors: Martin Salinga*1,2, Benedikt Kersting1,2, Ider Ronneberger1,2, Vara Prasad Jonnalagadda1, Xuan Thang Vu2, Manuel Le Gallo1, Iason Giannopoulos1, Oana Cojocaru-Mirédin2, Riccardo Mazzarello2, Abu Sebastian1 1 IBM Research-Zurich, CH-8803 Rüschlikon, Switzerland 2 RWTH Aachen University, D-52074 Aachen, Germany Contents: AIMD methodology .................................................................................................................................. 2 AIMD simulations of quenching rate variations ........................................................................................ 2 AIMD simulations with different densities ............................................................................................... 4 Finite size effects in AIMD simulations ..................................................................................................... 5 Simulation of Sb/SiO2 interface at high temperatures ............................................................................. 6 Observation of threshold switching ......................................................................................................... 6 Arrhenius behaviour of crystallization time .............................................................................................. 7 Definition of an amorphization window ................................................................................................... 7 Critical cooling rate for glass formation .................................................................................................... 8 Retention times in different memory applications ................................................................................... 9 References ............................................................................................................................................. 11 1/12 Supplementary information AIMD methodology For our Ab initio Molecular Dynamics (AIMD) simulations based on Density Functional Theory (DFT) we employed the second generation Car-Parrinello scheme1 which is implemented in the Quickstep code of the CP2K simulation package2. In this scheme the efficiency of the Car-Parrinello method is combined with the large time steps used in Born-Oppenheimer Molecular Dynamics (BOMD). Instead of solving the Kohn-Sham equations self-consistently at each molecular dynamics step or using the Car-Parrinello equations of motion, a predictor-corrector algorithm, more specifically the always stable predictor-corrector method by Kolafa3, is used to propagate the expansion coefficients of the wave function basis set. Using this scheme it is possible to maintain the system close to the Born- Oppenheimer surface without the need of using small time steps. As a consequence, a computational speedup of approximately a factor 20 to 30 with respect to the standard (BOMD) are typically achieved. However, the dynamics of this method is dissipative with some intrinsic friction coefficient of the system and thus cannot be used in the microcanonical ensemble (NVE). However, it can be used in the canonical ensemble (NVT). The implementation in CP2K utilizes a stochastic Langevin thermostat to control the intrinsic dissipation. This method has been proven to be to provide reliable results for many systems, for example in liquid H2O, Si, l-SiO21,4 and in particular it was successfully used for simulations of typical phase-change materials GeTe5, Ge2Sb2Te56, Ag4In3Sb67Te267. In the present simulation, we use this scheme for two types of systems, namely pure Sb and an interfacial model of Sb/SiO2. For the latter system, it was necessary to increase the number of predictor- corrector steps to properly thermalize the two subsystems (Sb and SiO2). Generalized gradient approximation (GGA) to the exchange-correlation function8 and scalar-relativistic Goedecker pseudopotentials9 (PP) are used for all the simulations. Triplezeta plus polarization Gaussian-type basis set are employed for the expansion of the Kohn-Sham orbitals and the charge density is expanded in plane waves with a cut-off of 300 Ry. Periodic boundary conditions are applied and the Brillouin zone is sampled at the Γ of the supercell. AIMD simulations of quenching rate variations For the generation of stable supercooled liquid and amorphous states of phase-change materials very large quenching rates are required to avoid crystallization. Due to limitations of the computational time scales, the quenching rate employed in simulations of typical phase-change materials6,7,10-12 are on the order of 1013 K/s which is around two orders of magnitude above typical experimentally realized values. Nevertheless, a variation of the quenching rates in the simulations on a computationally affordable time scale might give qualitative statements on the stability of supercooled and amorphous phase-change materials. In the present set of simulations, we performed AIMD simulations of pure Sb by changing the quenching rate γ over two orders of magnitude. For that we maximum quenching rate of "∞ K/ps" which is obtained by directly setting the target temperature from ~930 K to ~300 K (see Fig. S1). Our stochastic Langevin thermostat achieves the temperature within one picosecond which corresponds to ~630 K/ps. The quenching rates are realized by changing the considered a model containing 360 atoms in an orthorhombic supercell of 21.77 x 22.62 x 22.78 Å3 which corresponds to the density of liquid Sb and is equal to the experimental value of 6.49 g/cm3 (Ref. 13) at T=915 K. Five different simulations with quenching rates g = 3, 9.5, 30, 300 K/ps and an idealized target temperature in a stepwise fashion, e.g. g = 30 K/ps was achieved by decreasing the temperature by 15 K every 0.5 ps. After the quenching, the models were annealed at room temperature. 2/12 Supplementary information Fig. S1: The temperature profile of the models with different quenching rates, ranging γ = 3 K/ps to ∞ K/ps. The inset shows the zoomed-in region of fast quenching which is indicated by a rectangle. The results of this set of simulations are summarized in Fig. 1 of the main text. It shows the time evolution of the potential energy (dark colours) and the number of crystalline-like atoms NC (light colours) for all the models. The NC is determined based on a bond order correlation parameter, a definition of which is given in Ref. 14, Chapter 3, pages 60-63. We distinguish crystalline-like atoms by computing the bond order correlation parameter 𝑞&'()(𝑖) for each atom i and using a threshold value of 0.65. Liquid-like and of the 𝑞&'() parameter for different phases of Sb are shown in Fig. S2. The distributions in the liquid and crystalline-like if its 𝑞&'() exceeds that threshold. amorphous phases show negligibly small overlap with the distributions of perfectly crystalline and recrystallized phase at values close 0.65. We choose this value as a threshold and define an atom to be amorphous-like atoms display values close to 0, because the bond orientations between neighbouring atoms are uncorrelated, whereas crystalline-like atoms take values close to 1. The statistical distribution Fig. S2: The distributions of 𝑞&'() for liquid (red), amorphous (green), crystalline (blue) and recrystallized (purple) phases of Sb. For the corresponding computation of liquid, amorphous and recrystallized phases, segments from the trajectories of 360-atom models in the corresponding phases were taken. For the computation of crystalline phase, a short run of a perfectly crystalline model at 600 K was performed. Since the distribution in the crystalline phases are narrow, a second axis (indicated by the blue arrow) with a larger scale is used for the crystalline and recrystallized distributions. The threshold value used to define crystalline-like atoms is indicated as a dashed vertical line. 3/12 Supplementary information (360 atoms) were annealed at different temperatures ranging from 700 K to 300 K. Detailed discussion of the results for that set of simulations can be found in Ref. 14 (Chapter 7). All the models in the temperature range of 400-550 K crystallized very quickly whereas in the temperature range above The remaining four models of our present study reached room temperature without crystallization. The The model with the slowest quench rate of g = 3 K/ps crystallizes already during the quenching process. This result is consistent with the observation of very short crystallization times in a temperature range around ~500 to 600 K. Prior to the simulations discussed here, various models of supercooled Sb ~600 K and below ~300 K no fast crystallization occurred. model with g = 9.5 K/ps contains a small fraction of crystalline-like atoms (NC~10) at the time when the ~300 ps in both models. However, the subsequent crystallization of the faster quenched model (with after ~800 ps. Nonetheless, the number of crystalline-like atoms NC in this model grows slowly over target temperature was set to 300 K (t = 0) which steadily increases until the onset of crystallization. NC increases more slowly in the models with g = 30 K/ps and g = 300 K/ps. The onset of crystallization is at g = 300 K/ps) occurs notably slower resulting in a longer crystallization time. The model with the highest quenching rate g = ∞ K/ps. sets itself apart from all the other models in that it does not crystallize even time suggesting that it will eventually crystallize once it reaches a critical size. In our models we observe a consistent trend of increased stability of amorphous Sb with faster quenching rates. This result suggests that qualitatively employing higher quenching rates can lead to a more stable amorphous phase of Sb. Nevertheless, one should keep in mind that the crystallization is a stochastic process with varying crystallization times at a given temperature. A more quantitative analysis would require a larger number of independent simulations to improve the statistics of the crystallization times. This is, however, currently neither feasible with AIMD nor needed for the present study due to other limitations of the model (see sections below). AIMD simulations with different densities A parameter that is known to have an influence on the stability of the amorphous state is the density. In the current study we used the liquid density of rhigh = 6.49 g/cm3. Here, we address the question on how the density affects the stability of the supercooled and amorphous phase by generating an additional model with 7 % lower density of rlow = 6.03 g/cm3 by quenching with g = 30 K/ps. We chose a model with rhigh from the previous simulations at 500 K which falls into the range of fast crystallization and performed the same annealing run for the new model with rlow at that temperature. The results of these two models are compared in Fig. S3. As can be seen from both the potential energy and the number of crystalline-like atoms NC, the onset of crystallization for the model at lower density occurs at a significantly later point in time. For both models temperature and virial stress values were obtained averaging values in the first 20 ps of the simulation. For the virial stress, the average of the diagonal components 〈𝜎//〉=2𝜎33+𝜎55+𝜎667/ 3 resulted in 〈𝜎//〉=0.28±0.19 𝐺𝑃𝑎 at a temperature of 𝑇=503±23 𝐾 for the rlow model and 〈𝜎//〉=1.40±0.18 𝐺𝑃𝑎 at 𝑇=509±20 𝐾 for the rhigh model. While the rlow model is almost stress- free displaying values close to zero in the stress tensor, the average of the virial stress components of the high density model is five times larger than that of the low density model. This result suggests that supercooled and amorphous phase of Sb are destabilized by application of compressive stress. A similar finding was provided in Ref. 15 in which the stability of amorphous Ge-doped Sb (i.e. Ge7Sb93 and Ge6Sb94), closely related phase-change compounds, was shown to decrease experimentally upon compressive stress. 4/12 Supplementary information Fig. S3: The potential energy U (dark) and the number of crystalline-like particles NC (light) of two supercooled models at T=500 K with two different densities: rhigh = 6.49 g/cm3 (red) and rlow = 6.03 g/cm3 (blue). Both models are obtained with the same quenching rate of g = 30 K/ps. Finite size effects in AIMD simulations Finite size effects are inevitable in the typical system sizes employed in AIMD simulations. Because of the periodic boundary conditions atoms interact with their periodic images in a spurious way. In Ref. 14, we showed that the free energy barrier for nucleation in very small systems (on the order of 50 to 70 atoms) can take extremely small values. Simulations of such small systems can be inflicted with severe artefacts and should be therefore avoided. On the other hand, the system size is the main factor which determines the computational cost in AIMD simulations. Our system size of 360 atoms lies between the typical system sizes of 216 (GeTe) and 460 (Ge2Sb2Te5) atoms used in several AIMD studies 12,14,16 of phase-change materials. Currently the largest model used in AIMD simulations of crystallization contains 900 atoms (Ge2Sb2Te5)14. Classical molecular dynamics simulations are computationally much cheaper than AIMD. Thus finite system size effects can be easily reduced in such simulations by increasing the model size. However, classical simulations require reliable interatomic potentials which must be able to accurately describe chemical bonding in Sb. Currently no such potential for Sb is available. Therefore, we restrict ourselves to AIMD simulations of models with affordable sizes. We qualitatively assess the system size effects by employing two larger models of Sb (rhigh = 6.49 g/cm3) containing 540 and 720 atoms inside a cubic cell with dimensions 25.62 x 25.62 x 25.62 Å3 and 28.20 x 28.20 x 28.20 Å3, respectively. Supercooled models for these system sizes were obtained by quenching them with g = 30 K/ps. Each of these three models was annealed at T = 450 K resulting in the data plotted in Fig. S4. The crystallization time increases with model size. This is consistent with our argument about the reduction of the nucleation barrier with decreasing model sizes. Despite the difference in the nucleation barriers we expect qualitatively similar behaviour for the larger models regarding quenching rate and density effects, albeit on longer times scales. We would like to stress that the observation of crystallization itself in the larger models are very important, since this shows that fast crystallization in the smaller models are not merely artefacts of the model size. 5/12 Supplementary information Fig. S4: The potential energy U (dark) and the number of crystalline-like particles Nc (light) of supercooled Sb models at T=450 K with different system sizes of 360 (red), 540 (blue) and 720 atoms (green). All the models are obtained with the same quenching rate of g = 30 K/ps. Simulation of Sb/SiO2 interface at high temperatures A question relevant for the interpretation of our experimental data is whether we must expect mixing of the confining dielectric material with the phase-change material (Sb). The likelihood for such a mixing would certainly be highest during the most intense reset operation of the device involving melting of the Sb. We addressed this by running simulations of an interface model at high temperatures well above the melting temperature of Sb (Tm,Sb = 903 K). For this purpose, we constructed an interface model which is composed of a 360- atom model of liquid Sb and a separately generated model of amorphous SiO2 containing 288 atoms. Two different models with two different densities were considered. Four different simulation temperatures were chosen, namely, 1000 K, 2000 K, 3000 K and 4000 K. We find no mixing of atoms at 1000 K and 2000 K in the respective simulation times ranging between 35 ps to 55 ps. Only at temperatures above the melting point of SiO2, Tm,SiO2 ≈ 2000 K, (i.e. at 3000 K and 4000 K) the different atomic species start to mix with each other. At temperatures well above Tm,SiO2, the Si and O atoms show very high atomic mobility (diffusion) indicative of SiO2 becoming liquid. Based on these results we can exclude atomic mixing of Sb with the neighbouring dielectric material for temperatures up to 2000 K. Observation of threshold switching Besides the temperature dependence of the resistance (Fig. 2b) and the temporal evolution of the resistance (Fig. 2c) there is another characteristic attribute of amorphous phase change materials observable in our melt-quenched Sb devices: The sudden breakdown of resistance when the applied voltage reaches certain threshold (see exemplarily Fig. S5). Also the linear dependence between threshold voltage and device resistance is agreement with what has been observed for traditional phase change memory devices in the past. The device resistance can approximately be assumed to be a measure for the size of the amorphous mark along the length of the Sb line. 6/12 Supplementary information Fig. S5: Current response to voltage applied to an amorphized Sb device shows threshold switching. Device geometry: 5 nm thick Sb on 40 nm thick SiO2 heat barrier; ambient temperature: 100 K; electrical pulse inducing amorphization: 50 ns plateau, 3 ns leading and trailing edge, 1051 µW pulse power; voltage sweep testing the threshold behaviour: triangular pulse with 200 ns leading and trailing edge. The measurement was repeated five times. (a) Each of the resulting current-voltage characteristics is plotted demonstrating remarkable reproducibility. (b) Threshold voltage (defined as voltage where the current response surpasses 30 µA) as a function of device resistance. The colour code is according to the one given in Fig. 2d of the main text, i.e. representing various pulse powers of the amorphization pulse. The error bars denote the standard deviation determined from five identical electrical excitations. Arrhenius behaviour of crystallization time Fig. 4 of the main text shows that the temperature dependence of the crystallization time can be described reasonable well by an Arrhenius behaviour. The according activation energies resulting from the fits represented by the dashed lines in Fig. 4 are 1.09 ± 0.19 eV for 10 nm Sb, 1.03 ± 0.13 eV for 5 nm Sb, and 1.26 ± 0.20 eV for 3 nm Sb. The effect of uncontrolled impurities should be included in the margin of error of those measurement series (in Fig. 4), since experiments on several different devices have contributed to the crystallization data for each thickness. Definition of an amorphization window The concept of an amorphization window in the parameter space of power and trailing edge length of the electrical amorphization pulses is discussed in the main text. The definitions of the limits of such an amorphization window are given in the figure caption of Fig. 3 of the main text. Here (with Fig. S6), an illustration is added as a visual assistance for an easier understanding of how each limiting point in Fig. 3 is derived from a whole series of electrical measurements. 7/12 Supplementary information Fig. S6: Illustration of the derivation of the limits of the amorphization windows from series of amorphization experiments. (a) Device resistance resulting from an electrical excitation with a certain trailing edge length (horizontal axis) and pulse power (shades of red). The error bars denote the standard deviation determined from five identical electrical excitations. Filled triangles represent the longest trailing edge at a certain pulse power that still results in a device resistance higher than the crystalline state. More accurately, the limiting trailing edge value for each pulse power is read off where the extrapolation of the apparently linear decrease of the device resistance with trailing edge length hits the resistance value of the fully crystalline device (practically zero on a linear MΩ scale). These values of a trailing edge length and according pulse power form the upper bound of an amorphization windows in Fig. 3. (b) Device resistance resulting from an electrical excitation with a certain trailing edge length (shades of green) and pulse power (horizontal axis). The error bars denote the standard deviation determined from five identical electrical excitations. Empty triangles mark the minimum programming power below which no significant resistance increase was achievable with a certain trailing edge. More accurately, the minimum programming power lies between the highest programming power that did not yet increase the device resistance above its crystalline starting point on the one hand side and the lowest programming power that does significantly raise the device resistance on the other hand side. Those empty triangles are positioned in the middle between those two power values. The error bars of those data points marking the left border of the amorphization window in Fig. 3 reach from the first to the latter power value. Critical cooling rate for glass formation In the AIMD simulations shown in Fig. 1 crystallization took place during the cooling phase when the quenching rate was set to 3K/ps. Already with a quenching rate of 9.5K/ps there is no sign of crystallization during the cooling phase anymore. As we discuss in the methods section on "AIMD" and in the according paragraphs in the supplement ("AIMD simulations with different densities" and "Finite size effects in AIMD simulations") one has to expect that in reality the timescales for crystallization are significantly longer. That is why we prefer to only take qualitative trends from our simulations. When attempting to determine a critical cooling rate in case of the experiments there is another problem. In a self-heated nanoscale device, temperature across the material under test is very inhomogeneous. Also, the temporal evolution of the temperature during cooling is not simply linear. Thus, there is not one cooling rate the material under test experienced. So even if one determines the 8/12 Supplementary information temporal evolution of the temperature for every part of the molten material by simulating the device structure (e.g. via Finite Elements Method), which is highly unreliable anyways due to the lack of many significant input parameters (like thermal interface resistances) as a function of temperature, one will not be able to condense the outcome into a simple critical cooling rate. That is why we decided to modify the cooling of the material from the melt by using a quantity which we can control very well, i.e. the trailing edge of our electrical excitation. Evidently, it is possible to significantly influence the cooling times through a variation of the trailing edges in the range of few nanoseconds. Therefore, the cooling times themselves must be on a similar time scale. Thus, one can estimate that the temperature in the material was reduced by several hundred Kelvin (from above melting temperature to base temperature) in few nanoseconds. Under these conditions glass formation of pure Sb is apparently possible. Retention times in different memory applications In the application field of Storage Class Memories (the one currently addressed by Intel's and Micron's 3D-Xpoint technology) alone there can be different flavours: from storage-mapped (longer retention, but not as fast) to memory-mapped (faster and higher endurance, but allowed to be much more volatile). P. Cappelletti (Micron) writes in his 2015 IEDM article entitled „Non volatile memory evolution and revolution"17: "While present visibility gives emerging memories little chance to compete with NAND in cost and density, there is definitely a great opportunity for emerging memory to help close the constantly increasing performance gap between DRAM and NAND. However, even that space, often referred to as the Storage Class Memory space, is too wide to be covered by a single technology; it is more practical to consider two flavors of SCMs, associated to two different areas of applications: memory mapped and storage mapped (fig.8). The memory mapped SCMs (…) do not need to be "true" non-volatile (a retention time much longer than DRAM refresh time is sufficient)." In that fig. 8 the non- volatility criterion is specified for memory-mapped SCM as longer than ten times the DRAM refresh interval, which is currently 64 ms. So, the necessary retention time for such a memory application is less than a second. Fig. S7: Extrapolation of crystallization data towards long times: The vertical line marks room temperature (20°C), the horizontal line marks 10 years. Besides the experimental data depicted and fitted in Fig. 4, here we added the outcome of a single measurement at room temperature (50.8 hours) confirming the extrapolation of the data taken at elevated temperatures. 9/12 Supplementary information Beyond Storage Class Memories, there are other currently emerging applications of phase change memory devices. Given the explosive growth in data-centric cognitive computing and the imminent end of CMOS scaling laws, it is becoming increasingly clear that we need to transition to non-von Neumann computing architectures. In-memory computing and brain-inspired neuromorphic computing are two approaches that are being actively researched. In in-memory computing, the physical attributes and dynamics of memory devices are exploited to perform certain computational tasks in place. For example, PCM devices organized in a crossbar array can be used to solve linear equations, performing part of the computation in an iterative solver. In one such mixed-precision approach, depicted in Fig. S8, crossbar arrays of PCM devices are used to perform inexact analog matrix-vector multiplications using Ohm's and Kirchhoff's circuit laws in an iterative Krylov-subspace solver. A high-precision computing unit is used to iteratively improve the solution accuracy until a desired tolerance is reached. Fig. S8: Mixed-precision in-memory computing implementation of a linear equation solver according to ref. 18. The problem is to find the solution x of Ax = b, where A is a NxN matrix and b is a N-dimensional vector. In this application, matrix A is programmed once in the PCM array, and only read operations are performed subsequently in order to perform the matrix-vector multiplications to solve the linear system. Applications such as these require substantially smaller retention times than those needed for conventional memory/storage applications. To show this, we measured the runtime of the linear solver for different matrix sizes, using both an IBM POWER8 Central Processing Unit (CPU) and a NVIDIA P100 Graphical Processing Unit (GPU) as high-precision processing units. The linear solver was run for a tolerance of tol = 10-5. The PCM devices performing the matrix-vector multiplications were simulated with precision comparable to that achieved in the prototype chips used in the experiments presented in 10/12 Supplementary information ref.18. As it can be seen in Supplementary Fig. S9, in the mixed-precision implementations for all matrix sizes the runtime is less than 1 second. Therefore, a minimum retention time for the PCM device of 1 s is sufficient for this particular application. Such retention time would be easily fulfilled by the 3 nm Sb devices investigated in the main manuscript. Fig. S9: Measured runtime of linear solver for the mixed-precision implementation of Fig. S8 as well as CPU-only and GPU-only implementations. In the mixed-precision implementations the runtime is less than 1 second for all matrix sizes investigated. In the case of neuromorphic computing, certain computational tasks are performed using networks of neuronal and synaptic elements based on the computational principles of the brain. Phase-change memory (PCM) devices have been shown to capture both the neuronal and synaptic behavior in such an application 19,20. The requirements for retention for such PCM devices are significantly different from those for conventional memory and storage applications. For example, in a PCM-based neuron, the PCM device is RESET every time the neuron fires. Hence, if we assume a rather conservative mean spiking frequency of 10 Hz, the average retention time required for the PCM device is 0.1 s. This is substantially lower than the retention times exhibited by the 3 nm Sb devices. Even PCM-based synapses do not require a substantially long retention time. For example, an emerging application area is the use of PCM-based synapses for training deep neural networks 21. In such an application, once the training is complete, the synaptic weights are copied and stored elsewhere. For each layer, the forward and backward propagation as well as the synaptic weight update can be performed in less than 1 µs 22. For a network of 10 layers and a training set size of 100000, this translates to just 1 s of training time per epoch. Thus, a typical number of 20 training epochs would fit well within the retention time of the PCM devices presented in our work. References 1. Kühne, T. D., Krack, M., Mohamed, F. R. & Parrinello, M. Efficient and Accurate Car- Parrinello-like Approach to Born-Oppenheimer Molecular Dynamics. Phys. Rev. Lett. 98, 066401 (2007). 11/13 Supplementary information VandeVondele, J. et al. Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach. Computer Physics Communications 167, 103 -- 128 (2005). Kolafa, J. Time-reversible always stable predictor-corrector method for molecular dynamics of polarizable molecules. J. Comput. Chem. 25, 335 -- 342 (2003). Kuehne, T. D., Krack, M. & Parrinello, M. Static and Dynamical Properties of Liquid Water from First Principles by a Novel Car-Parrinello-like Approach. Journal of Chemical Theory and Computation 5, 235 -- 241 (2009). Mazzarello, R., Caravati, S., Angioletti-Uberti, S., Bernasconi, M. & Parrinello, M. Signature of tetrahedral Ge in the Raman spectrum of amorphous phase-change materials. Phys. Rev. Lett. 104, (2010). Ronneberger, I., Zhang, W., Eshet, H. & Mazzarello, R. Crystallization Properties of the Ge2Sb2Te5 Phase-Change Compound from Advanced Simulations. Adv. Funct. Mater. 25, 6407 -- 6413 (2015). Zhang, W. et al. How fragility makes phase-change data storage robust: insights from ab initio simulations. Scientific Reports 4, (2014). Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865 -- 3868 (1996). Goedecker, S., Teter, M. & Hutter, J. Separable dual-space Gaussian pseudopotentials. Phys. Rev. B 54, 1703 -- 1710 (1996). Akola, J. & Jones, R. O. Structural phase transitions on the nanoscale: The crucial pattern in the phase-change materials Ge2Sb2Te5 and GeTe. Phys. Rev. B 76, (2007). Hegedus, J. & Elliott, S. R. Microscopic origin of the fast crystallization ability of Ge-Sb-Te phase-change memory materials. Nature Materials 7, 399 -- 405 (2008). Kalikka, J., Akola, J., Larrucea, J. & Jones, R. O. Nucleus-driven crystallization of amorphous Ge2Sb2Te5: A density functional study. Phys. Rev. B 86, (2012). Crawley, A. F. & Kiff, D. R. Density and Viscosity of Liquid Antimony. Metallurgical Transactions 3, 157 -- & (1972). Ronneberger, I. Computational Study of Crystallization Kinetics of Phase Change Materials. (Doctoral thesis, 2016). doi:10.18154/RWTH-2017-00376 Eising, G., Pauza, A. & Kooi, B. J. Stress-Induced Crystallization of Ge-Doped Sb Phase- Change Thin Films. Crystal Growth & Design 13, 220 -- 225 (2013). Raty, J. Y. et al. Aging mechanisms in amorphous phase-change materials. Nature Communications 6, 1 -- 8 (2015). Cappelletti, P. Non volatile memory evolution and revolution. in 10.1.1 -- 10.1.4 (IEEE, 2015). doi:10.1109/IEDM.2015.7409666 Le Gallo, M. et al. Mixed-precision in-memory computing. Nature Electronics 1, 246 -- 253 (2018). Kuzum, D., Jeyasingh, R. G. D., Lee, B. & Wong, H. S. P. Nanoelectronic Programmable Synapses Based on Phase Change Materials for Brain-Inspired Computing. Nano Lett. 12, 2179 -- 2186 (2012). Tuma, T., Pantazi, A., Le Gallo, M., Sebastian, A. & Eleftheriou, E. Stochastic phase-change neurons. Nature Nanotechnology 11, 1 -- 8 (2016). Burr, G. W. et al. Neuromorphic computing using non-volatile memory. Advances in Physics: X 2, 89 -- 124 (2016). Gokmen, T. & Vlasov, Y. Acceleration of Deep Neural Network Training with Resistive Cross- Point Devices: Design Considerations. Front. Neurosci. 10, 1 -- 13 (2016). 12/12 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.
1805.11664	1	1805	2018-05-14T12:56:03	Additive Manufacturing of Nickel Based Superalloy	[ "physics.app-ph" ]	This paper reviews the use of relatively new manufacturing method called as additive manufacturing, most often mentioned as 3D printing in fabrication of high performance superalloys. The overview of the article describes the structure, property, processing, performance relationship of the fabrication process and the superalloys. The manufacturing methods such as Electron Beam Melting, Laser Beam Melting and Direct Energy Deposition used to fabricate commercially available alloys are explained. The microstructure / grain structure resulting from directional building, complex thermal cycles is discussed. An overview of the properties of the superalloys and their performance as well as applications is presented.	physics.app-ph	physics	Additive Manufacturing of Nickel Based Superalloys Graduate student, Michigan Technological University, Apoorv Kulkarni Houghton 49931USA as called Abstract – This paper reviews the use of relatively new manufacturing method additive manufacturing, most often mentioned as 3D printing in fabrication of high performance superalloys. The overview of the article describes the structure – property – processing – performance relationship of the fabrication process and the superalloys. The manufacturing methods such as Electron Beam Melting, Laser Beam Melting and Direct Energy Deposition used to fabricate commercially available alloys are explained. The microstructure / grain structure resulting from directional building, complex thermal cycles is discussed. An overview of the properties of the superalloys and their performance as well as applications is presented. Keywords: Additive manufacturing, Laser Beam Melting, Electron Beam Melting, Nickel Superalloys I. INTRODUCTION ASTM defines additive manufacturing as layer by layer deposition of materials and joining them, directly from the co-ordinate data obtained from a 3D model. Additive manufacturing is also known by many names such as 3D Printing, Freeform fabrication, Rapid Prototyping etc. [1]. Earlier development of additive manufacturing was focused on manufacturing prototypes from complex polymers. Thus, rapid prototyping. Recently there has been a development in manufacturing metal additive manufacturing route. This process facilitates production of complex, custom metal parts which cannot be produced by conventional methods [2] . The advantages of additive manufacturing processes include reduced time for the component manufacturing/ development, little to no waste, flexibility in manufacturing parts with directional solidification, control of the porosity of the part and the ability to produce finer microstructures due to rapid the name through parts heating and cooling cycles which is usually difficult by the conventional manufacturing methods [3].Most of the metal additive manufacturing technologies use feedstock which is then melted by a concentrated heat source such as laser or electron beam up to a certain depth to fuse it layer by layer and cooled afterwards to build a solid part. The feedstock used may be in the form of wire or powder [4]. and heating cooling Materials used for metal additive manufacturing undergo through very complex thermal processing, i.e. subsequent cycles. The microstructure developed for these components differs from the conventionally manufactured components. Also, there is a difference between the properties of the parts due to the processing. A lot of work is done with alloys such as Ti-6Al-4V. [1] Nickel base superalloy components have been the new focus of development with additive manufacturing processes where the applications have high precision requirement with low volume such as precision and medical applications [3]. engineering, aerospace In this article, the methods used for additive manufacturing of metals/ superalloys, the microstructure of the alloys due to the directionality of the fabrication process, and properties and performance of the superalloys is discussed. II. TECHNIQUES FOR ADDITIVE MANUFACTURING OF METALS Two methods based on the feedstock used for the manufacturing can be classified as Powder Bed Fusion processes (PBF) and direct metal deposition (DMD) or direct energy deposition (DED) [4]. Powder Bed Fusion is among some of the first commercial processes of Additive manufacturing. Powder bed fusion process uses thermal source to fuse the powder particles layer by layer. Different mechanisms are used to spread the powder layer over the build plane [5]. More methods are developed based on the basic Powder Bed Fusion process such as Laser Beam Melting (LBM), Electron Beam Melting Page 1 (EBM), Direct Metal Laser Sintering (DMLS), Laser Metal Fusion (LMF). These processes utilize similar approach of melting/ sintering the powder layer fusing it with the previous layer using various sources of hear such as Laser, Electron Beam [4]. The Direct Metal Deposition process manufactures parts by melting the metal simultaneously as it is deposited. These methods use a more focused heat source. Unlike the powder bed processes, this process does not melt the material powder that already present on the build platform. The usual mechanism contains two types of nozzles, one houses the metal delivery system which may be powder of wire. The other type of nozzle houses the energy sources which immediately melts the metal that is coming out of the nozzle while following the building path/trajectory fed by the CAD model [5]. Several types of additional processes have been developed with this method such as Laser Metal Deposition (LMD), Direct Energy Deposition (DED), Laser Engineered Net Shaping (LENS)etc. [4]. The two widely used processes used for manufacturing of metal alloy components are the Laser Beam Melting (LBM) or Selective Laser Melting (SLM) and Electron Beam Melting (EBM) [2]. A. Selective Laser Melting Laser Beam Melting (LBM) or Selective Laser Melting (SLM) is one of the most promising methods used for Manufacturing metal alloy components. Any additive manufacturing process starts with a 3D CAD model. The model is then sliced into 2D layers of a layer thickness decided by the energy source used for melting the powder [6]. The powder bed is created by rolling the powder on the build platform by using rollers or cassettes. The usual thickness the powder layer is several powder particles. [2] The cross-sectional area is then scanned by rastering the laser spot over the fed 3D CAD model. High power lasers such as CO2, Yttrium are used as the energy source [6]. Then the build plat form is lowered by the same dimension of the layer thickness, another layer of the powder is rolled over the build platform and the process is repeated till the part is manufactured [5]. The average layer thickness for laser melting methods ranges from 20 m – 100 m [4]. After the metal powder is fed by the hopper or a reservoir, the roller uses a device called recoater to make sure that the powder is spread with uniform thickness [4] [5]. The laser beam is then rastered with scan speed of maximum 15m/s [4]. The scan speed depends on the absorption of the radiation of the energy source into the material as well as the layer thickness that is to be manufactured and other parameters [5]. The thickness of the wall of the structure built depends on the spot size of the laser. The typical spot size ranges from 50 m – 180 m [4]. The transfer of energy from the laser to the powder layer is affected by factors like absorption, conduction, convection heat transfer, scattering, the flow of fluid when the powder is melted. The major factor is the energy density which controls the layer thickness and the scanning speed. Higher scanning speed translates to lower energy density resulting in Figure 1: Schematic of Selective Laser Melting Process [6] unstable melt pool of the powder. Which ultimately delivers poor surface roughness [4]. The primary features of the SLM process are the when concerned with metallurgical point of view is that, when the powder is heated to high temperature gradients, it causes non- equilibrium conditions at the solid/liquid interface. The raster of the laser beam causes cooling of the melt pool which causes it to transform from liquid to solid. Thus, changes in the microstructure of the are observed. Finer microstructures can be formed by increasing the cooling rates sufficiently. The grain structure of the part is controlled by the layer that is solidified before the current layer. The relation of processing parameters for SLM is shown in the Figure 1 [6]. B. Electron Beam Melting In Electron Beam Melting (EBM) the powder bed is created by similar method as Laser Beam Melting (LBM). The heat source used is electron beam instead of laser beam which is generated by an electron gun. As described in figure, the electron beam is accelerated with a certain accelerating voltage and then is focused using Page 2 Figure 2: (a) Schematic of an EBM system (b) SEM image of the precursor powder (c)Histogram of particle sizes of the powder [7] electromagnetic lenses. Before starting the build, the powder bed is preheated with the help of defocused beam. With the help of electron beam, higher temperatures can be achieved. This helps sinter the powder before staring the actual build scan. The beam current 5mA – 10 mA and scan speed 10mm/s are used for the subsequent scans [4]. The build is often conducted in an inert gas environment like Argon to prevent oxidation. C. Direct Metal Deposition With Direct Metal Deposition (DMD) or Laser Metal Deposition (LMD) or Direct Energy Deposition (DED) the part is built while simultaneously feeding the metal. The metal can be fed in the form of wire as well as powder. The metal powder/ wire is fed via nozzle, the nozzle can be a single nozzle or a coaxial nozzle surrounding the energy source [5]. LMD is more flexible in terms of the build parameters. Such that the layer thickness can be varied from 40 m to 1 mm, spot size can be varied from 0.3 mm to 3 mm and the scanning speeds from 150 mm/min to 1.5 m/min. The reason that the system is flexible is that the build plate is stationary, and the deposition head moves. The deposition head is like a 5-axis head which can move around. Some systems have been developed where the head remains stationary and the build plate moves [4]. LMD has wide usage as it can be used to repair cracks, broken parts in turbine systems and gear mechanisms which generally cannot be repaired by conventional methods. LMD can also be used in combination with subtractive methods such as milling machine and a new type of hybrid manufacturing technique has been developed. III. FEEDSTOCK FOR ADDITIVE MANUFACTURING OF METALS Most common materials used for additive manufacturing consist of Aluminum alloys, Titanium alloys, Nickel based superalloys. To generalize, any weldable material can be considered as a suitable material for metal additive manufacturing. Also, while using alloy powders it should observed that some alloys crack under higher solidification rates which are achieved while fabricating, are not appropriate. The crystal structures and mechanical properties of the components produced are different from the ones produced by the conventional manufacturing methods [5]. Page 3 A. Powder production Most of the metal powders are produced by atomization technologies which include water, plasma or gas atomization processes [4]. Powder characteristics such as particle size, morphology of the particle and chemical composition are changed with different powder production methods. The powder properties have considerable impact on the final part that is produced by the AM methods. The fluidity and packing factor of the powder properties have influence on the density on porosity of the part [4]. Water atomization process is the simplest and low-cost atomization process. Liquid metal is atomized by high speed water jets while the metal is falling through the atomization chamber. Particle sizes from ~ 5 m to 500 m can be produced by this method due to high cooling rates provided by the water. But this might lead to the particles having an irregular morphology. Which might not be suitable for AM as it reduces the packing factor of the powder. Water atomization also leads to higher oxygen content in the powder particles. This leads to oxide formation of the metals which leads to changes in powder flow behavior and the component composition. This may not be desirable for the materials where oxides have detrimental effects on the mechanical properties of the bulk material [4]. This disadvantage can be overcome by using inert environment with gas atomization. The method works for both reactive and non-reactive materials. The selection of gas influences the "development of microstructure" and in turn the microstructure of the bulk material of the part. The metal melt can be produced by two methods either by vacuum induction melting or electron induction melting where an electrode rod is made from the desired metal and then liquified by passing high intensity current through it. This method mostly results in spherical particles of the metal powder [4]. Plasma atomization can also be used to produce the powder required. In plasma atomization a thing wire is torched with plasma due to which it spherodizes and process spherical particles. The powder particles from plasma atomization are more uniform and fine having sizes of approximately 40 m [4]. IV. COMPOSITION OF NICKEL SUPERALLOYS Superalloys are the types of alloys that are able to retain high strength at elevated temperatures. These have a very complex composition compared to the usual alloys. These also showcase much higher resistance to corrosion and oxidation, and excellent resistance to creep and rupture at elevated temperatures. These are particularly suitable for applications which demand creep resistance at higher temperatures such as aircraft, gas turbines, rocket engines, chemical and petroleum plants as Nickel based superalloys are able to retain their strength when exposed to temperatures above 650C for long times. There are Nickel based, Nickel-Iron based and Cobalt based superalloys [7]. For additive manufacturing methods mostly Nickel based alloys are used. These superior properties of Nickel based superalloys are achieved by either solid solution strengthening or precipitation hardening. Most of these alloys are contain 10-20%Cr, up to 8%Al and Ti, 5 to 10% Co and other additions of Mo, Nb, Zr, B, W, Ta, Hf. A. General Microstructure of Nickel based superalloy The major phases that are present in the are as follows; • Gamma () – This the continuous matrix with FCC (face centered cubic) crystal structure. This is called Nickel based austenite. The phase is strengthened by solid solution elements such as Cr, Co, Mo, W, Fe, Ti and Al. Due to slow diffusion of some of the elements the alloy exhibits resistance to high temperature creep [7]. • Gamma Prime (') – This is the primary strengthening phase which can be precipitated by precipitation hardening treatments. The ' phase is precipitate in the FCC matric of the form A3B. The "A" is comprised of mostly electronegative elements such as NI, Co and Fe, and "B" is constitutes more electropositive elements such as Al, Ti or Nb. Usually ' is Ni3(Al,Ti). This phase exhibits the ordered L12 crystal structure. The difference in the lattice parameter of  and ' phase is about ~0.1% which lets ' precipitate homogeneously in the matrix. The ' is ductile, thus it provides strength to the matrix without compromising its fracture toughness. As ' is long-range ordered, the degree of the order increases with increase in temperature. This results in high strength in temperatures up to 800C [7]. • Carbides – Usually carbon is added from 0.02 – 2% by weight. It combines with reactive elements such as Ta, Ti and Hf to form carbides (TaC, TiC, HfC). As a result of heat treatment, the carbides Page 4 Figure 3: (a) SEM morphology of Inconel 625 powder (b) illustration of the laser scanning path [9] decompose to lower order such as M23C6 and M6C (M = Metal). The carbides generally have FCC structure. Carbides may be disadvantageous if their amount exceeds a certain percentage, i.e. carbon % is limited to 2%. Below that, carbides are considered to be advantageous when precipitated at grain boundaries, increasing the rupture strength at elevated temperatures. Usually M23C6 form at lower temperature heat treatments (760 - 980C) and • Topologically Close-Packed Phases – These are usually undesired phases formed during heat treatments. These phases are mostly brittle and form when the composition of the superalloy is not properly controlled. These from in the shape of plates parallel to {111}. These are detrimental because they tend to lower the rupture strengths and ductility. V. MICROSTRUCTURE OF ADDITIVELY MANUFACTURED SUPERALLOYS in A. Macroscopic surface morphology The generic test pieces of all the researches are found to be cubic blocks with the laser raster pattern as mentioned the Error! Reference source not found.(b). On the top surface a "V" shaped structure was observed similar to welding materials. The adjacent tracks of the raster are found to be overlapping closely with each other. This can be traced to the intense laser heat source which forms a molten pool of the powder as it moves. The powder at the start of the pool keeps continuously melting and the liquid alloy powder quickly solidifies as the laser beam follow the rastering path. This causes difference in the temperature at the head of laser and at the tail of the laser. This difference instigates a contrast in the density and surface tension which generates a convection current in the path of the laser. The convection causes stirring of the molten pool which leads to the formation of the "V" Figure 4: SEM micrograph of (a) top surface i.e. X-Z direction (b) scales formation in Y-Z direction (c) higher magnification of molten pool and the pool boundary (Inconel 625 via SLM) [9]. Page 5 morphology. The angle of the "V" depends on the speed of the lase rastering. The angle gets smaller as the laser speed increases, and in turn the melt pool gets narrower. This summarizes the morphology in the Z direction [8]. The Y-Z direction shows scale type structure. As the layer thickness of the SLM process is less than 0.05 mm, most of the previous (bottom) layer is re-melted while depositing the subsequent layer. The width of the melt pool is bigger than the spot size of the laser. Similar phenomenon is observed in the Z direction and a slight dip is seen in the middle of the pool. The energy dispersion of the laser into the material exhibit a Gaussian profile. The "scales" seem to match the profile [8]. B. Dendritic growth The microstructure of the superalloys Columnar dendrite structure which additively is highly manufactured nickel-based dependent on the processing parameters and the process history of the part. Hot isostatic processing is performed on most of the parts. As columnar grains formed via additive manufacturing, the microstructure exhibits anisotropy. [9]. The macroscopic properties are substantially affected by the microstructure of the part. The SLM process has very high heat signature which consequences low temperature gradient. As mentioned in [8] "the cooling speed can be as high as 106 K/s for SLM process". Due to this the parts manufactured by SLM exhibit distinct properties than cast superalloys and the wrought superalloys when comparing the microstructure and phase composition [8]. The directional anisotropy that is observed that is due to directional columnar grain growth caused by directional thermal conduction. the highlight of the additive manufacturing process are formed in the Z direction. Due to the dendrite structure the anisotropy in the microstructure is exhibited. The as processed alloys show a microstructure with significantly elongated grains. Most of the alloys show the dendrite growth in the direction <001> [10]. Most of the superalloys the matrix  has face centered cubic (FCC) crystal structure. While scanning the laser melts the top of the power layer, the layer on the bottom remains solid (it does not reach the melting temperature). This forms the temperature gradient flowing from top to the bottom. As the heat flux flows into the material, the angle between the direction of the flux and the crystallographic direction <001> is the lowest. Thus, the direction <001> is preferred for the dendrites to grow. Another reason for the directional grain growth can be attributed to the high temperature that is achieved during the SLM/EBM process. The temperatures of the molten pool may reach is up to 1800C. This superheats the melt pool which presents a difficulty in forming homogeneous nuclei. Thus, growth of columnar dendrites is promoted and the dendrites form from bottom to the top. However, temperatures of this order may also lead to high surface tension gradient. This endorses another phenomenon called Marangoni convection. The velocity of the flow of convection an may break the law of columnar dendrite growth and the dendrites may grow in a direction other than <001>. This phenomenon is rare [8]. Researches have shown that the texture of the dendrites is highly influenced by the input laser energy or energy density. Also, the microstructure can be further refined by heat treatment. Some investigations into phase identification of the constituting phases of the Inconel 718 show that the phases  and ' may be the composition of the dendritic structure. Although, distinguishing between the phases is very difficult as found in the XRD characterization, the peaks of  and ' overlap [10]. The two phases are difficult to separate in most of the Nickel superalloys as the ' phase shows ordered structure L12 and it precipitates coherently with the  [3]. The solidification in the case of metal additive manufacturing occurs the conventional methods. This affects he dendrite arm spacing which ranges from ~0.5 µm to 3 µm. Which is 2 orders of magnitude less than the dendrite arm spacing of established casting processes (100 µm – 300 µm). The dendrite spacing is specified by the formula compared rapidly to 𝑑 = 𝑎 ∗ −𝑏 Where d is the dendrite arm spacing; a and b are the material constants. For Nickel based alloys a = 50 µm and b = 1/3.  is the cooling rate of the melt pool. MC (Metal Carbides) were formed as precipitates in interdendritic regions. There are segregated during the laser scanning. MCs might help in strengthening and creep resistance if form in right amount, otherwise they are almost always detrimental as they precipitate along grain boundaries and may lead to intergranular fracture. Heat treatment is necessary if MC are precipitate din the microstructure [3]. The amount of the phases such , ' and MCs formed varies with different alloys. The MCs formed during SLM processing of the alloy have less effect on the microstructure. The primary structure of the superalloys was found to be austenitic FCC. Page 6 balls with porosities on the surface. As  is increased the laser scan tracks started becoming more continuous and uninterrupted. The balls of the materials reduced in size and the porosities were dispersed and ultimately vanished increasing the density of surface cross section. To compare, when the energy density was increased to 300 J/s, the surface was discovered to be very smooth and the number of metallic globules formed were significantly decreased. When the energy density was further increased to 330 J/m, the density increased to 98.4% as compared to 73.6% at  = 180 J/m. Other factors such as the scan speed, temperature and viscosity of the liquified metal also affect the surface morphology. As the scan speed of the laser is increased, the dwell time decreases and effectively the temperature at the melt pool is decreased. The combined effect of this is higher dynamic viscosity, which in turn increases the porosity of the surface [10]. Figure 5: (a) Dendritic microstructure of Nimonic 263 [11] (b) Densritic microstructure of IN100 [3] C. Effect of Energy Density on the Microstructure The research [10] suggests that as the energy density () of the laser increases the XRD diffraction peaks for the  and ' phase broaden, and the intensity of the peaks is significantly decreased. This suggests that as the energy density goes up the crystallite or grain size gets smaller. Also, the diffraction peaks were shifted to a higher angle as  increases. This peak shift indicates lattice distortion, that is increase in the lattice plane distant "d". The lattice distortion is because the  phase incorporates more uniformly when compared to lower  [10]. Initially looking at the change in the surface morphologies, research [10] considers 4 different cases where the energy density is varied as  = 180 J/m, 275 J/m, 300 J/m and 330 J/m. At the lower values of  the laser rastering tracks were found to be discontinuous. It was also observed that the microstructure had large sized Figure 6: Surface morphology of Inconel 718 (a) with 180 J/m (b) 275 J/m (c) 300 J/m (d) 330 J/m, (e) Change in density with the change in energy density [10].. When looking at the morphology of the dendritic structure, with faster scanning speeds and lower energy density the columnar dendrites formed were coarse and were mostly composed of the  phase. The columnar grains were said to be "long-cracked in the center of the whole trunk and short-cracked in the interdendritic region". When the scan speed is decreased, and the energy density increased, the dendrites get finer. The dendrites also start to cluster, and it becomes difficult to separate [12]. The energy density not only affects the morphology of the dendritic structure, but also affects the precipitates formed. With high energy density the cooling rates are low. With low cooling rates, more amount of ' phase is Page 7 precipitated. With fast cooling rates, less amount' phase gets precipitated in  matrix. In case of little to no precipitation of ' phase, heat treatment is necessary to segregate the dispersed MCs and to correctly precipitate the ' strengthening phase [9]. D. Effect Heat Treatment on the Microstructure For Nimonic 263, high The heat treatment differs for different type of alloys as they have different compositions and different amount of precipitate formed. Heat treatment brings the microstructure of the additively manufactured alloys to the conventional cast alloys. The first stage of any standard heat treatment is high temperature heating. This solutionizes the ' precipitate phase either partially or totally with the  matrix in conventional cases [9]. temperature heat treatment removes dislocations and homogenizes whatever segregations formed. The solutionizing effect might not be seen as compared to the conventional processing of the alloys. As less amount of ' phase is formed because of rapid cooling during the process, therefore solutionizing the ' phase becomes redundant. This is then followed by water quenching, to make sure that high number of nuclei are formed. The quenching may again be fast that the ' precipitates might not form. Annealing helps even distribution of the whatever nuclei are formed. Formation of secondary MC has been seen after annealing of the Nimonic 263. After the heat treatment the width of the dendritic elongated grains is seen to be increased. Thus, the aspect ratio of the columnar grains is decreased [11]. In case of Inconel 718 when it is processed through EBM route, the ' phase is precipitated in  matrix. The precipitates align parallel to the building direction. The heat treatment process in this case does not help significantly to refine the microstructure as ' phase is already precipitated. Aging treatment comes out to be more helpful to get rid of Laves phase [9]. With Inconel 718 manufactured with SLM the microstructure shows dendritic growth and some MCs and Laves phase precipitated in interdendritic region. Strengthening phases such as  and ' are not as clearly distinguished in the matrix. Considering this, heat treatment the strengthening phases properly. The heat treatment also helps with segregation of the MCs and Laves which in turn enhances macrostructural properties [9]. When looked at Inconel 100, the heat treatment pushes the lighter elements in the alloy to interdendritic regions and the heavier elements seem to get congregated is necessary to help precipitate Figure 7:Dendritic structure formed via AM processes with Inconel 718: (Dentridritic structure gets refined with increase in energy density [11] in the center of the dendtritic structure. 3 strep heat treatment process is preferred. After the heat treatment 3 types of precipitates primary ' which are larger in sizes (500 nm dia), secondary ' precipitates (100 nm) and tertiary ' precipitates (10 nm). These three types of precipitates are formed at different temperatures during different heat treatments. Primary ' were formed during solution treatment phase. Secondary 'phase was formed at cooling temperature of the annealing process. Tertiary ' phase was formed while aging the additively manufactured part [3]. Inconel 625 exhibits somewhat similar results to the heat treatment. The heat treatment makes the grains grow in size, but the grain size seems to be less than the grain size observed with the as cast Inconel 625. After the first heat treatment the grains prefer a certain orientation. The grains finally grow to be rectangular in shape. After the second heat treatment a zigzag type boundary is observed. The zigzag shape of the boundary is important for the ductility. Page 8 VI. PROPERTIES OF ADDITIVELY MANUFACTURED SUPERALLOYS The properties of the as processed alloys were found to be slightly inferior compared to the as cast alloys. But after heat treatment and aging, the properties improve drastically, and they are on par with the as cast alloys. As we can see from Figure 9 that the Yield Strength (849 MPa vs 738 MPa), % Elongation (22.8 vs 5), hardness (32.5 HRC vs 25 HRC) and Ultimate tensile strength (1126 MPa vs 862 MPa) is greater for as fabricated (SLM) compared to the cast alloy. When again the parts are heat treated the properties show significant increase and are greater than the wrought alloy [6]. From the research we can say that the "superalloy" properties can be retained for the additively manufactured Nickel-based superalloy. The properties of as fabricated alloys show anisotropy suggesting better properties in the build direction (Z- direction) compare to X and Y directions. After heat treatment, the properties seem to approach isotropy. Figure 8: (a) Hardness of SLM produced Inconel 625 with respect to temperature (b) Stress strain curve comparison of as cast and SLM processed Inconel 939 [6] VII. APPLICATIONS OF ADDITIVELY MANUFACTURED SUPERALLOYS show promising The applications of additive manufacturing are currently restricted to low volume production as mass production of additively manufactured parts is difficult and costly as of now. AM is appropriate for parts having very complex geometry. Usually development and manufacturing of such complex parts is usually more expensive than producing the part with AM. Am processes in developing personalized parts and complex structures for aerospace, automobile, energy and biomedical fields. Aerospace components have mostly very complex geometries and their usage is often subject to high strength, high temperatures which asks for high performance materials such as Titanium alloys, Nickel Superalloys, Special Steels etc. The demand for aerospace applications is also less compared to other applications (few thousand parts), therefore AM is very suitable for such applications [13]. Investigations show that the components manufactured suture Page 9 Figure 9: Stress strain curves comparing the properties of As fabricated, heat treated, cast and wrought Inconel 718 [6] via AM posses comparable and sometimes higher strength for the required applications. Many commercial components use additive manufacturing processes such as LENS, LMD, EBM etc. to manufacture components for satellites, spacecrafts, airplanes, helicopters etc. Complex components such as required AM has been applied for repairing of expensive parts. Some components in aerospace applications are very costly to reproduce if damaged. The damaged component can be repaired by filling the damage via AM processes. Figure 12 shows damaged blisk repaired using LENS [13]. Figure 11: Airfoils with duct manufactured from Inconel 738 via LMD process [13] Page 10 Figure 10: Turbine blades with thin walls and complex channels with Inconel 718 [14] "turbine blades with thin walls and complex channels" as shown in Figure 10 are manufactured with Inconel 718. Some companies have successfully manufactured and implemented additively manufactured vanes, stators, rotors, geometrically complex parts such as airfoils, blisks, ducts and diffusers. Figure 11 showcases airfoils manufactured from Inconel 738 via LMD process [13]. Hybrid manufacturing processes combining "multi axis laser deposition and CNC machining" have been developed which simultaneously deposit and machine the part according to the design. References [1] W. E. Frazier, "Metal Additive Manufacturing: A Review," Journal of Materials Engineering and Performance, vol. 23, no. 6, pp. 1917-1928, 2014. [2] S. M. G. D. A. R. E. M. J. H. K. N. A. P. W. S. F. R. M. R. B. W. Lawrence E. Murr, "Metal Fabrication by Additive Manufacturing Using Laser and Electron Beam Technologies," Journal of Material Science and Technology, vol. 28, no. 1, pp. 1-14, 2012. [3] G. . Bi, C. . Sun, H.-C. . Chen, F. L. Ng and C. C. K. Ma, "Microstructure and tensile properties of superalloy IN100 fabricated by micro-laser aided additive manufacturing," Materials & Design, vol. 60, no. , pp. 401-408, 2014. [4] D. . Herzog, V. . Seyda, E. . Wycisk and C. . Emmelmann, "Additive manufacturing of metals," Acta Materialia, vol. 117, no. , pp. 371-392, 2016. [5] I. . Gibson, D. . Rosen and B. . Stucker, "Additive Manufacturing Technologies," , vol. , no. , p. , 2015. [6] M. C. Karia, M. A. Popat and K. B. Sangani, "Selective laser melting of Inconel super alloy-a review," , 2017. [Online]. Available: http://aip.scitation.org/doi/abs/10.1063/1.4990166. [Accessed 31 3 2018]. [7] W. F. Smith, Structure and Properties of Engineering alloys, McGraw-Hill, Inc. , 1993. [8] Q. W. Y. S. Z. Z. D. Z. Shuan Li, "Microstructure Characteristics of Inconel 625 Superalloy Manufacture by Selective Laser Melting," Journal of Materials Science and Technology , vol. 31, pp. 946-952, 2015. [9] D. Deng, "Additively Manufactured Inconel 718 ： MIcrostructures and Mechanical properties," Division of Engineering Materials Department of Management and Engineering Linköping University SE-581 83 Linköping, Sweden, Linköping , 2018. [10] Q. . Jia and D. . Gu, "Selective laser melting additive manufacturing of Inconel 718 superalloy parts: Densification, microstructure and properties," Journal of Alloys and Compounds, vol. 585, no. , pp. 713-721, 2014. [11] C. J. B. L. N. M. S. T. Vilaro, "Microstructure and Mechanical approaches of the selective laser melting of the selective laser melting process applied to a nickel- base superalloy," Material Science and Engineering A, vol. 534, pp. 446-451, 2012. [12] H. . Helmer, C. . Körner and R. F. Singer, "Additive manufacturing of nickel-based superalloy Inconel 718 by selective electron beam melting: Processing window and microstructure," Journal of Materials Research, vol. 29, no. 17, pp. 1987-1996, 2014. [13] M. C. L. Nannan Guo, "Additive manufacturing: Technology, applications and research needs," Frontiers Page 11 Figure 12: Damaged area repaired with LENS process VIII. CONCLUSION cast To summarize, additive manufacturing is a novel but expensive method to fabricate metal parts. The method is now developed to be able to fabricate complicated and intricate parts that are required for special purpose applications such as aerospace, biomedical, energy and automobile. Nickel superalloys as a material can be a suitable option for theses special purpose applications. The properties of Nickel such as High temperature creep resistance, high temperature rupture strength, retaining the elongation and hardness which makes it viable for this. The microstructure of parts manufactured from Nickel using AM differs from the conventional and wrought methods. The microstructure exhibits dendrite like structure (which can also be seen in cast alloys) with aspect ratio of about 5 to 6.5 (much greater than cast). This directional growth of the dendrites affects the isotropy of the properties. Thus, making the part stronger in the build direction of the AM process. The heat treatment done on the parts help increase the width of the dendrites, in turn reducing the aspect ratio and getting the properties closer to isotropy. The cost and time required to develop parts from conceptual stage via AM comes out to be less than using conventional methods. Yet, AM cannot be used for mass manufacturing, but low volume requirement can utilize this method efficiently. of Mechanical Engineering , vol. 8, no. 3, pp. 215-243, 2013. [14] Y. . Idell, C. E. Campbell, L. E. Levine, F. . Zhang, G. B. Olson and D. . Snyder, "Characterization of Nickel Based Superalloys Processed Through Direct Metal Laser Sintering Technique of Additive Manufacturing," Microscopy and Microanalysis, vol. 21, no. , pp. 465- 466, 2015. [15] H. . Wei, T. . Mukherjee and T. . DebRoy, "Grain Growth Modeling for Additive Manufacturing of Nickel Based Superalloys," , 2016. [Online]. Available: https://link.springer.com/content/pdf/10.1007/978-3- 319-48770-0_39.pdf. [Accessed 31 3 2018]. [16] A. . Basak and S. . Das, "Additive Manufacturing of Nickel-Base Superalloy IN100 Through Scanning Laser Epitaxy," JOM, vol. 70, no. 1, pp. 53-59, 2018. [17] L. E. Murr, "Metallurgy of additive manufacturing: Examples from electron beam melting," Additive manufacturing, vol. 5, no. , pp. 40-53, 2015. [18] L. E. Murr, E. . Martinez, K. N. Amato, S. M. Gaytan, J. . Hernandez, D. A. Ramirez, P. W. Shindo, F. . Medina and R. . Wicker, "Fabrication of Metal and Alloy Components by Additive Manufacturing: Examples of 3D Materials Science," Journal of materials research and technology, vol. 1, no. 1, pp. 42-54, 2012. [19] L. E. Murr, E. . Martinez, S. M. Gaytan, D. A. Ramirez, B. . Machado, P. W. Shindo, J. . Martinez, F. . Medina, J. . Wooten, D. . Ciscel, U. . Ackelid and R. . Wicker, "Microstructural Architecture, Microstructures, and Mechanical Properties for a Nickel-Base Superalloy Fabricated by Electron Beam Melting," Metallurgical and Materials Transactions A-physical Metallurgy and Materials Science, vol. 42, no. 11, pp. 3491-3508, 2011. Page 12
1903.09593	1	1903	2019-03-22T16:41:27	Design Principles for Self-forming Interfaces Enabling Stable Lithium Metal Anodes	[ "physics.app-ph", "physics.chem-ph" ]	The path toward Li-ion batteries with higher energy-densities will likely involve use of thin lithium metal (Li) anode (<50 $\mu$m in thickness), whose cyclability today remains limited by dendrite formation and low Coulombic efficiency. Previous studies have shown that the solid-electrolyte-interface (SEI) of Li metal plays a crucial role in Li electrodeposition and stripping. However, design rules for optimal SEIs on lithium metal are not well-established. Here, using integrated experimental and modeling studies on a series of structurally-similar SEI-modifying compounds as model systems, we reveal the relationship between SEI compositions, Li deposition morphology and coulombic efficiency, and identify two key descriptors (ionicity and compactness) for high performance SEIs through integrated experimental and modeling studies. Using this understanding, we design a highly ionic and compact SEI that shows excellent cycling performance in LiCoO$_2$-Li full cells at practical current densities. Our results provide guidance for the rational selection and optimization of SEI modifiers to further improve Li metal anodes.	physics.app-ph	physics	Design Principles for Self-forming Interfaces Enabling Stable Lithium Metal Anodes Yingying Zhu,ë, é Vikram Pande,^, é Linsen Li,ë, †, * Sam Pan,† Bohua Wen,† David Wang,† Venkatasubramanian Viswanathan,^, * Yet-Ming Chiang†, * ë Department of Chemical Engineering, Shanghai Electrochemical Energy Devices Research Center (SEED), Shanghai Jiao Tong University, Shanghai, China 200240 † Department of Materials Science & Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA ^ Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh PA 15213， USA é These authors contribute equally to the work. * Corresponding authors Abstract The path toward Li-ion batteries with higher energy-densities will likely involve use of thin lithium metal (Li) anode (<50 µm thickness), whose cyclability today remains limited by dendrite formation and low Coulombic efficiency (CE). Previous studies have shown that the solid-electrolyte-interface (SEI) of the Li metal plays a crucial role in Li electrodeposition and stripping behavior. However, design rules for optimal SEIs on Li metal are not well established. Here, using integrated experimental and modeling studies on a series of structurally-similar SEI-modifying model compounds, we reveal the relationship between SEI compositions, Li deposition morphology and CE, and identify two key descriptors (ionicity and compactness) for high-performance SEIs. Using this understanding, we design a highly ionic and compact SEI that shows excellent cycling performance in high specific energy LiCoO2-Li full cells at practical current densities. Our results provide guidance for rational design of the SEI to further improve Li metal anodes. Main Text Matching high-voltage oxide cathodes (> 4 V vs Li+/Li) with thin lithium (Li) metal (<50 µm in thickness) anodes promises Li-ion batteries with specific energy exceeding 300 Wh kg-1.1 However, the cycle life of thin Li metal anodes is severely limited by short-circuits (i.e. "sudden death") caused by Li dendrite formation and low Coulombic efficiency (CE) as a result of side reactions between Li metal and electrolyte (i.e. "gradual death").2-4 Recently published work has shown that dendrite formation may be suppressed to some extent by employing three-dimensional current collectors,5,6 functionalized 1 separators,7-9 electrolyte additives,10-13 surface coatings,14-17, concentrated electrolytes,18-20 and solid electrolytes.21-23 With such improvements, it is urgent to address the low CE problem, for otherwise "gradual death" (running out of available Li or electrolyte dry-out) would likely occur before "sudden death" (short-circuit), and limit the cycle-life. Li metal is highly reductive and reacts instantaneously with electrolyte constituents upon contact to form a surface film generally referred to as the solid-electrolyte-interface (SEI).24,25 SEI formation consumes active Li+ ions and leads to coulombic inefficiency. To minimize such loss, it is necessary for the SEI formation reaction to be self-limiting. It has also been shown recently that the microstructure and properties of SEI can impact the crystal growth behavior of Li metal during electro-deposition (charging of the cell)26 and how the Li deposits are stripped during battery discharge.27 Therefore, SEI tuning may be a promising strategy for improving Li metal anode performance. A variety of compounds, such as Li3PO4,15 LiF,28,29 LiBr,30 LiI,31 LiNO3,32 Li2S8,33 AlI3,34 SnI2,35 Al2O3,36 and Cu3N,37 have been used to modify the composition and morphology of the SEI and have been shown to be effective in improving the cycling performance of Li metal anode. However, these studies have typically been conducted with Li metal anodes of larger thickness (usually >250 µm). For high energy density, feasibility must be demonstrated with thin Li metal anode. It also appears that SEI tuning often follows a trial-and-error approach that results in incremental improvement. Thus, it is necessary to establish clear selection criteria for effective SEI modifiers. Here, we first quantify the impact of Li metal thickness and CE on energy density and cycle-life of Li-metal rechargeable batteries. Then, using a model series of structurally-similar SEI-modifying compounds, we show the interrelationship between SEI compositions, Li deposition behavior, and CE. We identify two key descriptors (i.e. ionicity and compactness) for high performance SEIs using integrated experimental and modeling studies. Using this approach, electrolytes that result in a highly ionic and compact SEI enriched with LiF, Li2CO3, and Li2SO3, have been discovered, which form a dense Li film during electrodeposition (charge) and achieve both dendrite-free and high CE cycling. Li metal full batteries based on thin Li metal anodes (50 µm in thickness) and LiCoO2 cathodes (theoretical areal capacity ~4.2 mAh cm-2) demonstrate stable cycling exceeding 240 cycles (to 80% capacity retention) at practical C-rates (0.2 C charge/0.5 C discharge, 1 C » 3.7 mA cm-2). For even thinner Li metal anodes (20 µm thickness), full cells still cycle for 130 cycles. Our results provide guidance for rational selection and optimization of SEI modifiers to enable practical Li metal rechargeable batteries. 2 Impact of electrode thickness and coulombic efficiency on energy-density, specific energy and cycle- life of Li metal batteries The calculated energy density, specific energy, and cycle life of Li metal batteries consisting of a high-area-capacity LiCoO2 cathode (>4.2 mAh cm−2) and a lithium metal anode of various thicknesses (i.e. 20 µm-thick Li corresponds to ~4.12 mAh cm−2) are shown in Figure 1. The fraction of Li passed per cycle (Fp) can be calculated from the areal capacity of the cathode and the anode Fp = "#$%&'() "#$%&'()"+, [1] Here, Qcathode and QLi are the area capacities of the cathode and Li metal anode, respectively; Fp varies inversely with the percentage of Li excess of the battery, and naturally the battery reaches its highest energy density in the anode-free case, Fp = 1. Thus in Figure 1 the energy density/specific energy decreases as Fp decreases (Li excess increases), with the being calculated based on the mass and volume of the cathode, anode, current collectors, separator, and liquid electrolyte, and including a packaging factor, as given in detail in the Supplementary Information. The cycle life of the battery, n, may be predicted based on the CEavg, which is the Coulombic efficiency averaged over the number of cycles until the all the available Li, Qcathode + QLi, runs out. CEavg is calculated as follows: = 1 - "#$%&'()"+, -"#$%&'() CEavg = 1 - "#$%&'()"+, -"+, .$//)( .)0 #1#2) [2] (Qcathode + QLi)/n is the average Li loss per cycle. Here we assume charging to 100% state-of-charge, and that CE loss occurs only at the Li metal anode, not at the cathode. Therefore, QLi passed per cycle is equivalent to Qcathode during cycling. From Equation 1 and 2, the relation between Fp and cycle life (n) can be determined as: Fp = 3 -(3567$89) [3] The Fp versus n plots in Figure 1 are constructed based on Equation 3 at several selected values of CEavg. As shown in Figure 1, if the CEavg is as low as 80%, the battery cannot survive more than 60 cycles even with a 250-µm-thick Li metal anode (low Fp, large Li-excess). With a CEavg of 99.9%, even an anode-free battery can last more than 1000 cycles, approaching the cycle life of existing Li-ion batteries using graphite anodes. The cycle life of Li metal battery can be further increased to more than 2000 cycles when a thin Li metal anode (e.g. 20 µm or 50 µm) is employed. Due to the low density of Li metal (0.534 g cm-3), this only leads to a small reduction in gravimetric energy density but a large reduction in volumetric energy density. Thus the increase in the thickness of the lithium metal foil improves the cycle life, but at 3 the cost of specific energy and energy density. In this work, we demonstrate cycling of lithium metal foils with 20 and 50 µm thickness. Figure 1 Prediction of energy density and cycle life of Li metal batteries. The energy density, specific energy and cycle life are functions of the average Coulombic efficiency (CEavg) and fraction of Li passed (Fp) per cycle. The mass and volume of the electrodes, current collectors, separator, electrolyte, and packaging are included in the calculation of gravimetric and volumetric energy density respectively, as detailed in Supplementary Information. The 20 µm, 50 µm, 100 µm, 250 and 750 µm-thick Li metal films are commercially available. Selecting SEI modifiers The central theme of this work is to establish the selection criteria for SEI modifiers that create a stable Li metal anode interface, and one which has the potential to self-heal upon formation of cracks. The spontaneous reaction between Li metal and iodine (I2), which is historically used in primary batteries and 4 leads to the formation of a Li+ conducting layer and solid separator during operation,38 offers a starting -/I2, this limits the voltage capability paradigm. However, due to the shuttle reactions associated with I-/I3 of cathodes to less than 3.2 V.39,40 Similar shuttle reactions are present for other halide species involving bromide and chloride at potentials below that of 4 V cathodes.41 Among the halide series, this criterion leaves only LiF; the benefits of a LiF-rich SEI have been documented before.12,29 However, direct addition of LiF as a salt in common organic electrolytes is not effective due to its extremely limited solubility (<0.002 mM in dimethyl carbonate). Therefore, we chose to explore an approach whereby LiF-rich SEIs are formed through intentional decomposition of fluorinated electrolyte constituents at the Li metal surface. Density-functional theory (DFT) calculations were used to probe reactions of a wide range of fluorinated organic compounds and lithium metal surface to determine their propensity to form a desired SEI (Supplementary Fig. 1). Note that these compounds can be considered either as solvents or additives depending on the amount added into the electrolyte. We begin by focusing on a series of structurally similar fluorinated organic compounds, namely fluoroethylene carbonate (FEC), di-fluoroethylene carbonate (DFEC) and 3,3,3-trifluoropropylenecarbonate (also known as trifluoromethyl ethylene carbonate, CF3-EC) (see molecular structure in Figure 2a). The result from ethylene carbonate (EC) is also included for comparison. As shown in Figure 2b and 2c, FEC spontaneously decomposes to form LiF, unstable CO- anion and lithium salt of glycolaldehyde. On the other hand, DFEC decomposes partially upon ring opening, leading to formation of LiF and a large lithium alkoxide. Interestingly, CF3EC, despite containing more F in its molecular structure, does not decompose to form LiF. These results clearly show that not every fluorinated organic solvent decomposes to form LiF at the Li metal surface. Interestingly, during the screening process for SEI modifiers, we have identified another approach to enriching LiF in the SEI, whereby the F atoms are extracted from the electrolyte salt LiPF6. We discovered that 1,3,2- dioxathiolan-2,2-oxide (DTD, see molecular structure in Figure 2a) decomposes along with LiPF6 to form 2- anion (Figure 2b & LiF, PF5, ethane-1,2-diolate (similar to a decomposition product of FEC) and SO2 2c). These results indicate the potential for systematically tuning the inorganic and organic content in the SEI through spontaneous reactions between the organic electrolyte constituents and Li metal. 5 Figure 2 Decomposition of selected molecules at the Li surface studied by DFT calculations and XPS. a. Molecular structure of ethylene carbonate (EC), fluoroethylene carbonate (FEC), di- fluoroethylene carbonate (DFEC), 3,3,3-trifluoropropylenecarbonate (CF3EC), and 1,3,2-dioxathiolan- 2,2-oxide (DTD). b and c are top and side views, respectively, of final decomposition products of EC, FEC, DFEC, CF3EC, and DTD at Li (100) surface at the presence of LiPF6, as predicted by DFT calculations. The purple atoms represent Li, red represent O, grey represent C, green represent P, blue represent F, yellow represent S and silver represent H. FEC and DTD break down completely, while DFEC decomposes partially and CF3EC does not undergo significant breakdown. DTD also catalyzes the decomposition of LiPF6 leading to formation of LiF. d, e, and f are narrow-scan XPS spectra of F 1s, C 1s, and O 1s, showing that DFEC and CF3EC decomposes differently from FEC, despite their structural similarity. 6 To experimentally corroborate the DFT calculations, X-ray photoelectron spectroscopy (XPS) measurements were made to probe the SEIs of Li films deposited in electrolytes containing 1 M LiPF6 dissolved in EC-DMC (EL-0, DMC = dimethyl carbonate), FEC-DMC (EL-1), DFEC-DMC (EL-2), CF3EC-DMC (EL-3), and FEC-DMC + 3 wt% DTD (EL-4). Approximately the same amount of Li (~4.2 mAh cm-2) was deposited on Li/Cu substrates (50 µm-thick Li, 15 µm-thick Cu) by charging LiCoO2-Li cells to 4.5 V at 0.1 C. The deposited Li films were rinsed with fresh dimethyl carbonate (DMC) and dried under argon atmosphere before transferring to XPS measurements using an air-proof sample holder. Wide- scan XPS spectra of the SEIs (Supplementary Fig. 2) show that the F content in the SEI increases in the order of EL-0 (1.5 at.%), EL-3 (1.7 at.%), EL-1 (5.8 at.%), EL-4 (7.2 at.%), and EL-2 (10.6 at.%), which is in agreement with the trend predicted by the DFT calculations. Narrow-scan XPS spectra of F, C, and O are analyzed and the results are summarized in Figure 2d, 2e, and 2f, respectively. The F 1s spectra for all four SEIs show similar peaks that can be assigned to C-F ad LiF (Figure 2d). Interestingly, unlike the other three cases, there is more C-F than LiF in the SEI when EL-2 (DFEC-DMC) is used. Although XPS is a semi-quantitative method, it is safe to conclude that the SEI formed in EL-2 contains less LiF than that formed in EL-1 and EL-4. The F 1s spectrum of EL-2 may be explained by the partial decomposition of DFEC as suggested by the DFT calculation. The high intensities of the C-O peaks observed for the SEIs formed in EL-2 and EL-3, but not in EL-1 and EL-4 also support the conclusion that DFEC and CF3EC do not decomposes as completely as FEC (Figure 2e). According to the O 1s spectra (Figure 2e), ROLi species are formed when FEC and DFEC are used, while a large amount of alkyl lithium carbonate (ROCO2Li) is observed when CF3EC is used, which confirms the DFT results. It is also observed in EL- 4 that the presence of DTD in FEC-DMC promotes the formation of Li2CO3 over ROLi (compare EL-4 and EL-1). For EL-4, the S 1s spectrum was also collected and analyzed (Supplementary Fig. 3). Surprisingly, there is no S-containing species observed in the SEI. This may be explained by the decomposition rate of DTD versus FEC at Li surface. DFT calculations suggest that FEC decomposes directly on Li while DTD co-decomposes with LiPF6. Thus we may expect the DTD decomposition reaction to be slower. Once the Li metal surface is passivated by the decomposition products of FEC (such as LiF), its reactivity toward DTD is significantly reduced. However, on the Li metal film that was cycled 20 times and 50 times, ROSO2Li, Li2SO4, and Li2SO3 were indeed observed. This result suggests that S-containing species are gradually incorporated into the SEI during Li deposition/stripping cycles and may provide a healing function when the SEI cracks. 7 SEI and Li metal electrodeposition behaviors To investigate the relationship between SEI composition and Li metal electrodeposition behavior, scanning electron microscopy (SEM) was performed on the Li film deposited in EL-0, EL-1, EL-3, and EL-4 electrolytes using the LiCoO2-Li cells. The same Li films on which the XPS measurements were made were examined by SEM. The SEM used for this study was installed inside an argon-filled glovebox so that the samples were never exposed to air. Top-view SEM images showed that Li particles of several microns were deposited in EL-0, EL-1, and EL-4 (Figure 3a, 3b, and 3d). Smaller particles were observed in EL-3 (Figure 3c). The difference in Li electrodeposition behavior among the four electrolytes is most clearly seen in the cross-sectional SEM images in Figure 3e-3h. The deposited Li films were clearly thicker in the EL-0 (~35 µm) and EL-3 electrolytes (~37 µm) than EL-1 (~25 µm), and EL-4 electrolytes (~26 µm), despite having the same areal capacity or areal mass. Since the deposited capacity of 4.2 mAh cm-2 corresponds to a thickness of ~20 µm if we assume that the Li film is fully dense, we can estimate the porosity of the deposited Li film in the four electrolytes using the thickness of the deposited Li observed in cross-sectional SEM. These densities are 43% for the EL-0, 20% for the EL-1, 54% for the EL-3, and 22% for the EL4. It appears that the LiF and Li2CO3-rich SEIs formed in the EL-1 and EL-4 electrolytes promote the deposition of dense Li films and with less of a whisker-like Li morphology. This morphology appears to correspond to a high CE. It has been shown that whisker-like Li particles are prone to cracking during stripping and thereby lose contact to become "dead Li",27 which likely leads to a low CE. More whisker-like Li particles and a wavy surface of the deposited Li film were indeed observed when a thick Li foil was used as the counter electrode instead of the LiCoO2 electrodes (Supplementary Fig. 4). 8 Figure 3 SEM characterization of the deposited Li film on the Li/Cu substrates. a, b, c, and d are top-view SEM images of the deposited Li films on the Li/Cu substrates (50 µm-thick Li, 15 µm-thick Cu) in 1 M LiPF6 EC-DMC (EL-0), 1 M LiPF6 FEC-DMC (EL-1), 1 M LiPF6 CF3EC-DMC (EL-3), and 1 M LiPF6 FEC-DMC + DTD (EL-4), respectively. The Li films were deposited in LiCoO2-Li cells by charging at 0.1 C to 4.5 V vs Li+/Li. The same amount of Li (~4.2 mAh cm-2) was deposited for all five cases. Micro-sized Li particles were observed in EL-0, 1, and 4. Smaller particles were observed in EL-3. e, f, g, and h are the corresponding cross-sectional SEM images, showing three-layer structure consisting of the deposited Li, the original 50 µm-thick Li, and the underlying Cu substrate. The deposited Li films are thicker in EL-0 and EL-3 than EL-1, and EL-4. Panel i, j, k, and l are cross-sectional SEM images at higher magnifications. 9 Coulombic efficiency measurements using asymmetric Li-Li cells In order to accurately measure Coulombic efficiency during plating and stripping, it is necessary to have a limited lithium source so that the loss of working lithium can be traced. This can be done with a full cell using an intercalation cathode, but any losses at the positive electrode may be difficult to separate from those occurring at the lithium metal electrode. We developed an asymmetric Li-Li cell test that is able to accurately quantify the average Coulombic efficiency occurring over a number of cycles, which we denote CEavg (Figure 4a). The asymmetric cell consists of a two Li metal electrodes, one of which has a low area capacity that is systematically consumed during cycling. In the present study we used a 20 µm- thick Li film coated on a copper foil (QLi = 4.12 mAh cm−2) as this working electrode, while the counter electrode is a 750 µm-thick Li foil with a large excess of capacity. The two electrodes were assembled into a coin-cell with a polyethylene separator and liquid electrolyte. In the first half-cycle, a known amount of Li, in this instance 3.0 mAh cm−2 per cycle (QLi passed per cycle), is deposited on the thin working electrode (here, at a current density of 0.6 mA cm−2). The same 3.0 mAh cm−2 is then stripped from the working electrode. With each successive cycle, the same QLi passed per cycle is stripped and plated. Any Coulombic inefficiency erodes the initial 20 µm-thick Li film on the working electrode. Barring a short- circuit event, the original thin Li electrode is gradually consumed by the side reactions, either forming SEI or being isolated by SEI during cycling (Figure 4b) and forming so-called "dead Li." When all the initial Li at the working electrode is consumed, a voltage spike is observed. Three selected examples are shown in Figure 4c. Crucially, the appearance of voltage spike (denoted by the black arrows in Figure 4c) shows that short-circuits are absent. The average Coulombic efficiency over the number of cycles the cell experienced up to the voltage spike is CEavg, and is calculated from an equation similar to Equation 2: CEavg = 1 - ( "%'%$2 +, - ) ∙ ( 3 "+, .$//)( .)0 #1#2) ) = 1 - "+, .$//)( .)0 #1#2) "+, -"+, .$//)( .)0 #1#2) [4] 10 Figure 4 Asymmetric Li-Li cell design and test. a. Schematic illustration of an asymmetric Li-Li cell consisting of a thin Li metal electrode (working electrode, WE), a separator, and a thick Li metal electrode (counter electrode, CE). In the first half-cycle, a fixed amount of Li is electrochemically deposited onto the thin Li electrode, and then this amount is stripped and deposited repeatedly. b. Evolution of the Li film on the thin Li electrode (WE) during the test. The Li originally coated on the Cu substrate was gradually consumed by the side reactions, and some becomes "dead Li" insulated by a thick SEI layer. c. Voltage curves of three selected examples of the Li-Li asymmetric cell tests. The cycling current density was 0.6 mA cm-2 . The cycling area-capacity is 3.0 mAh cm-2. Li was first deposited on the thin Li electrode and then stripped. The final voltage spikes denoted by the black arrows indicate the end of the tests when there is no Li available for stripping anymore and the absence of short-circuits during the tests. Longer cycle time before the voltage spikes indicate higher CEavg, based on Equation 4. We tested the effectiveness of this approach by measuring the CEavg of the thin Li anodes in several selected electrolytes reported in previous literature. The results are summarized in Supplementary Table 2. Different values of CEavg were clearly observed for different electrolytes. The Li metal anodes were reported to cycle well in 1 M LiTFSI EC/tetrahydropyran,42 1 M LiAsF6 EC/2-methyl-tetrahydropyran,43 and 1 M LiTFSI 1,3-dioxolane/1,2-dimethoxyethane (1:1 v) + 1 wt% LiNO3 electrolytes44 but poorly in 1 M LiPF6 propylene carbonate.45 Correspondingly, high CEavg values were observed for the former three and a low CEavg was found for the latter. The CEavg of the thin Li metal anodes in the electrolytes EL-0 to 4 was then measured. The commonly used LiPF6 EC-DMC electrolyte (EL-0) showed a CEavg of 92.6%. Replacing EC with a fluorinated EC such as FEC and DFEC significantly improved the CEavg to 97.0% and 96.2%, respectively. However, this beneficial effect was not observed for CF3EC, which showed a very low CEavg of 20.9%. The EL-4 electrolyte with 3wt% DTD additive showed the highest CEavg of 11 97.9%. High CEavg was observed for the cases where the SEI contained more ionic compounds and the deposited Li was denser. Table 1 Average Coulombic efficiency of the thin Li anodes in different electrolytes Lithium Salt Solvents & Additives EC-DMC (1:1 v), EL-0 FEC-DMC (1:1 v), EL-1 DFEC-DMC (1:1 v), EL-2 CF3EC-DMC (1:1 v), EL-3 CEavg 92.6% 97.0% 96.2% 20.9% (1 M) LiPF6 LiPF6 LiPF6 LiPF6 LiPF6 FEC-DMC (1:1 v) + DTD, EL-4 97.9% Test conditions: 0.6 mA cm-2 current density, 3.0 mAh cm-2 cycling areal capacity. The asymmetric Li-Li cell test provides a useful platform to evaluate and compare the effectiveness of different electrolyte components (salts, solvents, and additives). It is able to quantify both the CE and area-specific impedance (ASR) of the Li metal electrodes, whereas the conventional symmetric Li-Li cell tests can only quantify ASR. Furthermore, "soft" short circuits can be difficult to differentiate from a low ASR in symmetric Li-Li cells, whereas the asymmetric configuration yields unrealistically high CE (i.e. outliers) when short circuits are present. The asymmetric Li-Li cell test is also more directly relevant to practical applications than the widely used asymmetric Li-Cu (or Ni, or stainless steel) cell test in which Li is deposited on bare metal current collector, since the use of thin Li anodes in a full cell in most instances will provide a better compromise between energy density and cycle life than the so-called "anode-free" configuration where deposition occurs on metal current collector. The cycling performance of the thin Li metal anodes in the presence of the different electrolytes and SEIs was further evaluated in Li metal full cells consisting of a "high-voltage" LiCoO2 cathode (area- capacity ~4.2 mAh cm-2 when charged to 4.5 V vs Li+/Li), and a 20 µm-thick Li anode coated on a copper foil (areal capacity ~4.12 mAh cm-2), and a single-layer polyethylene separator. 2025-type coin cell cases made of 316L grade stainless steel and a single-layer polyethylene separator, were used for the tests. The full cells were cycled at 0.2 C charge-0.5 C discharge between 4.5 and 3.0 V. Three formation cycles were performed at 0.1 C before the long-term cycling tests. The LiCoO2-Li (20 µm) battery with the EL-0 electrolyte showed rapid capacity fade (Figure 5a and 5b). The cycling performance was improved with 12 the EL-1 electrolyte. The best cycling performance was achieved with the EL-4 electrolyte containing the DTD additive. 80% of the initial capacity was retained for 131 cycles (Figure 5b). It is also interesting to compare the CE of the three cells. The best-performing cell using the EL-4 electrolyte provides higher CE over those using the EL-0 and EL-1 electrolytes (Figure 5c). When a 50-µm Li metal anode was employed instead of the 20-µm Li metal anode, the LiCoO2-Li full battery retained more than 80% of the initial capacity for 246 cycles (Figure 5d). This shows that thickness of lithium provides an effective way to strike an application dependent trade-off between cycle life and specific energy. Figure 5 Electrochemical tests of LiCoO2-Li full batteries. a. Voltage profiles of three LiCoO2-Li full batteries containing 1 M LiPF6 EC-DMC (EL-0), 1 M LiPF6 FEC-DMC (EL-1), and 1 M LiPF6 FEC- DMC + DTD (EL-4). The batteries were first cycled at 0.1 C for three cycles and then charged at 0.2 C and discharged at 0.5 C repeatedly between 4.5-3.0 V. The LiCoO2 electrodes are ~23 mg cm-2 in mass 13 loading and ~4.2 mAh cm-2 in areal capacity. The Li electrodes are ~20 µm in thickness, ~4.12 mAh cm-2 in areal capacity. Data from the cycle 1 to 100 is shown. b. Cycling performance and c. Coulombic efficiency of the LiCoO2-Li full batteries containing the three different electrolytes. EL-4 provides better cycling performance and higher Coulombic efficiency. d. Cycling performance of a LiCoO2-Li full battery containing the 1 M LiPF6 FEC-DMC + DTD (EL-4) electrolyte with a Li metal anode of 50 µm in thickness, ~10.30 mAh cm-2 in areal capacity. 80% of the initial capacity was retained after 246 cycles. It is important to note that the present results were obtained in the absence of any significant applied pressure. Pressure applied to lithium metal cells can dramatically improve their cycling stability. Unlike intercalation electrodes that undergo a comparatively small volume change during cycling, Li metal electrodes undergo a large volume change as they dissolve during discharge, which effectively reduces the stack pressure applied to the electrodes and likely changes the Li electrodeposition behavior. Observations made in the present external-pressure-free cells may differ from what occurs in wound cells that are under winding pressure, or stacked cells that may be under pressure from the cell casing or externally applied pressure. We expect that the present results would be further improved in cell configurations with significant applied pressure. The performance of the Li metal full batteries with the EL-4 electrolyte is among the best reported for Li metal rechargeable batteries. Under the same test conditions, the EL-4 electrolyte provides even better cycling performance and CE than the previously reported dual salt electrolyte10 (Supplementary Fig. 6). We further compared our results with others using an updated version of the "ARPA-E plot" previously shown in ref. 1 (Figure 6). The data points in the "ARPA-E plot" are analyzed in terms of four parameters: cumulative capacity plated (Ah cm-2), plated current density (mA cm-2), per-cycle areal capacity (mAh cm-2), and fraction of Li passed per cycle (Fp). Compared with other previous work using liquid electrolytes (point 15 to 35), our work stands out for high Fp (Fp = 0.42 for point 36; Fp = 0.23 for point 37) and relatively high per-cycle areal capacity (>3.2 mAh cm-2), which means higher energy density based on the analysis shown in Figure 1. The data points in red color (i.e. low Fp values) are from cells having relatively low overall energy density and far from meeting the DOE goals (point 1 and 2, green color, Fp = 0.8). In this work, we have demonstrated cells with a specific energy of 370 Wh/kg lasting 130 cycles and cells with a specific energy of 355 Wh/kg lasting 246 cycles at practical current densities and cathode loadings. We also use 20 𝜇m thin lithium and have more than twice the lithium utilization (~42%) compared to any previous published work. Further, the voltage profiles clearly rule out the possibility of 14 a soft short. Finally, our cells exhibit linear capacity fade, distinct from the typically observed non-linear drop in capacity after few cycles. Figure 6 Comparison of cycling performance of Li metal anode. This figure is an updated version of the one in Reference 1 ("Points 1 and 2 are DOE goals, 3 -- 6 are for LiPON thin-film cells, 7 -- 9 are PEO- based solid polymer electrolytes, 10 -- 12 are solid inorganic separators, 13 and 14 are custom nanostructures, and 15 -- 26 are liquid electrolytes"). Data points 27-37 are newly added based on recent literatures, which are from liquid electrolytes as well, just like points 15-26. See Supplementary Table 2 of this manuscript for references for each points. Data point 36 and 37 are from our work. Data point 36 and 37 are from LiCoO2 (~4.2 mAh cm-2)-Li (20 µm) and LiCoO2 (~4.2 mAh cm-2)-Li (50 µm) full cells, respectively. EL-4 electrolyte (1 M LiPF6 FEC-DMC + 3 wt% DTD) were used. The cycling data were shown in Figure 5. Compared with other previous work using liquid electrolytes, our work stands out for higher fraction of Li passed per cycle (Fp = 0.42 for point 36; Fp = 0.23 for point 37), which means higher energy density based on the analysis shown in Figure 1 of this manuscript. Descriptors for Effective SEIs In order to formulate the structure-property relationship between the electrolyte components and their function in full cells, we identify two descriptors that are key for a high performance, self-formed SEI. 15 The two key factors are (i) the ionicity of the SEI, which is needed for ensuring low electronic conductivity46 and (ii) the porosity of the SEI, which will control the morphology and packing density of the inorganic and organic phases. Here, we propose to use the number of electrons transferred from the lithium surface to the electrolyte molecule as a descriptor for the iconicity of the SEI. This descriptor is intuitive as the more ionic the SEI is, the more electrons should be transferred from the lithium surface, as shown in Table 2. On the other hand, describing the porosity and morphology of the SEI is a more challenging problem. We propose that the volume of the organic species left behind from the decomposition is a good descriptor to describe the ability to form a compact SEI. This volume can be quantified by the Bader Volume of the largest SEI species, typically the organometallic salt component. More details on the rationale behind the two descriptors is discussed in the Supplementary Information. These two descriptors, used in conjunction, can rationalize the experimentally observed trends. Based on the descriptor for the ionicity of the SEI, among the present fluorinated solvents, the trend is CF3EC << FEC < DFEC. Using just this descriptor, we would conclude that DFEC leads to a more ionic SEI. However, when comparing the descriptor for compactness of the SEI, the incomplete decomposition of DFEC leads to much larger moieties for DFEC than FEC. These two factors taken together indicates that FEC should fare better due to a combination of an ionic and compact SEI. The case of DTD is special and it suggests a new strategy to enrich the LiF content in the SEI. Unlike the other electrolyte molecules considered, DTD decomposes along with LiPF6 to form LiF, PF5, ethane- 2- anion. The latter eventually leads to the formation of Li2SO4, 1,2-diolate (similar to FEC) and SO2 Li2SO3 and ROSO2Li as observed in the experiments. For the two descriptors identified, DTD leads to the formation of a more ionic SEI while maintaining high density quite similar to FEC. We attribute the improved behavior seen in FEC-DTD case to increased LiF and Li2CO3 formation from the DTD addition as well 4as the formation of Li2SO3 as confirmed by the XPS results. These results suggest that these two descriptors together can provide rational design principles for the selection of SEI forming compounds. Table 2 Theoretical Descriptors for Solvent decomposition on Li metal Solvent Electrons Bader Volume Transferred of Largest SEI Specie (Å3) DMC EC 2.0 2.5 145 132 16 FEC DFEC CF3EC DTD 3.3 4.4 0.5 4.1 89 125 169 97 Conclusions We have shown that using thin Li metal anode and having high CE are two prerequisites for high energy-density and long cycle-life Li metal batteries. It is discovered that the decomposition of FEC and DTD (at the presence of LiPF6) at Li metal surface enriched the SEI with ionic compounds such as LiF, Li2CO3, and Li2SO3 rather than other organic species, whereas structurally similar DFEC and CF3-EC decompose and modify the SEI differently. A compact and ionic SEI promotes dense Li electrodeposition and high CE during cycling. LiCoO2-Li full cells demonstrated stable cycling performance over 240 cycles at practically relevant areal-capacity (>3.2 mAh cm-2) and C-rates (0.2 C charge/0.5 C discharge, 1 C = 3.7 mA cm-2). This work establishes a methodology for rational selection and optimization of the SEI modifiers to further improve the cycling performance of Li metal anode. Methods Materials. The 20 µm-thick lithium film coated on Cu foil was purchased from Honjo, Japan. The 50 µm- thick lithium film coated on Cu foil was purchased from China Lithium Energy Co., Ltd. The electrolytes were prepared by (Shanghai Songjing New-Energy Technology) using battery-grade reagents except EL- 2 because DFEC was only available at 95% purity (from BOCSCI Inc.).The LiCoO2 electrode sheets were prepared by coating the slurry of LiCoO2 cathode powder (LC-95, Hunan Shanshan), carbon black, and PVDF (weight ratio 96:2:2) on aluminum foils. The electrodes were calendared to ~60 µm. The mass loading of LiCoO2 was ~23 mg cm-2. Characterizations. XPS measurements were carried out using a Kratos X-ray photoelectron spectrometer. To avoid electrode contaminations caused by exposure to air, the Li electrodes were rinsed using fresh dimethyl carbonate, dried, sealed in a specialized holder inside the Ar-filled glovebox, and then transferred into the chamber of the XPS instrument. SEM characterizations were carried out using a Phenom-Pro scanning electron microscope installed inside an Ar-filled glovebox. The cross-section of the deposited Li 17 films were prepared by slowly tearing the electrodes. Using a razor blade to cut the electrodes is not advised because it easily deforms the soft Li metal and appears to "smooth out" the deposited Li films. Electrochemical measurements. Electrochemical performances were measured using CR2025 coin-type cells. The Li metal full batteries were assembled using a LiCoO2 cathode, a thin Li metal anode (20 or 50 µm in thickness) and one piece of polyethylene separator (Tonen), and <75 µL electrolyte per cell. Electrochemical tests were performed using battery cyclers (Shenzhen Neware, BTS4000-5V, 10/1.0 mA version, 0.05% current/voltage resolution). The coin cells were cycled inside temperature chambers set at 30 ± 0.2 °C. The lithium metal full batteries were first cycled at 0.1 C between 4.5-3.0 V for three cycles and then charged at 0.2 C and discharged at 0.5 C repeatedly. 1 C equals to ~3.7 mA cm-2. DFT simulations. Self-Consistent DFT calculations were performed using the real space projector- augmented wave method47,48 implemented in the GPAW code49,50 and employing the PBE exchange correlation functional51. We performed the DTF calculations on the Li(100), (110) and (111) surfaces. The Li surfaces comprised of four layers with the bottom two layers constrained at the bulk lattice constants. Each layer consisted of 3´3 Li unit cell. The solvent molecule along with Li + and PF6 - ions were placed on top of Li surface and the structure was allowed to relax to determine the decomposition. Periodic boundary conditions were used for x and y directions and a vacuum of 10 Å was used in the z direction perpendicular to the surface on both sides of the slab. A real-space grid spacing of 0.16 Å was used and the Brillouin zone was sampled using the Monkhorst Pack scheme52 with a k-point grid of (6´6´1). All simulations were converged to a force < 0.05 eV Å -- 1. Lastly Bader analysis53 was used to determine the amount of charge transferred from the lithium to the solvent during the decomposition and also volumes of the various decomposed species. Data availability. The supporting data for the included charts/graphs within this paper, as well as other findings from this study, are available from the corresponding author upon reasonable request. Acknowledgements. This work was supported by the Assistant Secretary for Energy Efficiency and Renewable Energy, Office of Vehicle Technologies of the US Department of Energy (DOE) through the Advanced Battery Materials Research (BMR) Program under contract no. DE-EE0007810 and supported by the start-up fund of Shanghai Jiao Tong University (to L.S.L). Author contributions. L.S.L., V.V., and Y.-M.C. conceived and supervised the research. Y.Y.Z. and L.S.L. performed the experiments with the help from M.S.P., B.H.W., D.W.. V.P. carried out the DFT simulations and analysis. L.S.L., V.P., V.V., and Y.-M.C prepared the manuscript with input from other co-authors. 18 Additional information. Supplementary information is available for this paper. Correspondence and requests for materials should be addressed to L.S.L ([email protected]), V.V. ([email protected]), or Y.-M.C. ([email protected]). Reference 1. Albertus, P., Babinec, S., Litzelman, S. & Newman, A. Status and challenges in enabling the lithium metal electrode for high-energy and low-cost rechargeable batteries. Nature Energy 3, 16-21 (2018). 2. Zhang, J.-G., Xu, W. & Henderson, W. A. Lithium metal anodes and rechargeable lithium metal batteries. 1 edn, 194 (Springer International Publishing, 2017). 3. Lin, D., Liu, Y. & Cui, Y. Reviving the lithium metal anode for high-energy batteries. Nature Nanotech. 12, 194 (2017). 4. Wood, K. N., Noked, M. & Dasgupta, N. P. Lithium Metal Anodes: Toward an Improved Understanding of Coupled Morphological, Electrochemical, and Mechanical Behavior. ACS Energy Lett. 2, 664-672 (2017). 5. Lin, D. et al. Layered reduced graphene oxide with nanoscale interlayer gaps as a stable host for lithium metal anodes. Nature Nanotech. 11, 626 (2016). 6. Yang, C.-P., Yin, Y.-X., Zhang, S.-F., Li, N.-W. & Guo, Y.-G. Accommodating lithium into 3D current collectors with a submicron skeleton towards long-life lithium metal anodes. Nature Commun. 6, 8058 (2015). 7. Bai, P. et al. Interactions between lithium growths and nanoporous ceramic separators. Joule 2, 2434- 2449 (2018). 8. Zhao, C.-Z. et al. An ion redistributor for dendrite-free lithium metal anodes. Sci. Adv. 4, eaat3446 (2018). 9. Liu, Y. et al. Making Li-metal electrodes rechargeable by controlling the dendrite growth direction. Nature Energy 2, 17083 (2017). 10. Zheng, J. et al. Electrolyte additive enabled fast charging and stable cycling lithium metal batteries. Nature Energy 2, 17012 (2017). 11. Ren, X. et al. Guided lithium metal deposition and improved lithium Coulombic efficiency through synergistic effects of LiAsF6 and cyclic carbonate additives. ACS Energy Lett. 3, 14-19 (2018). 12. Zhang, X.-Q., Cheng, X.-B., Chen, X., Yan, C. & Zhang, Q. Fluoroethylene Carbonate Additives to Render Uniform Li Deposits in Lithium Metal Batteries. Adv. Funct. Mater. 27, 1605989 (2017). 13. Qian, J. et al. Dendrite-free Li deposition using trace-amounts of water as an electrolyte additive. Nano Energy 15, 135-144 (2015). 14. Kim, M. S. et al. Langmuir -- Blodgett artificial solid-electrolyte interphases for practical lithium metal batteries. Nature Energy 3, 889-898 (2018). 15. Li, N.-W., Yin, Y.-X., Yang, C.-P. & Guo, Y.-G. An artificial solid electrolyte interphase layer for stable lithium metal anodes. Adv. Mater. 28, 1853-1858 (2016). 16. Tu, Z. et al. Fast ion transport at solid -- solid interfaces in hybrid battery anodes. Nature Energy 3, 310-316 (2018). 17. Liang, X. et al. A facile surface chemistry route to a stabilized lithium metal anode. Nature Energy 2, 17119 (2017). 18. Suo, L. et al. Fluorine-donating electrolytes enable highly reversible 5-V-class Li metal batteries. Proceedings of the National Academy of Sciences 115, 1156-1161 (2018). 19 19. Fan, X. et al. Highly Fluorinated Interphases Enable High-Voltage Li-Metal Batteries. Chem 4, 174- 185 (2018). 20. Qian, J. et al. High rate and stable cycling of lithium metal anode. Nature Commun. 6, 6362 (2015). 21. Han, X. et al. Negating interfacial impedance in garnet-based solid-state Li metal batteries. Nature Mater. 16, 572 (2016). 22. Duan, H. et al. Dendrite-free Li-metal battery enabled by a thin asymmetric solid electrolyte with engineered layers. J. Am. Chem. Soc. 140, 82-85 (2018). 23. Li, J., Ma, C., Chi, M., Liang, C. & Dudney, N. J. Solid Electrolyte: the Key for High-Voltage Lithium Batteries. Adv. Energy Mater. 5, 1401408 (2015). 24. Bai, P., Li, J., Brushett, F. R. & Bazant, M. Z. Transition of lithium growth mechanisms in liquid electrolytes. Energy Environ. Sci. 9, 3221-3229 (2016). 25. Kushima, A. et al. Liquid cell transmission electron microscopy observation of lithium metal growth and dissolution: Root growth, dead lithium and lithium flotsams. Nano Energy 32, 271-279 (2017). 26. Li, Y. et al. Atomic structure of sensitive battery materials and interfaces revealed by cryo -- electron microscopy. Science 358, 506-510 (2017). 27. Li, Y. et al. Correlating structure and function of battery interphases at atomic resolution using cryoelectron microscopy. Joule 2, 2167-2177 (2018). 28. Lu, Y., Tu, Z. & Archer, L. A. Stable lithium electrodeposition in liquid and nanoporous solid electrolytes. Nature Mater. 13, 961 (2014). 29. Markevich, E., Salitra, G., Chesneau, F., Schmidt, M. & Aurbach, D. Very Stable Lithium Metal Stripping -- Plating at a High Rate and High Areal Capacity in Fluoroethylene Carbonate-Based Organic Electrolyte Solution. ACS Energy Lett. 2, 1321-1326 (2017). 30. Lu, Y., Tu, Z., Shu, J. & Archer, L. A. Stable lithium electrodeposition in salt-reinforced electrolytes. J. Power Sources 279, 413-418 (2015). 31. Wu, F. et al. Lithium iodide as a promising electrolyte additive for lithium -- sulfur batteries: mechanisms of performance enhancement. Adv. Mater. 27, 101-108 (2015). 32. Yan, C. et al. Lithium nitrate solvation chemistry in carbonate electrolyte sustains high-voltage lithium metal batteries. Angew. Chem. Int. Ed. 57, 14055-14059 (2018). 33. Li, W. et al. The synergetic effect of lithium polysulfide and lithium nitrate to prevent lithium dendrite growth. Nature Commun. 6, 7436 (2015). 34. Ma, L., Kim, M. S. & Archer, L. A. Stable artificial solid electrolyte interphases for lithium batteries. Chem. Mater. 29, 4181-4189 (2017). 35. Ishikawa, M., Morita, M. & Matsuda, Y. In situ scanning vibrating electrode technique for lithium metal anodes. J. Power Sources 68, 501-505 (1997). 36. Kazyak, E., Wood, K. N. & Dasgupta, N. P. Improved Cycle Life and Stability of Lithium Metal Anodes through Ultrathin Atomic Layer Deposition Surface Treatments. Chem. Mater. 27, 6457-6462 (2015). 37. Liu, Y. et al. An artificial solid electrolyte interphase with high Li-Ion conductivity, mechanical strength, and flexibility for stable lithium metal anodes. Adv. Mater. 29, 1605531 (2017). 38. Weinstein, L., Yourey, W., Gural, J. & Amatucci, G. G. Electrochemical impedance spectroscopy of electrochemically self-assembled lithium -- iodine batteries. J. Electrochem. Soc. 155, A590-A598 (2008). 39. Behl, W. K. & Chin, D. T. Electrochemical overcharge protection of rechargeable lithium batteries: I. Kinetics of iodide/tri-Iodide/iodine redox reactions on platinum in solutions. J. Electrochem. Soc. 135, 16-21 (1988). 20 40. Behl, W. K. & Chin, D. T. Electrochemical overcharge protection of rechargeable lithium batteries: II. Effect of lithium iodide-iodine additives on the behavior of lithium electrode in solutions. J. Electrochem. Soc. 135, 21-25 (1988). 41. Behl, W. K. Anodic oxidation of lithium bromide in tetrahydrofuran solutions. J. Electrochem. Soc. 136, 2305-2310 (1989). 42. Xianming, W., Yasukawa, E. & Kasuya, S. Electrochemical properties of tetrahydropyran-based ternary electrolytes for 4 V Lithium metal rechargeable batteries. Electrochimica Acta 46, 813-819 (2001). 43. Yamaki, J.-I. The development of lithium rechargeable batteries. J. Power Sources 20, 3-7 (1987). 44. Zhang, S. S. Role of LiNO3 in rechargeable lithium/sulfur battery. Electrochimica Acta 70, 344-348 (2012). 45. Ding, F. et al. Effects of Carbonate Solvents and Lithium Salts on Morphology and Coulombic Efficiency of Lithium Electrode. J. Electrochem. Soc. 160, A1894-A1901 (2013). 46. Han, F. et al. High electronic conductivity as the origin of lithium dendrite formation within solid electrolytes. Nature Energy (2019). 47. Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 17953-17979 (1994). 48. Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758-1775 (1999). 49. Mortensen, J. J., Hansen, L. B. & Jacobsen, K. W. Real-space grid implementation of the projector augmented wave method. Phys. Rev. B 71, 035109 (2005). 50. Enkovaara, J. et al. Electronic structure calculations with GPAW: a real-space implementation of the projector augmented-wave method. J. Phys. Cond. Mater. 22, 253202 (2010). 51. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865-3868 (1996). 52. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations. Phys. Rev. B 13, 5188- 5192 (1976). 53. Tang, W., Sanville, E. & Henkelman, G. A grid-based Bader analysis algorithm without lattice bias. J. Phys. Cond. Mater. 21, 084204 (2009). 21 Supplementary Information for Design Principles for Self-forming Interfaces Enabling Stable Lithium Metal Anodes Yingying Zhu,ë, é Vikram Pande,^, é Linsen Li,ë, †, * Sam Pan,† Bohua Wen,† David Wang, † Venkatasubramanian Viswanathan,^, * Yet-Ming Chiang†, * ë Department of Chemical Engineering, Shanghai Jiao Tong University, Shanghai, China 200240 † Department of Materials Science & Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA ^ Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh PA 15213， USA é These authors contribute equally to the work. * Corresponding authors 1 2 Supplementary Figure 1 Decomposition of fluorinated compounds at Li surface predicted by DFT calculations. We show here the decomposition of 15 compounds. Most fluorinated compounds do lead to the formation of LiF, but there are some compounds which do not release the fluorine such as fluorinated epoxides, ethers and compounds with CF3 and CF2 groups. We also see that some compounds such as fluorinated DTD, fluorinated dioxane lead to the formation of additional LiF by decomposing the LiPF6 2- anion which we believe would salt. Lastly all sulfate and sulfur groups decompose readily to form SO2 ultimately lead to the formation of Li2SO3 and ROSO2Li. 3 Supplementary Figure 2 Decomposition of FEC in the presence of LiPF6 salt on Li (100), (110) and (111) surfaces. The pink atoms represent Li, green represent P and purple represent F. The decomposition products are identical in all cases. We suggest that the surface energy effects do not significantly affect the decomposition pathway for solvent decomposition on Li. We also see identical results for other solvents considered in this study. We hypothesize that the reason behind this is surface energy difference between different surface ~0.2-0.5 eV while the energy difference is an order of magnitude higher for the decomposition reaction ~2-5 eV. This shows that even during SEI healing, the exposed Li cracks will react to give similar reaction products assuming there are sufficient free Li atoms to complete the decomposition. In some cases, it is possible that due to passivation, the complete decomposition does not happen. 4 Supplementary Figure 3 Wide-scan XPS spectra of five different electrolytes. Wide-scan XPS spectra of the Li metal SEI formed in a. 1 M LiPF6 EC-DMC (EL-0), b. 1 M LiPF6 FEC-DMC (EL-1), c. 1 M LiPF6 DFEC-DMC (EL-2), d. 1 M LiPF6 CF3EC-DMC (EL-3), and e. 1 M LiPF6 FEC-DMC + DTD (EL-4). The fluorine content in the SEI increases in the order of EL-0, EL-4, EL-1, EL-4, and EL-2. 5 Supplementary Figure 4 Narrow-scan XPS spectra of S 1s and peak analysis. Narrow-scan XPS spectra of S 1s collected from the SEI formed on the Li electrode surface at the 1st charge, 20th charge, and 50th charge cycle. There is little S (0.1 at.%) in the SEI formed at the 1st charge. S content increases to 1.7 at.% and 2.2 at.% at the 20th and 50th charge, respectively. Based on peak analysis, there are three different S species, namely Li2SO4, ROSO2Li, and a small amount of Li2SO3. 6 Supplementary Figure 5 Decomposition of solvent molecules on Li(100) surface with one mono layer of LiF. The pink atoms represent Li, red represent O, grey represent C, green represent P, purple represent F, yellow represent S and white represent H. There is no chemical decomposition of the solvent in any of the cases. This is also validated from the electrons transferred from the Li slab which is less than 0.5 electrons as shown in Supplementary Table 1. Even in the case of DTD, the co-decomposition of DTD and LiPF6 is stopped due to the unavailability of Li. This clearly shows that a monolayer of LiF is sufficient to chemically passivate a surface from further solvent decomposition (chemically). This proves that in general increased fraction of LiF in the SEI will lead to a more compact and dense SEI. 7 Supplementary Figure 6 SEM characterization of the deposited Li film on the Li/Cu substrates from Li-Li cells. a, Top-view SEM images of the deposited Li film on the Li/Cu substrates (50 µm-thick Li, 15 µm-thick Cu) in 1 M LiPF6 EC-DMC (EL-0). The Li film was deposited in a Li-Li (thick Li foil vs 50 µm-thick Li on Cu) cell by charging at 0.42 mA cm-2 for 10 h. 4.2 mAh cm-2 of Li was deposited. More whisker-like Li particles were observed when a thick Li foil was used as the counter electrode instead of the roller-pressed LiCoO2 electrode shown in Figure 3 of the main text. b and c are the corresponding cross-sectional SEM images, showing three-layer structure consisting of the deposited Li, the original 50 µm-thick Li, and underlying Cu substrate. The deposited Li film consisted of whisker-like Li particles and had a wave-like surface, making it difficult to measure the thickness of the deposited Li film. The surface of the Li films deposited using LiCoO2-Li cells were more flat, as shown in Figure 3 of the main text. 8 Supplementary Figure 7 Electrochemical tests of LiCoO2-Li full batteries. a, The cycling performance of the LiCoO2-Li full cells using the EL-4 electrolyte (1 M LiPF6 FEC-DMC + 3wt% DTD) shown in comparison with those using the dual salt electrolyte (0.6 M LiTFSI + 0.4 M LiBOB in EC- EMC 4/6 by weight + 0.05 M LiPF6) previously reported in Zheng, J. et al. Nature Energy 2, 17012 (2017). The cells were tested under the same conditions. Data from two cells was shown in each case. The EL-4 electrolyte provides better cycling performance than the dual-salt electrolyte. b, Coulombic efficiency of the four cells. The EL-4 electrolyte provides higher CE than the dual-salt electrolyte. 9 Supplementary Tables Supplementary Table 1 Theoretical descriptors for solvent decomposition on a mono layer of LiF formed on Li(100) Solvent Electrons Transferred on LiF DMC EC FEC DiFEC CF3-EC DTD 0.3 0.2 0.2 0.2 0.1 0.3 10 Supplementary Table 2 Average Coulombic efficiency of the thin Li anodes in different electrolytes Lithium Salt Solvents & Additives (1 M) LiPF6 LiTFSI LiPF6 LiTFSI LiAsF6 LiAsF6 LiTFSI PC DOL-DME (1:1 v), 0.1 M LiI EC-EMC (1:1 v) EC-THP (1:1 v) EC-DMC (1:1 v) EC-2MeTHF (1:1 v) DOL-DME (1:1 v), 1 wt% LiNO3 CEavg 73.6% 83.0% 90.1% 95.5% 95.9% 96.4% 96.7% Test conditions: 0.6 mA cm-2 current density, 3.0 mAh cm-2 cycling areal capacity. PC = propylene carbonate, DOL = 1, 3-dioxolane, DME = 1,2-Dimethoxyethane, EMC = ethyl methyl carbonate, THP = tetrahydropyran, 2-MeTHF = 2-methyl tetrahydrofuran Supplementary Table 3 is a separate Microsoft Excel file. Supplementary Text DFT Calculation Details Self-Consistent DFT calculations were performed using the real space projector-augmented wave method[1,2] implemented in the GPAW code[3,4] and employing the PBE exchange correlation functional[5]. We chose the Li (100), (110) and (111) surfaces for the DFT calculations. The Li surface comprised of four layers with the bottom two layers constrained at the bulk lattice constants. Each layer consisted of 3x3 Li unit cell. The solvent molecule was placed on top of Li surface at a distance of 2 Å. We explored different conformers of the solvent molecule, chosen based on placing electronegative atoms such as F and O close to the surface. Li+ and PF# solvent molecule. LiPF6 salt was placed for decomposition studies as it is known that salt ions affect the $ ions were placed at a distance of 2 Å on top of the stability of solvent molecules by renormalizing the molecular energy levels of the solvent.[6,7] In addition, the salt ions may themselves participate in the reaction. It is worth highlighting that there are numerous possible configurations of the salt ions and solvent, we believe that given the consistency between the 11 structures, trends in reactivity are well captured with this approach. The internal coordinates of these structures were allowed to relax to determine the decomposition products. Periodic boundary conditions were used for x and y directions and a vacuum of 10 Å was used in the z direction perpendicular to the surface on both sides of the slab. A real-space grid spacing of 0.16 Å was used and the Brillouin zone was sampled using the Monkhorst Pack scheme[8] with a k-point grid of (6´6´1). The calculations were converged to < 5meV accuracy with respect to k-points and grid spacing. A Fermi-Dirac smearing of 0.05 eV All simulations were converged to a force < 0.05 eV Å -- 1. Bader analysis[9] was used to determine the amount of charge transferred from the lithium to the solvent during the decomposition and also volumes of the various decomposed species. Calculation of Bader Charges and Bader Volumes of different species To calculate the Bader charges and volumes, initially the electron density as a function of spatial coordinates was stored in a ".cube" file. The Bader analysis was performed on the ".cube" file. For charges, the Bader analysis was done by setting the vacuum charge density to zero. This was done to ensure that all charges are assigned to the appropriate molecular species. For calculating Bader Volumes of the atoms, the Bader analysis was done by setting the vacuum charge density to 0.0001 e Å -- 3. This was the error of the electron density in the DFT calculations performed. Thus, a cutoff lower than this value would not be consistent. A larger value for the vacuum charge density cutoff leads to incorrect assignment of the electrons to different atoms. We checked for some different values in the appropriate range and found that the Bader Volumes calculated scale with those chosen number, but the trends in the volumes of different species remain the same as shown in the table shown below. This means that the specie with the largest Bader volume is invariant implying that the descriptor used will provide the correct trend for the classification problem. The Bader charge transferred to the solvent while decomposition was calculated as negative of the charge on the Li(100) slab because the overall system is charge neutral. The decomposed species were identified by considering bond distances between two atoms. Two atoms were considered chemically bonded if the distance between them was less than 1.75 Å. Thus, all the decomposed species were identified and their charge and volumes calculated by summing over the charges and volumes of the individual atoms. 12 Solvent Bader Volume (Å3) for Cutoff Density 0.0001 (e/Å3) 0.0005 (e/Å3) 0.001 (e/Å3) DMC EC FEC DiFEC CF3-EC DTD 145 132 89 125 169 97 120 110 75 103 137 81 107 99 67 92 122 74 Passivation of Li(100) surface covered with LiF In order to explore the extent of passivation by LiF, we performed calculations by placing a monolayer of LiF on a 6 layer Li(100) slab. The structure was generated by placing F atoms on top of the Li surface. After relaxation, this spontaneously led to the formation of a LiF monolayer and 5 layers of Li(100). For simulating solvent decomposition on this structure, the bottom two layer were constrained to the bulk lattice constant of Li and the other layers and solvent molecule along with LiPF6 was allowed to relax. A similar Bader charge analysis was done to determine the charge transferred to the solvent. This shows us whether there is any Li oxidation and hence any possible reaction with the solvent. If the charge transferred is less than 0.5 e- then, this would indicate that the surface is passivated and no further reaction with the solvent is likely. 13 Calculation of Energy Density in Figure 1 LCO mass (mg) Separator mass (mg) Li (mg) Cu (mg) Al Electrolyte (mg) (mg) Total Mass (mg) Energy (Wh) Energy Density (Wh/kg) Total Volume (cm^3) Energy Density (Wh/L) Volumetric 29.13 3.80 1.35 8.51 2.57 4.18 49.55 0.02156 370 0.0178 1029 LCO-Li (20um Li) LCO-Li (50 29.13 3.80 3.38 8.51 2.57 4.18 51.58 0.02156 355 0.0216 848 um Li) LCO-Li (100 um Li) LCO-Li (250 um Li) LCO-Li (450 um Li) LCO-Li (750 um Li) LCO- Cu (anode free) 29.13 3.80 6.76 8.51 2.57 4.18 54.96 0.02156 333 0.0280 656 29.13 3.80 16.91 8.51 2.57 4.18 65.10 0.02156 281 0.0470 390 29.13 3.80 30.44 8.51 2.57 4.18 78.63 0.02156 233 0.0723 253 29.13 3.80 50.73 8.51 2.57 4.18 98.93 0.02156 185 0.1103 166 29.13 3.80 0.00 8.51 2.57 4.18 48.19 0.02156 380 0.0153 1199 Cathode Anode Electrolyte Current collector LCO Lithium EC-EMC Cu Al Separator Capacity [mAh/g] Density [g/cm3] porosity thickness [um] 185 3860 5 1.2 8.96 2.7 1.2 0.25 0 0.5 60 15 15 25 Packaging factor = 0.85 Electrode area = 1.267 cm2 14 Supplementary References [1] Blöchl, P. E., Projector augmented-wave method. Physical review B, 50(24), 17953 (1994). [2] Kresse, G., & Joubert, D., From ultrasoft pseudopotentials to the projector augmented-wave method. Physical Review B, 59(3), 1758 (1999). [3] Mortensen, J. J., Hansen, L. B., & Jacobsen, K. W., Real-space grid implementation of the projector augmented wave method. Physical Review B, 71(3), 035109 (2005). [4] Enkovaara, J. E., Rostgaard, C., Mortensen, J. J., Chen, J., Dułak, M., Ferrighi, L. & Kristoffersen, H. H., Electronic structure calculations with GPAW: a real-space implementation of the projector augmented- wave method. Journal of Physics: Condensed Matter, 22(25), 253202 (2010). [5] Perdew, J. P., Burke, K., & Ernzerhof, M., Generalized gradient approximation made simple.Phys. Rev. Lett., 77(18), 3865 (1996). [6] Kumar, N., & Siegel, D. J. (2016). Interface-induced renormalization of electrolyte energy levels in magnesium batteries. J. Phys. Chem. Lett., 7(5), 874-881. [7] Khetan, A., Luntz, A., & Viswanathan, V. (2015). Trade-offs in capacity and rechargeability in nonaqueous Li -- O2 batteries: Solution-driven growth versus nucleophilic stability. J. Phys. Chem. Lett.,, 6(7), 1254-1259. [8] Monkhorst, H. J., & Pack, J. D. Special points for Brillouin-zone integrations. Physical review B, 13(12), 5188 (1976). [9] Tang, W., Sanville, E., & Henkelman, G., A grid-based Bader analysis algorithm without lattice bias. Journal of Physics: Condensed Matter, 21(8), 084204 (2009). 15
1810.07293	1	1810	2018-10-16T22:03:47	Effect of Graded Bias Voltage on the Microstructure of arc-PVD CrN Films and its Response in Electrochemical & Mechanical Behavior	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The effect of graded or constant bias voltages (-40 V, -80 V and -40/60/80 V) on size grain and surface defects of arc PVD deposited CrN films was investigated. Corrosion resistance evaluated using electrochemical impedance spectroscopy (EIS) and potentiodynamic curves (Tafel) and the mechanical behavior evaluated by means of instrumented nanoindentation and scratch testing was correlated with the microstructural changes. It was found that the bias voltage variation affects corrosion behavior due to the presence of defects (i.e. open voids, droplets) which also affects the failure mechanisms and increasing spallation. High bias voltage (-80 V) increases nano-hardness and the elastic modulus due to the dense microstructure of the CrN coating.	physics.app-ph	physics	Effect of Graded Bias Voltage on the Microstructure of arc-PVD CrN Films and its Response in Electrochemical & Mechanical Behavior M.L. Cedeño-Ventéa, D.G. Espinosa-Arbeláeza, J. Manríquez-Rochab, G.C. Mondragón Rodrígueza, A.E. Gómez-Ovallea, J. González-Hernándeza, J.M. Alvarado-Orozcoa a Centro de Ingeniería y Desarrollo Industrial, Surface Engineering Dept., Av. Pie de la b Centro de Investigación y Desarrollo Tecnológico en Electroquímica, Pedro Escobedo, Cuesta 702, 76125 Santiago de Querétaro México. 76730 Santiago de Querétaro México. Abstract The effect of graded or constant bias voltages (-40 V, -80 V and -40/60/80 V) on size grain and surface defects of arc PVD deposited CrN films was investigated. Corrosion resistance evaluated using electrochemical impedance spectroscopy (EIS) and potentiodynamic curves (Tafel) and the mechanical behavior evaluated by means of instrumented nanoindentation and scratch testing was correlated with the microstructural changes. It was found that the bias voltage variation affects corrosion behavior due to the presence of defects (i.e. open voids, droplets) which also affects the failure mechanisms and increasing spallation. High bias voltage (-80 V) increases nano-hardness and the elastic modulus due to the dense microstructure of the CrN coating. Keywords: Chromium nitride, graded voltage, microstructure, defects, corrosion. 1. Introduction Chromium nitride coatings deposited by physical vapor deposition technologies (PVD) have been wide interest owing to the ability to improve corrosion, oxidation and wear resistance; as well as the low friction coefficient [1]. For those reasons, it is a suitable candidate for uses in mechanical devices, cutting and forming tools, die casting molds and plastic extrusion [2]. Some PVD technology variants such as cathodic arc (arc-PVD) and magnetron sputtering (MS) have been widely used to deposit CrN. Particularly, the cathodic arc has gained importance for the high deposit speeds and excellent substrate adhesion [3]. However, during the deposition process, unavoidable defects are generated which reduce the mechanical properties and the corrosion resistance of the coatings. These defect such as open voids, pinholes, cone-like defects and droplets act as ways for the development and propagation of corrosion phenomenon. For example, the pinholes derived from the shadowing effect and the voids caused by the removal of particles weakly bound to the coating can act as routes of diffusion of corrosive media towards the substrate, allowing the degradation and subsequent delamination [4], also it reduces the densification of the film, which leads to a decrease in the mechanical performance of the piece. On the other hand, droplets generated by molten cathode material can decrease coating adhesion, increasing the roughness and porosity or acting as abrasive particles in the coating [5]. Additionally, these defects can accelerate the substrate corrosion by the generation of localized galvanic corrosion [6,7]; however, their effects are sometimes controversial [8][9]. To avoid the droplets inconvenience, various plasma filtering techniques have been developed, but the up-scalability of many filters is not easy to get due to their geometry. Additionally, the effect of process parameters has been considered as a good way for getting this aim. An appropriate control of parameters as cathode voltage, substrate temperature, gases ratio, bias voltage improve coatings performance and they can reduce defects density. In CrN films an increase in N2 pressure produce phase's transformation, changes in preferential orientation and it can increase macroparticles size and numbers [10]. The bias voltage, which is applied to the substrate affects surface morphology, the mechanical and electrochemical properties of films. Constant voltages higher reduce size and number of droplets, this phenomenon can be originated for two reasons: 1) a higher bias voltage produces the accelerated particles have higher momentum and more macroparticles are removed of the coating. 2) The electrical repulsion force due to change the charging state of the macroparticles the higher the bias voltage [11]. Recently, graded bias voltage has been implemented to find deposition parameters that assure improved coating properties, however, there are few studies about graded bias in nitride films and its effect on microstructure and corrosion resistance of coatings. Shuyong Tan [12] found the hardness of bias-graded films is between that of films with the low and high constant voltage. Also, the graded bias voltage leads to the modification of the morphology and microstructure of the CrN film, moving from a pyramidal or cellular conical structure to a mixture of both. In this work, we study the influence of constant and graded bias voltage on the microstructure and formation of surface defects and their relationship with electrochemical behavior and mechanical in CrN coating systems. 2. Experimental Methodology 2.1. Coatings deposition CrN thin films were deposited by arc sputtering using Oerlikon Balzers PVD equipment, model DOMINO MINI MZR 303/4/46. Chromium targets with 99.95% of purity were used. Substrates M2 steel were polished to a roughness of Ra 0.025 micrometer and they were chemically cleaned with acetone. Finally, the cleaning was finished with an ions cleaning process into the chamber using a working pressure of 1x10-2 mbar and bias voltage -250 V for 30 min. A reactive nitrogen discharge using a flow of 500 sccm was used to deposit CrN coatings. The current and temperature of the substrate were kept at 150 A and 400 °C, respectively. To study the effect of the bias voltage, the substrates were polarized with constant voltages of -40V and -80V. A graded bias voltage of -40/-60/-80V was also applied. The remaining experimental conditions were maintained constant for all samples. 2.2. Characterization of films Microstructure, morphology, and chemistry The crystallite size and crystal phases of the samples were analyzed by X-ray diffraction (XRD) using a SmartLab diffractometer (Rigaku), with a Cu Kα radiation (λ = 0.15405 nm). Grazing incidence scans in the range 30° -- 90° at 1° incidence, with a step size of 0.02° were used. The coating thickness and composition as a function of the deep were measured with glow discharge optical emission spectrometry (GD-OES, Horiba) equipment which was operated at 650 Pa and 35W of pressure and power, respectively. The GD-OES data was complemented by a scanning electron microscope (SEM) and energy dispersive spectroscopy (EDS). The samples microstructure was studied in a Helios Nanolab 600 SEM using an electron secondary detector at 5 kV of accelerating voltage. The surface roughness of coated and uncoated specimens was studied using a Dektak Xt-A surface profilometer (Bruker, Billerica, MA, USA) with a conical tip of 2 μm radius and a scan rate of 66 µm/s. Optical microscopy was carried out at the surface of the specimen using a VHX-5000 digital (Keyence). Images at 500X magnifications were used to determine the number and average size of defects. Optical images with the same properties were analyzed by ImageJ software program and a complementary analysis with 3D profilometry images was made on 500µm x 500µm. Mechanical properties A scratch tester (Anton Paar) with a 200 μm diamond tip was used to study the film adhesion. The tests were carried out with progressive loads from 1 N to 35 N at a loading speed of 68 N/min. The cohesive and adhesive critical loads were identified through the correlation between the signal of an acoustic sensor and the images of the scratch tracks. Failure mechanisms were identified and studied by scanning electron microscopy. Oliver and Pharr method was used to evaluate the hardness and reduced elastic modulus using a Nanoindentation Hysitron TI-700 UBI 1 with a Berkovich (BK2) indenter at a normal load of 6 mN for 10 s. 60 nanoindentations were performed per sample and Weibull statistics was used to analyze the experimental data. Corrosion testing Corrosion tests were carried out using a BioLogic potentiostat-galvanostat SP300. A typical three-electrode cell was used with a platinum mesh counter electrode and an Ag/AgCl reference electrode. The coated and uncoated M2 steels were employed as a working electrode with a 0.95 cm2 working area. The electrolyte consisted of a 3.5% NaCl solution in deionized water at room temperature (25 °C). Potentiodynamic polarization tests were obtained with 2 of mV/s scan rate and potentials between -1 V to +1 V with respect to the open circuit potential. The Ecorr parameter, anodic (βa) and cathodic (βc) slopes, and the corrosion current density (icorr) were extracted from Tafel plots. The protective efficiency (Pe) was determined from the corrosion current densities of the coatings (𝐼𝑐𝑜𝑟𝑟) and the substrate (𝐼𝑐𝑜𝑟𝑟 ). 0 𝑃𝑒(%) = (1 − 𝐼𝑐𝑜𝑟𝑟 0 ) × 100 𝐼𝑐𝑜𝑟𝑟 (1) 3. Results and discussion Microstructure, morphology and chemistry Figure 1 shows XRD patterns of CrN coatings which exhibit diffraction peaks corresponding to the B1 type structure of CrN phase (ICSD card # 01-083-5612, in figure 1a) with strong reflection in (200), as can be seen in figure 1b. The samples were compared with previously reported of unstrained CrN and they are presented in table I. A compressive stress state are present in the samples under study which are according to the shifting (200) peak to higher angles observed for 40Vb, 40/60/80Vb, and 80Vb samples. This behavior is related to the residual stresses due to atomic peening effect (APE) during the film growth [13]. A higher lattice strain was found from the 80Vb specimen caused by APE that promotes changes in nucleation and growth kinetics. Table I shows that crystallite size increases with the corresponding increment of the bias voltage. It is due to the enhanced mobility of adatoms on the surface and hence the growth and coalescence of nuclei were accelerated [14]. Likewise, it was observed a correlation between Bias Voltage and coating thickness finding a less thickness and major density of the coating for 80Vb [15]. a) b) Figure 1. XRD patterns of CrN films: a) at 40 Vb, 40/60/80 Vb and 80 Vb, and b) zoom in of (200) peak. The GDOES measurements show the N/Cr ratio in the CrN films was reduced with the increase in negative bias voltage [15]. The reduction of N/Cr ratio was related to the re- sputtering phenomenon of the N atoms which is originated when the ions impinging increase their energy. The high-energy ions selectively sputter the N atoms because they are lighter than Cr atoms. 30405060708001234CrN (222)CrN (311)CrN (220)CrN (200)CrN (111) ICSD # 01-083-5612 80 V 40/60/80 V 40 VIntensity (a.u)2(deg.)4243444546CrN (200) ICSD # 01-083-5612 80 V 40/60/80 V 40 VIntensity (a.u)2(deg.)Table I. Crystallographic parameters and structure properties Crystallite Thickness Lattice strain (%) 35.000 39.000 60.000 Lattice constant (Å) 4.1660 4.1637 4.1656 Unstrained 4.1668 Specimen 40Vb 40/60/80Vb 80Vb ICSD card # 01-083- 5612 size (Å) 179 220 232 - (µm) 5.060 5.010 4.020 N/Cr (%wt) 0.166 ± 0,004 0.219 ± 0,004 0.102 ± 0,001 - - Figure 2 shows the roughness values obtained by profilometry. The films roughness and defects percent have a clear tendency to be reduced for coatings obtained with a higher bias voltage; such as is previously reported by Warcholinski et al. who found a decrease of the microparticles density when the bias voltage increase [17]. However, we found an increase in the particle average size that could be related to higher arcs formation during the process. Figure 2. Relationship between size and defects percentage with the bias voltage. A detailed analysis of the SEM micrographs show a honeycomb structure at CrN films similar behavior was found by some authors for CrN coatings surface, a deposited by arc-PVD (see Figure 3) [17,19]. The honeycomb structure is Open void more defined in the 40/60/80Vb coating due to the gradual acceleration and momentum that suffer the ions and particles during the deposit process. On the other hand, flakes-like defects were observed that could arise from particles detachment of the chamber walls. Additionally, open void defects were found in all samples. This kind of defects are widely known and it is mainly due to a droplets removal or others particles deposited on the substrate surface. [20]. Figure 3. SEM micrographs of CrN coatings: a) 40Vb, b) 40/60/80 Vb, and c) 80 Vb. Mechanical properties Figure 3 show hardness (H), reduce elastic modulus (E), and the H3/E2 ratio of the CrN coatings. An increase 19.74 GPa to 25.38 GPa in hardness and 245.66 GPa to 293.41 GPa in the reduced elastic modulus was reached when bias voltage shift from 40 to 80 V. Values higher obtained to 80 V can be attributed to more dense microstructure and major lattice strain, which indicates compressive stresses that enhance the mechanical resistance of the system [21][22][23][24]. On the other hand, intermediate values of H and E were obtained with 40/60/80 V due to a combination of mechanical properties between 40Vb and 80V. Additionally, It is important to consider that the values found have been higher to reported by Tan [12], who obtained a hardness between 7 GPa to 15 Gpa with CrN coatings deposited with a graded bias voltage (from -20 V to -200 V with intervals of 5V, 10V, and 60V). Besides, figure 3a shows the H3/E2 ratio of CrN coatings deposited at various voltages. This ratio represents the plastic deformation resistance of the coatings, which is related to the contact behavior of films. Higher values indicate the ability of the coating to absorb more energy. The behavior found is similar to the obtained with the hardness, in this case, the H3/E2 shows a growth from 113.45 GPa to 189.96 GPa with increasing of V, thus, this coating should be more rigid to the contact compare with the others. FlakeOpen voidDropletDropleta)b)c) a) b) Figure 4. Effect of bias voltage on mechanical properties of CrN coatings: a) H, E and H3/E2; b) Critical loads obtained by scratch tester. Figure 4b shows the critical loads cohesive Lc1 (black) and adhesive Lc2 (red) related with resistances to initiation of cracks and to delamination. LC1 is influenced by H3/E2 ratio, thus, its behavior is similar to figure 3, where higher values were obtained with coatings deposited to 80 V. However, different behavior was found with LC2, in this case, the 40/60/80 V presents major interfacial resistance followed by coatings of 40 V and 80 V respectively. The phenomenon is related to the residual stresses observed by DRX, due to the high compressive stress in coatings with 80 V, the adhesion of the system of substrate/coating is reduces 4040/60/808010012014016018020020222426 H(GPa)220240260280300E(GPa) (H3/E2)*103Voltage (V)4040/60/808010152025 Lc1 Lc2Critical load (N) Voltage (V)[11][12]. On the other hand, the graded microstructure obtained with 40/60/80Vb diminishes the stress state produced by the mechanic contact. Figure 5 shows the scratch tracks of the coatings. It was observed that at a load lower than LC1 the scratch track presents lateral cracks, corresponding to cohesive failures. When the load slightly overpasses LC1, recovery spallation at the border of the scratch crack and arc tensile cracks are formed, which are more marked by the increase of the load. After the Lc2 load, the scratch track shows buckling spallation. The buckling spallation is more severe 40/60/80V coatings likely by their honeycomb structure more pronounced. CrN 40 V CrN 40/60/80V CrN 80V Figure 5. Scratch tracks of the coatings. LC1-Black line; LC2- Red line Figure 6 shows the failure mechanism of the coatings. It can be seen that cohesive failures start with lateral cracks, which bordered droplets and they cross defects as voids and adhered weak particles. These cracks increase the depth when the load is rising up and they can interact with each other for originating recovery spallation at the border of the scratch crack (with values majors or equal to LC1), which act as cluster start point of buckling spallation. On the other hand, arc tensile cracks which are simultaneously generated with recovery spallation can aid to crack proliferation at end of the scratch track; this phenomenon is more notable when the honeycomb structure is more pronounced (40/60/80V). Additionally, the mechanism of failure depends on the H3/E2 ratio and the residual stress of films. For that reasons, at the scratch track end, the films deposited with 80 V has minor amount spallations. Figure 6. Failure mechanism of the coatings (CrN 80V). Electrochemical properties Figure 7 shows the potentiodynamic polarization curves of M2 steel without and with the coating, in presence of chlorides and at room temperature. The electrochemical parameters obtained by Tafel analysis are detailed in Table II. Table II. Electrochemical parameters Material Ecorr (mV) Icorr (µA) βa βc (mV/decade) (mV/decade) Acero M2 CrN 40V CrN 40/60/80 -619,87 ± 1,25 514,09 ± 7,47 371,89 ± 13,14 48,55 ± 82,1 ± 410,90 ± 0,05 ± 0,01 0,25 ± 0,04 122.70 ± 0.01 188,10 ± 13.54 200,05 ± 37.54 185,15 ± 10.96 Protection efficiency (%) N.A. 99.88 94.87 V CrN 80 V -263,48 ± 4,70 2,48 ± 0,44 249,8 ± 62.49 115,35 ± 10.39 99.48 Where, Ecorr= corrosion potential, Icorr= corrosion current density, βa= anodic slope, βc= cathodic slope. It was observed that M2 steel presents potential and corrosion current values of -619.87 mV and 48.55 µA respectively. In the anodic zone, its active region in the increases fasts with the applied potential until 0.3 V in which starting the formation of unstable and un-protective iron oxide layer, which allows the dissolution of the metal to continue taking place. On the other hand, the CrN improves the corrosion resistance of M2 steel, and this is higher when reducing the bias voltage. That is showed with the increase in the corrosion potential and corrosion current lower values that were obtained with CrN 40Vb. In all cases coated steel exhibits a passive behavior, and the anodic current density increases slower than M2 steel, as shown by the larger value of the slopes of the anode region (Table II). Besides, all coatings have protection efficiency upper 90%. Figure 7. Polarization curves of un-coating and coating steel. CrN coatings obtained to 40/60/80 V and 80 V have passive zones between potentials of - 0.077 and 0.69 V derivated of the formation of protective oxide of Cr2O3. Besides, to potentials a little lower than transpassive potentials there are fluctuations in current due to an increase of dissolution points generated by pitting corrosion that the steel experiments when defects are removed of the coating. In films obtained to 40 V the pitting potential was not observed, only one active region could be identified. However, these films present the lower corrosion currents and the higher corrosion potentials, that indicated that this coating improves greater the corrosion resistance of M2 steel. Analysis of the surface corroded and corrosion mechanism Scanning electron microscope was used to analyze the surface of the specimens after the corrosion process. It was observed that the specimens are susceptible to medium chloride and present corrosion. A lower damage was observed in coated-steel. Figure 8 shows that M2 steel basically presents uniform corrosion, while that in coated-steel the localized corrosion is prevalent. In coated samples, the corrosion process is dependent on the voltage. As shown in figure 6 coatings deposited at 80 V had larger pitting and higher iron dissolution, while the corrosion process of CrN 40 V and CrN 40/60/80 V as was slower. 10-610-510-410-310-210-1100101-1,0-0,50,00,51,0 40V 40/60/80V 80V M2 Ewe/V (V)log (I/mA) (mA)Figure 8. SEM images of the surface after corrosion test. The corrosion phenomenon is triggered by the diffusion of the electrolyte in the porosities of the film, derived from defects formed during the deposition process. Pinholes can act as routes of diffusion of corrosive media towards the substrate. We found that the size of the droplets related to the corrosion mechanism, larger sizes generate a poor adhesion of the particle with the coating and can increase the residual stresses in the film favoring the separation of this inside the coating and subsequently the generation of spaces through which the electrolyte can penetrate, as shown in the figure 9. For that reason, CrN 80 V presented the lowest corrosion resistance. Figure 9. Corrosion mechanism of the coatings. Besides, with the increase of the potential, the droplet detachment is produced and pore formation. The pores provide the direct diffusion ways for electrolyte and generate pitting corrosion. Conclusions The effect of constant and graded bias voltage (40V, 80V, and 40/60/80V) in corrosion and mechanical properties were studied. We showed that the increase of bias voltage enhances the hardness and elastic modulus due to major lattice strain and compressive stresses in the film. However, the adhesion had an opposite behavior where CrN 40/60/80V and CrN 40V presented major critical load LC1. The better adhesion observed with CrN 40/60/80V was attributed to the graded microstructure, which diminishes the stress state produced by the mechanic contact. In all cases, CrN coatings improved corrosion resistance of M2 steel with a protection efficiency upper 90%. The corrosion resistance was increased when the bias voltage was reduced to 40V, and the corrosion mechanism was major dominated by localized corrosion. The phenomenon was related to the defect sizes. A major size increase the corrosion current and shift down the potential, reducing the corrosion resistance. References [1] K.C. Mutyala, E. Ghanbari, G.L. Doll, Effect of deposition method on tribological performance and corrosion resistance characteristics of Cr x N coatings deposited by physical vapor deposition, Thin Solid Films. 636 (2017) 232 -- 239. B. Navinšek, P. Panjan, I. Milošev, Industrial applications of CrN (PVD) coatings, [2] deposited at high and low temperatures, Surf. Coat. Technol. 97 (1997) 182 -- 191. I.G. Brown, CATHODIC ARC DEPOSITION OF FILMS, Annu. Rev. Mater. Sci. [3] 28 (1998) 243 -- 269. S.H. Ahn, J.H. Lee, J.G. Kim, J.G. Han, Localized corrosion mechanisms of the [4] multilayered coatings related to growth defects, Surf. Coat. Technol. 177-178 (2004) 638 -- 644. P.J. Kelly, R.D. Arnell, Magnetron sputtering: a review of recent developments and [5] applications, Vacuum. 56 (2000) 159 -- 172. [6] D.B. Lewis, S.J. Creasey, C. Wüstefeld, A.P. Ehiasarian, P. Eh. Hovsepian, The role of the growth defects on the corrosion resistance of CrN/NbN superlattice coatings deposited at low temperatures, Thin Solid Films. 503 (2006) 143 -- 148. U.K. Wiiala, I.M. Penttinen, A.S. Korhonen, J. Aromaa, E. Ristolainen, Improved [7] corrosion resistance of physical vapour deposition coated TiN and ZrN, Surf. Coat. Technol. 41 (1990) 191 -- 204. S.-G. Wang, X.-D. Bai, B.-C. Wang, Y.-D. Fan, The positive roles of metallic droplets [8] in deposition of alloy films by cathodic arc plasma deposition, J. Mater. Res. 11 (1996) 1137 -- 1143. [9] H.W. Wang, M.M. Stack, S.B. Lyon, P. Hovsepian, W. -D Münz, The corrosion behaviour of macroparticle defects in arc bond-sputtered CrN/NbN superlattice coatings, Surf. Coat. Technol. 126 (2000) 279 -- 287. [10] M.-S. Li, L.I. Ming-sheng, C.-J. Feng, F.-H. Wang, Effect of partial pressure of reactive gas on chromium nitride and chromium oxide deposited by arc ion plating, Trans. Nonferrous Met. Soc. China. 16 (2006) s276 -- s279. [11] X.S. Wan, S.S. Zhao, Y. Yang, J. Gong, C. Sun, Effects of nitrogen pressure and pulse bias voltage on the properties of Cr -- N coatings deposited by arc ion plating, Surf. Coat. Technol. 204 (2010) 1800 -- 1810. [12] S.Y. Tan, X.H. Zhang, X.J. Wu, F. Fang, Effect of Graded Bias on Microstructure and Hardness of CrNX Films, Adv. Mat. Res. 154-155 (2010) 1476 -- 1480. [13] B.-Y. Shew, J.-L. Huang, D.-F. Lii, Effects of r.f. bias and nitrogen flow rates on the reactive sputtering of TiA1N films, Thin Solid Films. 293 (1997) 212 -- 219. [14] M. Marinov, Effect of ion bombardment on the initial stages of thin film growth, Thin Solid Films. 46 (1977) 267 -- 274. [15] Y. Chunyan, T. Linhai, W. Yinghui, W. Shebin, L. Tianbao, X. Bingshe, The effect of substrate bias voltages on impact resistance of CrAlN coatings deposited by modified ion beam enhanced magnetron sputtering, Appl. Surf. Sci. 255 (2009) 4033 -- 4038. [16] Q.M. Wang, K.H. Kim, Effect of negative bias voltage on CrN films deposited by arc ion plating. II. Film composition, structure, and properties, J. Vac. Sci. Technol. A. 26 (2008) 1267 -- 1276. [17] B. Warcholinski, A. Gilewicz, J. Ratajski, Z. Kuklinski, J. Rochowicz, An analysis of macroparticle-related defects on CrCN and CrN coatings in dependence of the substrate bias voltage, Vacuum. 86 (2012) 1235 -- 1239. [18] P. Panjan, M. Čekada, M. Panjan, D. Kek-Merl, Growth defects in PVD hard coatings, Vacuum. 84 (2009) 209 -- 214. [19] Y. Ye, Y. Wang, H. Chen, J. Li, Y. Yao, C. Wang, Doping carbon to improve the tribological performance of CrN coatings in seawater, Tribol. Int. 90 (2015) 362 -- 371. [20] P. Panjan, D. Kek Merl, F. Zupanič, M. Čekada, M. Panjan, SEM study of defects in PVD hard coatings using focused ion beam milling, Surf. Coat. Technol. 202 (2008) 2302 -- 2305. [21] Y.X. Ou, J. Lin, S. Tong, H.L. Che, W.D. Sproul, M.K. Lei, Wear and corrosion resistance of CrN/TiN superlattice coatings deposited by a combined deep oscillation magnetron sputtering and pulsed dc magnetron sputtering, Appl. Surf. Sci. 351 (2015) 332 -- 343. [22] G.-H. Song, L.U.O. Zhuo, L.I. Feng, L.-J. Chen, H.E. Chun-lin, Microstructure and indentation toughness of Cr/CrN multilayer coatings by arc ion plating, Trans. Nonferrous Met. Soc. China. 25 (2015) 811 -- 816. [23] R. Bayón, A. Igartua, X. Fernández, R. Martínez, R.J. Rodríguez, J.A. García, A. de Frutos, M.A. Arenas, J. de Damborenea, Corrosion-wear behaviour of PVD Cr/CrN multilayer coatings for gear applications, Tribol. Int. 42 (2009) 591 -- 599. [24] X. Chu, S.A. Barnett, Model of superlattice yield stress and hardness enhancements, J. Appl. Phys. 77 (1995) 4403 -- 4411.
1810.06863	1	1810	2018-10-16T08:00:10	On sound propagation in three-dimensional poroelastic field: correct field expressions	[ "physics.app-ph" ]	The correct three-dimensional poroelastic field expressions, under monochromatic harmonic incidence, were revealed in this paper. They were confirmed satisfactory using the results got by other researchers. Former incomplete expressions were found to overestimate the random sound transmission loss and lead to division-by-zero, though their results were amply acceptable.	physics.app-ph	physics	On sound propagation in three-dimensional poroelastic field: correct field expressions Hou Qiaoa, Zeng Hea,b, Wen Jianga,b,∗, Weicai Pengc aDepartment of Mechanics, Huazhong University of Science &Technology, Wuhan, China bHubei Key Laboratory for Engineering Structural Analysis and Safety Assessment, Huazhong University cNational Key Laboratory on Ship Vibration and Noise, China Ship Development and Design Center, of Science & Technology, Wuhan, China Wuhan, China Abstract The correct three-dimensional poroelastic field expressions, under monochromatic harmonic incidence, were revealed in this paper. They were confirmed satisfactory us- ing the results got by other researchers. Former incomplete expressions were found to overestimate the random sound transmission loss and lead to division-by-zero, though their results were amply acceptable. Keywords: Poroelastic field; Sound transmission; Correct field expressions; Three-dimensional 1. Introduction Recently, much interest has been shown on modeling sound absorbing materials, especially poroelastic materials, i.e., porous materials which have an elastic frame. The Biot theory [1], which is the most widely used modeling technique for these materials, is studied extensively during the last decades. Despite the fruitful numerical developments [2, 3, 4] in Biot theory, the related theoretical study is scarce and mainly about the two-dimensional (2D) problems [5, 6]. Closed-form 2D poroelastic field expressions, which are among the most notable progress, were obtained by Bolton [6] using the simplifications of Deresiewicz [5] and ∗Corresponding author Email address: [email protected] (Wen Jiang) 1 Allard [7]. When it comes to the three-dimensional (3D) case, the theoretical develop- ment is insufficient until the work by Zhou [8], where he obtained the 3D poroelastic field expressions. These field expressions have been cited extensively [9, 10] since then. However, the 3D poroelastic field expressions are not sufficiently revealed by now. For example, in the work by Zhou [8, 11], he used an inappropriate hypothesis (refer to Equation (13) in [11]), which leads to an incomplete set of expressions. It can be erroneous if the correct expressions are not revealed. The aim of this discussion is to provide their formal derivation, and the correct expressions. The field expressions obtained here, with satisfying consistency with the theoretical results obtained by other researchers, are confirmed adequate in 3D poroelastic applications. 2. Derivations With the time dependence ejωt (j = solid displacement us and fluid displacement u f are [12] −ω2(cid:16) ∗ ∗ 11us + ρ ρ 12u f(cid:17) −ω2(cid:16) √−1), the poroelastic equations formulated by 12us(cid:17) = (A + N)∇∇ · us + N∇2us + Q∇∇ · u f = R∇∇ · u f + Q∇∇ · us ∗ ∗ 22u f + ρ ρ 11, ρ∗ 12, ρ∗ (1) (cid:113) where the porous parameters ρ∗ 22, A, N, Q, R can be found in [6, 8, 13]. When two scalar potentials ϕs = ∇·us, ϕ f = ∇·u f and two vector potentials Ψs = ∇×us, Ψ f = ∇×u f are introduced, the above equations can be reduced to a fourth-order scalar wave equation and a second-order vector equation [6]. The three propagation wavenumbers corresponding to the wave equations are 2} = A1/2 ± 11R− 2ρ∗ {k2 1, k2 (2) 12)/(PR− Q2), where A1 = ω2(ρ∗ P = A + 2N; k1, k2 correspond to the scalar wave equation, while k3 correspond to the vector wave equation. 11 − ρ 1/4 − A2, k2 ∗ ∗ ∗ ∗ = ω2/N(ρ A2 22) 12ρ 12/ρ 22 − ρ∗ 22P)/(PR− Q2), A2 = ω4(ρ∗ 12ρ∗ 11ρ∗ 12Q + ρ∗ 3 When a 3D monochromatic incident wave Φi = ejωt−jkr, where k = (kx, ky, kz), r = (x, y, z), transmits through the porous domain along its normal direction z (the time 2 dependence ejωt is omitted henceforth), the solid phase strain es = ϕs and Ωs = Ψs are C1e−jk1zz + C2ejk1zz + C3e−jk2zz + C4ejk2zz(cid:17) es = e−j(kx x+kyy)(cid:16) (cid:20)(cid:16) (cid:16) C7e−jk3zz + C8ejk3zz(cid:17) C5e−jk3zz + C6ejk3zz(cid:17) (cid:21) (cid:16) C9e−jk3zz + C10ejk3zz(cid:17) ix + iz + iy Ωs = e−j(kx x+kyy) (cid:113) i − k2 k2 x − k2 y (i=1,2,3); ix, iy, iz are the unit vectors along x, y, and z; the where kiz = unknown constants C1 −C10 are not independent and they can be determined by proper boundary conditions. Subsequently, the fluid phase strain e f = ϕ f and Ω f = Ψ f are e f = e−j(kx x+kyy)(cid:16) b1C1e−jk1zz + b1C2ejk1zz + b2C3e−jk2zz + b2C4ejk2zz(cid:17) (3) (4) (5) (6) (7) (8) (9) Ω f = gΩs where b1, b2, g can be found in [6, 8]. Bolton [6] pointed out that, in 2D case, all displacement field variables should contain contributions from all the six wave components. Accordingly, in the 3D case here, the components of the solid phase displacement us = (us y, us z) are (cid:18) x, us x = e−j(kx x+kyy) D1e−jk1zz + D2ejk1zz + D3e−jk2zz us (cid:18) + D4ejk2zz + D5e−jk3zz + D6ejk3zz D7e−jk1zz + D8ejk1zz + D9e−jk2zz (cid:18) + D10ejk2zz + D11e−jk3zz + D12ejk3zz D13e−jk1zz + D14ejk1zz + D15e−jk2zz + D16ejk2zz + D17e−jk3zz + D18ejk3zz (cid:19) (cid:19) (cid:19) z = e−j(kx x+kyy) us y = e−j(kx x+kyy) us where D1 − D18 are the contribution coefficients of the six wave components to the dis- placement components. These coefficients are not independent and they are expressed by C1 − C10 subsequently. Corresponding results of the fluid phase, which can be obtained in a similar manner as the solid phase, are not discussed in detail herein. As the unknown constants C1 − C10 are not independent, the condition ∇ · Ωs = 0 [14] which can uniquely determine us from es and Ωs, is used first to eliminate the 3 dependent constants among them. Subsequently, three different cases are identified according to their incident wavenumbers. The first case, when the incident wave is not in the x-z plane (ky (cid:44) 0), leads to C7 = − kx ky C5, C8 = − kx ky C6, C9 = C10 = 0 The second case, when the incident wave is not in the y-z plane (kx (cid:44) 0), leads to C5 = − ky kx C7, C6 = − ky kx C8, C9 = C10 = 0 (10) (11) The third case, when the incident wave is along the z axis (normal incidence, kx = ky = 0), leads to C5 = C6 = C7 = C8 = C9 = C10 = 0 (12) The above cases, to be brief, are denoted as incident case 1 (ky (cid:44) 0), incident case 2 (kx (cid:44) 0) and incident case 3 (kx = ky = 0) in the following. These cases are not mutually exclusive; however, they are separated for mathematical convenience. It can be concluded from Eqs.(10)-(12) that there are six independent constants at most in the 3D poroelastic field here. According to Eqs.(3)-(4), (7)-(12), the poroelastic field expressions can be ex- pressed in closed form using four or six independent constants. Once the field ex- pressions are obtained, the independent unknown constants, along with the poroelastic problem, can be solved by proper boundary conditions. 2.1. Incident case 1 When ky (cid:44) 0, according to Eq.(10), the solid phase strain Ωs is Ωs = e−j(kx x+kyy)(cid:16) C5e−jk3zz + C6ejk3zz(cid:17)(cid:32) (cid:33) iy ix − kx ky z ]T is y , u f Subsequently, the displacement u = [us y, us x, us x , u f z, u f u = e−j(kx x+kyy)Y1EC where E = diag(e−jkm 1zz, ejkm 1zz, e−jkm 2zz, e−jkm C = [C1, C2, C3, C4, C5, C6]T 2zz, ejkm 3zz, ejkm 3zz) 4 (13) (14) (15) (16) Here, the matrix E is a 6 × 6 diagonal matrix. The non-zero elements of the coefficient matrix Y1 are provided in Appendix A. 2.2. Incident case 2 When kx (cid:44) 0, according to Eq.(11), the solid phase strain Ωs is Ωs = e−j(kx x+kyy)(cid:16) C5e−jk3zz + C6ejk3zz(cid:17)(cid:32) (cid:33) iy − ky kx ix y , u f z ]T is Subsequently, the displacement u = [us y, us x, us z, u f x , u f u = e−j(kx x+kyy)Y2EC (17) (18) (19) (20) Here, the matrix E and the vector C are provided in Eq.(15)-(16). The non-zero ele- ments of the coefficient matrix Y2 are provided in Appendix B. 2.3. Incident case 3 When ky = kx = 0, the solid phase strain Ωs = 0. Subsequently, the displacement u = [us x, us y, us z, u f x , u f y , u f z ]T is u = e−j(kx x+kyy)Y3 E C where E = diag(e−jkm 1zz, e−jkm 1zz, ejkm 2zz, ejkm C = [C1, C2, C3, C4, 0, 0]T 2zz, 0, 0) (21) Here, the matrix E is a 6 × 6 diagonal matrix. The non-zero elements of the coefficient matrix Y3 are given in Appendix C. Once the poroelastic displacements in the porous media are obtained, their force components can be solved using the stress-strain relations in Biot theory [12]. The poroelastic field is revealed subsequently. 5 3. Discussions 3.1. Comparison with the 3D results by Zhou The problems proposed by Zhou [8] (a 3D double panel with porous lining under different boundary conditions) are solved using the field expressions obtained here. Their boundary formulations and parameters can be found in [8]. The results obtained here versus Zhou's results, in the Mach number M = 0 case, are provided in Fig.1(a). It is found that Zhou overestimated the results slightly, though the same trends are shared, owing to the incomplete 3D poroelastic field expressions in [8]. His field expressions, when kx = k1cosϕ1cosθ1 = 0 (the same definitions and notations as [8]), are invalid and lead to division by zero. As k1 (cid:44) 0, ϕ1 ∈ [π/10, π/2] and θ1 ∈ [0, 2π], these division- by-zero cases lead to ϕ1 = π/2, or θ1 = π/2, 3π/2. These three incident cases, which are not properly discussed by Zhou, should be correctly solved to obtain the STL. Fig. 1. Comparison of the random STL results with Zhou: (a) the results here (the lines) versus Zhou's results (the marks); (b) the results here exclude division-by-zero cases (the lines) versus Zhou's results (the marks) To study the influence of the division-by-zero cases, the results here with the division- by-zero cases excluded, versus Zhou's results, are provided in Fig.1(b). It shows the consistency of the two results is more satisfactory than in Fig.1(a). These results con- firm that, the results by Zhou, though amply acceptable, fail to reveal the division-by- zero cases and lead to an overestimate of the STL. 6 102103104Frequency(Hz)020406080STL(dB)102103104Frequency(Hz)020406080STL(dB)(a)(b)Fig. 2. Comparison of the random STL results with Bolton: the results here (the lines) versus Bolton's results (the marks) (a) the BB, BU, and UU cases (b) the B and U cases 3.2. Validation with the 2D results by Bolton In his paper [6], Bolton obtained 2D poroelastic field expressions and studied the random STL of a 2D double panel with porous lining under different boundary condi- tions. These 2D results are compared with the results obtained using the field expres- sions provided here (with ky = 0) in Fig.2 (the same configurations and parameters in [6] are used). All the five boundary cases are in good consistency. These results show the correctness of the 3D field expressions obtained here. 3.3. Verification of the division-by-zero cases The division-by-zero cases, owing to incomplete expressions [8], are tested to ver- ify the field expressions got here. Three 3D oblique incident conditions are identified here, which are ϕ1 = π/2, θ1 = π/2 or θ1 = 3π/2 (where the elevation angle ϕ1 and the azimuth angle θ1 are defined in [8]). However, as θ1 = π/2 or 3π/2 correspond to the 2D plane incident cases which were validated using Boltons results, only ϕ1 = π/2, i.e., the normal incident case, is verified here. A problem proposed by Allard [12], which is a poroelastic layer bonded on to a rigid impermeable wall in a normal acoustic field, is discussed to test the normal incidence solution (the same configuration and parameters in [12] are used). Result comparisons with Allard are provided in Fig.3. As shown, the consistency between the results got here and Allards results is satisfactory. This consistency confirms the effectiveness of the 3D poroelastic field expressions got here. 7 102103104Frequency(Hz)020406080STL(dB)102103104Frequency(Hz)01020304050STL(dB)(a)(b)Fig. 3. Comparison of the normalized surface impedance at normal incidence when the porous layer thickness is (a) 10 cm (b) 5.6 cm (the lines - the results obtained here; the marks - Allard's results) 4. Conclusions The correct 3D poroelastic field expressions, under monochromatic harmonic inci- dence, were revealed here. Result comparisons of these field expressions with former expressions showed that the results got formerly, though acceptable, overestimated the random STL slightly and led to division-by-zero. The division-by-zero cases and vali- dations with 2D theoretical results were subsequently tested. Their result comparisons were all satisfactory and confirmed the soundness of the 3D field expressions got here. As a practicable and reliable tool, these 3D field expressions can benefit to understand the acoustic properties of poroelastic materials. Acknowledgements This work is supported by the National Natural Science Foundation of China (NSFC) No.11572137. The authors wish to thank Professor J. Stuart Bolton, Professor Yu Liu and Dr. Jie Zhou for their warm-hearted discussions. 8 0.20.40.60.811.21.41.6-4-20240.20.40.60.811.21.41.6Frequency(Hz)-202(a)(b)Appendix A. The coefficient matrix of incident case 1 The non-zero elements of matrix Y1 are , Y1(1, 3) = Y1(1, 4) = jkx Y1(1, 1) = Y1(1, 2) = jkx k2 k2 2 1 , Y1(1, 6) = −Y1(1, 5) Y1(1, 5) = − jkxk3z k2 3ky , Y1(2, 3) = Y1(2, 4) = jky Y1(2, 1) = Y1(2, 2) = jky k2 k2 2 1 , Y1(2, 6) = −Y1(2, 5) Y1(2, 5) = − jk3z k2 3 , Y1(3, 2) = −Y1(3, 1) Y1(3, 1) = j k1z k2 1 , Y1(3, 4) = −Y1(3, 3) Y1(3, 3) = j k2z k2 2 x +k2 Y1(3, 5) = j(k2 y ) , Y1(3, 6) = Y1(3, 5) k2 3ky b1, Y1(4, 3) = Y1(4, 4) = jkx Y1(4, 1) = Y1(4, 2) = jkx k2 k2 1 2 g, Y1(4, 6) = −Y1(4, 5) Y1(4, 5) = − jkxk3z k2 3ky Y1(5, 1) = Y1(5, 2) = jky b1, Y1(5, 3) = Y1(5, 4) = jky k2 k2 1 2 g, Y1(5, 6) = −Y1(5, 5) Y1(5, 5) = − jk3z k2 3 b1, Y1(6, 2) = −Y1(6, 1) Y1(6, 1) = j k1z k2 1 b2, Y1(6, 4) = −Y1(6, 3) Y1(6, 3) = j k2z k2 2 x +k2 Y1(6, 5) = j(k2 y ) g, Y1(6, 6) = Y1(6, 5) k2 3ky b2 b2 (A.1) 9 Appendix B. The coefficient matrix of incident case 2 The non-zero elements of matrix Y2 are Y2(1, 1) = Y2(1, 2) = jkx k2 1 , Y2(1, 3) = Y2(1, 4) = jkx k2 2 Y2(1, 5) = jk3z k2 3 , Y2(1, 6) = −Y2(1, 5) Y2(2, 1) = Y2(2, 2) = jky k2 1 , Y2(2, 3) = Y2(2, 4) = jky k2 2 , Y2(2, 6) = −Y2(2, 5) Y2(2, 5) = jkyk3z k2 3kx , Y2(3, 2) = −Y2(3, 1) Y2(3, 1) = j k1z k2 1 , Y2(3, 4) = −Y2(3, 3) Y2(3, 3) = j k2z k2 2 Y2(3, 5) = − j(k2 x +k2 y ) , Y2(3, 6) = Y2(3, 5) k2 3kx Y2(4, 1) = Y2(4, 2) = jkx b1, Y2(4, 3) = Y2(4, 4) = jkx k2 k2 1 2 g, Y2(4, 6) = −Y2(4, 5) Y2(4, 5) = jk3z k2 3 Y2(5, 1) = Y2(5, 2) = jky b1, Y2(5, 3) = Y2(5, 4) = jky k2 k2 1 2 g, Y2(5, 6) = −Y2(5, 5) Y2(5, 5) = jkyk3z k2 3kx b1, Y2(6, 2) = −Y2(6, 1) Y2(6, 1) = j k1z k2 1 b2, Y2(6, 4) = −Y2(6, 3) Y2(6, 3) = j k2z k2 2 Y2(6, 5) = − j(k2 x +k2 y ) g, Y2(6, 6) = Y2(6, 5) k2 3kx b2 b2 Appendix C. The coefficient matrix of incident case 3 The non-zero elements of matrix Y3 are Y3(3, 1) = j k1 Y3(3, 3) = j k2 Y3(6, 1) = j k1 Y3(6, 3) = j k2 , Y3(3, 2) = −Y3(3, 1) , Y3(3, 4) = −Y3(3, 3) b1, Y3(6, 2) = −Y3(6, 1) b2, Y3(6, 4) = −Y3(6, 3) 10 (B.1) (C.1) References [1] M. A. Biot, Theory of propagation of elastic waves in a fluidsaturated porous solid. i. lowfrequency range, J. Acoust. Soc. Am. 28 (1956) 168 -- 178. [2] N. Atalla, R. Panneton, P. Debergue, A mixed displacement-pressure formulation for poroelastic materials, J. Acoust. Soc. Am. 104 (1998) 1444 -- 1452. [3] N. Atalla, F. Sgard, C. K. Amedin, On the modeling of sound radiation from poroelastic materials, J. Acoust. Soc. Am. 120 (2006) 1990 -- 1995. [4] M. Schanz, Fast multipole method for poroelastodynamics, Eng. Anal. Boundary Elem. 89 (2018) 50 -- 59. [5] H. Deresiewicz, J. T. Rice, The effect of boundaries on wave propagation in a liquid-filled porous solid: iii. reflection of plane waves at a free plane boundary (general case), Bull. Seismol. Soc. Am. 52 (1962) 595 -- 625. [6] J. Bolton, N.-M. Shiau, Y. Kang, Sound transmission through multi-panel struc- tures lined with elastic porous materials, J. Sound Vib. 191 (1996) 317 -- 347. [7] J. F. Allard, C. Depollier, P. Rebillard, W. Lauriks, A. Cops, Inhomogeneous biot waves in layered media, J. Appl. Phys. 66 (1989) 2278 -- 2284. [8] J. Zhou, A. Bhaskar, X. Zhang, Sound transmission through a double-panel con- struction lined with poroelastic material in the presence of mean flow, J. Sound Vib. 332 (2013) 3724 -- 3734. [9] Y. Liu, A. Sebastian, Effects of external and gap mean flows on sound trans- mission through a double-wall sandwich panel, J. Sound Vib. 344 (2015) 399 -- 415. [10] R. Talebitooti, K. Daneshjou, M. Kornokar, Three dimensional sound transmis- sion through poroelastic cylindrical shells in the presence of subsonic flow, J. Sound Vib. 363 (2016) 380 -- 406. 11 [11] J. Zhou, A. Bhaskar, X. Zhang, Optimization for sound transmission through a double-wall panel, Appl. Acoust. 74 (2013) 1422 -- 1428. [12] J. F. Allard, N. Atalla, Propagation of Sound in Porous Media, John Wiley & Sons, Ltd, 2009. [13] H. Qiao, Z. He, W. Jiang, W. Peng, composite structure lined with porous core: arXiv:1808.06590 (2018). Sound transmission of periodic rib-stiffened double panel case, [14] K. F. Graff, Wave motion in elastic solids, Ohio State University Press, 1975. 12
1907.05027	1	1907	2019-07-11T07:20:52	Improving the Signal-to-noise Ratio for Heat-Assisted Magnetic Recording by Optimizing a High/Low Tc bilayer structure	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We optimize the recording medium for heat-assisted magnetic recording by using a high/low $T_{\mathrm{c}}$ bilayer structure to reduce AC and DC noise. Compared to a former work, small Gilbert damping $\alpha=0.02$ is considered for the FePt like hard magnetic material. Atomistic simulations are performed for a cylindrical recording grain with diameter $d=5\,$nm and height $h=8\,$nm. Different soft magnetic material compositions are tested and the amount of hard and soft magnetic material is optimized. The results show that for a soft magnetic material with $\alpha_{\mathrm{SM}}=0.1$ and $J_{ij,\mathrm{SM}}=7.72\times 10^{-21}\,$J/link a composition with $50\%$ hard and $50\%$ soft magnetic material leads to the best results. Additionally, we analyse how much the areal density can be improved by using the optimized bilayer structure compared to the pure hard magnetic recording material. It turns out that the optimized bilayer design allows an areal density that is $1\,$Tb/in$^2$ higher than that of the pure hard magnetic material while obtaining the same SNR.	physics.app-ph	physics	Improving the Signal-to-noise Ratio for Heat-Assisted Magnetic Recording by Optimizing a High/Low Tc bilayer structure O. Muthsam,1, a) F. Slanovc,1 C. Vogler,1 and D. Suess1 University of Vienna, Physics of Functional Materials, Boltzmanngasse 5, 1090 Vienna, Austria (Dated: 12 July 2019) We optimize the recording medium for heat-assisted magnetic recording by using a high/low Tc bilayer structure to reduce AC and DC noise. Compared to a former work, small Gilbert damping α = 0.02 is considered for the FePt like hard magnetic material. Atomistic simu- lations are performed for a cylindrical recording grain with diameter d = 5 nm and height h = 8 nm. Different soft magnetic material compositions are tested and the amount of hard and soft magnetic material is optimized. The results show that for a soft magnetic material with αSM = 0.1 and Jij,SM = 7.72 × 10−21 J/link a composition with 50% hard and 50% soft magnetic material leads to the best results. Additionally, we analyse how much the areal density can be improved by using the optimized bilayer structure compared to the pure hard magnetic recording material. It turns out that the optimized bilayer design allows an areal density that is 1 Tb/in2 higher than that of the pure hard magnetic material while obtaining the same SNR. I. INTRODUCTION Heat-assisted magnetic recording (HAMR) [1 -- 7] is a promising recording technology to further increase the areal storage densities (ADs) of hard disk drives. Con- ventional state-of-the-art recording technologies are not able to overcome the so-called recording trilemma [8]: Higher ADs require smaller grains. These grains need to have high uniaxial anisotropy to be thermally sta- ble. However today's write heads are not able to pro- duce fields that are strong enough to switch these high anisotropy grains. In the HAMR process a heat pulse is included in the recording process to locally heat the recording medium. This leads to a drop of the coercivity, making the high anisotropy recording medium writeable. The medium is then quickly cooled and the information reliably stored. To reach high linear densities it is necessary to reduce AC and DC noise in recording media [9]. AC noise de- termines the distance between neighboring bits in bit- patterned [10 -- 12] media or the transition between grains in granular media. DC noise restricts the maximum switching probability of grains away from the transition. It has been shown, that pure hard magnetic grains do not switch reliably [13] if bit-patterned media are con- sidered whereas non-optimized exchange coupled bilayer structures [14 -- 19] of hard and soft magnetic material ex- perience high AC noise [20]. A work to reduce noise in recording media by optimizing a high/low Tc bilayer structure (see Ref. [21]) showed that an optimial bilayer structure consists of 80% hard magnetic and 20% soft magnetic material. However, in the former work the Gilbert damping was assumed to be αHM = 0.1 which is hard to achieve in a FePt like hard magnetic material in reality. In realistic hard magnetic recording materi- als, the damping constant is α = 0.02, according to the Advanced Storage Technology Consortium (ASTC) [22]. a)Electronic mail: [email protected] Since it has been shown that the damping constant has a strong influence on the maximum switching probabil- ity and the down-track jitter, we follow the optimization approach and optimize a bilayer structure for the ASTC parameters. After the optimization, we study how the optimized material differs from that with αHM = 0.1. Additionally, we investigate how much the areal storage density (AD) can be improved when using the optimized recording material instead of the pure hard magnetic one. This is done with the help of the signal-to-noise ratio (SNR), which gives the power of the signal over the power of the noise and is a good indicator for the quality of writ- ten bits. The structure of this work is as follows: In Section II, the HAMR model and the material parameters are pre- sented. In Section III, the results are shown and they are discussed in Section IV. II. HAMR MODEL the stochastic Landau-Lifshitz-Gilbert The optimization simulations are performed with the atomistic simulation program VAMPIRE [23] which solves (LLG) equation. In the simulations, a cylindrical recording grain with a diameter d = 5 nm and a height h = 8 nm is used. It can be considered as one recording bit in bit-patterned media. A simple cubic crystal structure is used and only nearest neighbor interactions are consid- ered. The effective lattice parameter a and the exchange interaxtion Jij are adjusted in order to lead to the exper- imentally obtained saturation magnetization and Curie temperature. [24; 25]. The write head is assumed to move with a velocity of v = 15 m/s. A continuous laser pulse is assumed with the Gaussian temperature profile T (x, y, t) = (Twrite − Tmin)e − x2+y2 = Tpeak(y) · e 2σ2 + Tmin − x2 2σ2 + Tmin (1) (2) with FWHM(cid:112)8 ln(2) . σ = (3) The full width at half maximum (FWHM) is assumed to be 60 nm. Both, the down-track position x and the off-track position y are variable in the simulations. The initial and final temperature is Tmin = 300 K. The applied field is modeled as a trapezoidal field with a write field duration of 0.57 ns and a field rise and decay time of 0.1 ns. The field is applied at an angle of 22 deg with respect to the normal. The field strength is assumed to be +0.8 T and -0.8 T in z-direction. Initially, the magnetization of each grain points in +z-direction. The trapezoidal field tries to switch the magnetization of the grain from +z-direction to −z-direction. At the end of every simulation, it is evaluated if the bit has switched or not. A. Material parameters for The material parameters the hard magnetic material can be seen in Table I. For the soft magnetic material, the atomistic spin moment is assumed to be µs = 1.6 µB which corresponds to a saturation polariza- tion Js = 1.35 T. The uniaxial anisotropy constant ku,SM in the soft magnetic layer is initially set to 0 but later varied. The Gilbert damping αSM and the exchange interaction Jij,SM within the soft magnetic material are varied. Experimentally, it is possible to increase the damping constant by doping the soft magnetic material with transition metals like Gd or Os [26 -- 30]. Thus, also enhanced damping constants αSM larger than 0.02 are considered in the simulations. 2 FIG. 1. Switching probability phase diagram of a pure FePt like hard magnetic grain. The contour lines indicate the transition between areas with switching probability less than 1% (red) and areas with switching probability higher than 99.2% (blue). The dashed lines mark the switching probabil- ity curves of Figure 2. the pure hard magnetic grain shows only two small ar- eas with switching probability larger than 99.2%. This threshold is used, since 128 simulations per phase point are performed and a switching probability of 100% corre- sponds to a number of successfully switched trajectories larger than 1 − 1/128 = 0.992. To determine the down-track jitter σ, a down-track switching probability curve P (x) for −20 nm ≤ x ≤ 6 nm at a fixed temperature Tpeak = 760 K is determined for pure hard magnetic material (see Figure 2). The switch- ing probability curve is fitted with a Gaussian cumulative function III. RESULTS A. Hard magentic grain Φµ,σ2 = 1 2 with (1 + erf( )) · P x − µ √2σ2 (cid:90) x −τ 2 (4) (5) First, a switching probability phase diagram for the pure hard magnetic material is computed where the switching probability is depending on the down-track po- sition x and the off-track position y. With eq. (2) each off-track position y can be transformed into an unique peak temperature Tpeak, if the write temperature Twrite is fixed, and vice versa. Thus, the switching probability in Figure 1 is shown as a function of the down-track position x and the peak temperature Tpeak that corresponds to y. The resolution of the phase diagram in down-track direc- tion is ∆x = 1.5 nm and that in temperature direction is ∆Tpeak = 25 K. In each phase point, 128 trajectories are simulated with a simulation length of 1.5 ns. Thus, the phase diagram contains more than 30.000 switching trajectories. From the phase diagram it can be seen that erf(x) = 2 √π e 0 dτ, where the standard deviation σ, the mean value µ and the mean maximum switching probability P ∈ [0, 1] are the fitting parameters. The standard deviation σ deter- mines the steepness of the transition function and is a measure for the transition jitter. In the further course it will be called σdown. The fitting parameter P is a measure for the average switching probability for suffi- ciently high temperatures. The resulting fitting parame- ters of the hard magnetic material can be seen in Table V. Note, that the calculated jitter values only consider the down-track contribution of the write jitter. The so-called a−parameter is given by Curie temp. TC [K] Damping α 693.5 0.02 Uniaxial anisotropy. ku [J/link] 9.124 × 10−23 Jij [J/link] 6.72 × 10−21 µs [µB] 1.6 TABLE I. Material parameters of a FePt like hard magnetic granular recording medium. 3 FIG. 2. Down-track switching probability curve P (x) at a peak temperature Tpeak = 760 K for a pure hard magnetic grain. (cid:113) a = σ2 down + σ2 g (6) FIG. 3. Down-track jitter σdown as a function of the damp- ing constant and the exchange interaction. The contour line indicates the transition between areas with down-track jitter larger than 0.5 nm (light red, blue) and areas with down-track jitter smaller than 0.5 nm (dark red). where σg is a grain-size-dependent jitter contribution [31]. The write jitter can then be calculated by (cid:114) σwrite ≈ a S W (7) where W is the reader width and S = D + B is the grain diameter, i.e. the sum of the particle size D and the nonmagnetic boundary B [32; 33]. B. Media Optimization To find the best soft magnetic material composition, down-track switching probability curves P (x) similar to Figure 2 are computed for 50/50 bilayer structures with different damping constants αSM and different exchange interactions Jij,SM. The range in which the parameters are varied can be seen in Table II. Note, that P (x) is computed at different peak temperatures for the different exchange interactions, since there holds Jij = 3kBTC z , (8) where kB is the Boltzmann constant, z is the number of nearest neighbors and is a correction factor from the mean-field expression which is approximately 0.86 [23]. The temperature at which P (x) is calculated is chosen to be TC + 60 K. The down-track switching probability curves are then fitted with eq. (4). The down-track jitter parameters as a function of the damping constant and the exchange interaction can be see in Figure 3. The maximum switching probability is 1 for α ≥ 0.1. From the simulations it can be seen that a Gilbert damping αSM = 0.1 together with Jij,SM = 7.72 × 10−21 J/link leads to the best results with the smallest down-track jitter σdown = 0.41 nm and a switching proa- bility P = 1. The last soft magnetic parameter that is varied, is the uniaxial anisotropy ku,SM. It is known that the small- est coercive field in an exchange spring medium can be achieved if KSM = 1/5KHM [34; 35]. Here Ki = natku,i a3 i ∈ {SM, HM} (9) are the macroscopic anisotropy constants in J/m3 with the unit cell size a = 0.24 nm and the number of atoms nat per unit cell. ku,SM is varied between 0 and 1/2ku,HM = 4.562× 10−23 J/link. The damping constant −15−10−5000.20.40.60.81down-trackx[nm]switchingprobabilityHMParameter min. value αSM Jij,SM [J/link] 5.72 × 10−21 ku,SM [J/link] 0.02 0 max.value 9.72 × 10−21 0.5 1/2ku,HM = 4.562 × 10−23 4 TABLE II. Range in which the different soft magnetic material parameters are varied. ku,SM × 10−23 [J/link] σdown [nm] P 1.0 0 1.0 0.562 1.0 1.8428 [= 1/5 ku,HM] 1.0 3.124 4.562 [= 1/2 ku,HM] 1.0 0.41 0.919 1.04 0.898 1.01 TABLE III. Resulting down-track jitter parameters and mean maximum switching probability values for soft magnetic materials with different uniaxial anisotropy constants ku,SM. is αSM = 0.1. The resulting fitting parameters are sum- marized in Table III. It can be seen that the switching probability is one for all varied ku,SM. However, the down-track jitter increases for higher ku,SM. Since for ku,SM = 0 J/link the jitter is the smallest, this value is chosen for the optimal material composition. In conclusion, the material parameters of the optimized soft magnetic material composition can be seen in Ta- ble IV. Next, simulations for different ratios of hard and soft magnetic material are performed. Down-track switching probability curves P (x) are computed for different ratios at Tpeak = 780 K and the down-track jitter and the mean maximum switching probability are determined. The re- sults are listed in Table V. It can be seen that a structure with 50% hard magnetic and 50% soft magnetic materials leads to the smallest jitter and the highest switching probability. This result differs from the optimized material composition in Ref. [21], where the optimal composition consists of 80% hard magnetic and 20% soft magnetic materials. In Figure 4, a switching probability phase diagram of the optimized bilayer structure with 50% hard and 50% soft magnetic material can be seen. It is visible that the switching probability of the structure is larger than 99.2% for a bigger area of down- track positions and peak temperatures. This shows the reduction of DC noise in the optimized structure. FIG. 4. Switching probability phase diagram of recording grain consisting of a composition of 50% hard magnetic ma- terial and 50% soft magnetic material with ku,SM = 0 J/link and Jij,SM = 7.72 × 10−21 J/link. The contour lines indicate the transition between areas with switching probability less than 1% (red) and areas with switching probability higher than 99.2% (blue). C. Areal Density To analyse the possible increase of areal density by us- ing the optimized bilayer structure instead of the pure hard magnetic recording medium, the signal-to-noise ra- tio is calculated. With the help of an analytical model of a phase diagram developed by Slanovc et al [33] it is pos- sible to calculate a switching probability phase diagram from eight input parameters. The input parameters are Pmax, σdown, the off-track jitter σoff , the transition cur- vature, the bit length, the half maximum temperature and the position of the phase diagram in Tpeak direc- tion and the position of the phase diagram in down-track direction. The σdown and Pmax values are those result- ing from the simulations for pure hard magnetic material and the optimized bilayer structure. All other model in- put parameters are obtained by a least square fit from a switching probability phase diagram computed with a coarse-grained LLB model [36]. The phase diagram is mapped onto a granular recording medium where the switching probability of the grain corresponds to its po- sition. The writing process is repeated for 50 different randomly initialized granular media. The SNR is then computed from the read-back process with the help of a SNR calculator provided by SEAGATE [37]. The SNR is analysed for areal densities of 2 to 5 Tb/in2. For the bitsize (bs) at a certain areal density, there are different track width and bit length combinations (t, b) Damping αSM 0.1 Uniaxial anisotropy. ku [J/link] 0 Jij [J/link] 7.72 × 10−21 µs [µB] 1.6 TABLE IV. Resulting material parameters for the optimal soft magnetic material composition. 5 0.974 1.06 0.813 HM/SM σdown [nm] P HM 90/10 80/20 70/30 60/40 50/50 0.6 0.8 0.41 0.95 0.969 0.998 0.988 0.999 1.0 TABLE V. Resulting down-track jitter parameters and mean maximum switching probability values for hard magnetic material and three different hard/soft bilayer structures with different damping constants in the soft magnetic material. that yield bs = t · b. (10) To compute the SNR for a certain (t, b) combination, the reader was scaled in both the down-track and the off- track direction according to the bit length and the track width, respectively. The reader resolution R in down- track direction is scaled by R = R0 · b b0 (11) where b is the bit length, R0 = 13.26 nm is the initial reader resolution and b0 = 10.2 nm denotes the mean ini- tial bit length according to ASTC. In off-track direction, the reader width is scaled to the respective track width t. The initial track width is 44.34 nm. In Figure 5(a) and (b) the SNR is shown as a function of the bit length and the track width for pure hard magnetic material and the optimized bilayer structure, respectively. Additionally, the phase plots include the SNR curves for (t, b) combina- tions that yield areal densities from 2 to 5 Tb/in2. From the phase diagram it is visible that higher SNR values can be achieved for the optimized structures than for the pure hard magnetic material in the same bit length − track width range. For example, the SNR for an areal density of 2 Tb/in2 for the bilayer structure is larger than 15 dB whereas it is between 10 dB and 15 dB for pure hard mag- netic material. For each AD there is a (t, b) combination for which the SNR is maximal and which is marked by a dot in the phase plot. In Figure 6 the maximum SNR over the areal density is displayed for both structures. The results show that the SNR that can be achieved with the optimized structure is around 2 db higher than that of the hard magnetic material, if the same areal density is assumed. To get the same SNR, the optimized design allows for an areal density that is 1 Tb/in2 higher than for the hard magnetic one. Summarizing, the bit length − track width combinations at which the maximum SNR is achieved are given in Table VI. FIG. 5. Signal-to-noise ratio (in dB) as a function of the bit length and the track width for (a) pure hard magnetic ma- terial and (b) the optimized hard/soft bilayer structure. The red lines indicate the bit length − track width combinations that yield 2, 3, 4 and 5 Tb/in2 areal density. The dots indicate the combination at which the SNR is maximal. IV. CONCLUSION To conclude, we optimized a recording medium with high/low TC grains for heat-assisted magnetic record- ing with a low Gilbert damping in the hard magnetic part αHM = 0.02. The simulations for a cylindrical recording grain with d = 5 nm and h = 8 nm were AD [Tb/in2] Max. SNR [dB] (HM) x [nm] (HM) y [nm] (HM) Max. SNR [dB] (HM/SM) x [nm] (HM/SM) y [nm] (HM/SM) 2 3 4 5 13.85 11.07 9.46 7.16 10.0 6.23 5.0 4.3 32.26 34.52 32.26 30.01 16.08 13.37 11.55 9.16 8.06 5.37 5.0 4.69 40.02 37.53 32.26 27.51 TABLE VI. Resulting bit length x and track width y combinations for the maximum SNR at different areal densities (AD) for pure hard magnetic material (HM) and the optimized bilayer structure (HM/SM). 6 formance. In the former work the improved performance due to higher damping was not an issue since the damp- ing constant was 0.1 in both layers. This explains the different ratios of hard and soft magnetic material. Furthermore, we analyzed the increase of the areal den- sity can be improved if the optimized bilayer structure is used instead of pure hard magnetic recording mate- rial. This was done by analyzing the signal-to-noise ra- tio (SNR). The results showed that the areal density of the optimized bilayer structure could be increased by 1 Tb/in2 to achieve the same SNR as for the pure hard magnetic structure. In other words, that means that at a certain areal density, the SNR was increased by 2 dB by using the optimized structure. Concluding, the opti- mized bilayer structure is a promising design to increase the areal storage density by just modifying the recording material. FIG. 6. Maximum SNR for different areal densities for pure hard magnetic material and the optimized bilayer structure. V. ACKNOWLEDGEMENTS performed with the atomistic simulation program VAM- PIRE. The damping constant of the soft magnetic mate- rial was assumed to be enhanced by doping the soft mag- netic material with transition metals. The simulations showed that larger damping constants lead to smaller jit- ter and higher switching probabilities. A damping con- stant αSM = 0.1, in combination with an exchange in- teraction Jij,SM = 7.72 × 10−21 J/link and an uniaxial anisotropy constant ku,SM = 0 J/link, led to the best re- sults in terms of small down-track jitter and high switch- ing probability in a wide range of down-track and off- track positions. Interestingly, the soft magnetic com- position is almost the same as for the structure with αHM = 0.1 obtained in a previous work [21]. In further simulations the amount of hard and soft magnetic material was varied. Surprisingly, the results showed that a higher amount of soft magnetic material leads to smaller down-track jitter. This is not as expected since for αHM = 0.1 an increase of the soft magnetic ma- terial led to larger AC noise [21]. However, it can be easily explained why a higher amount of soft magnetic material leads to better jitter results. Studying the in- fluence of the damping constant on the down-track jitter shows that an increase of the damping constant from 0.02 to 0.1 reduces the down-track jitter by almost 30%. Ad- ditionally,the maximum switching probability increases to 1. Since it can be seen that higher damping leads to smaller jitter and higher maximum switching probability, it is reasonable that a higher amount of soft magnetic material with αSM = 0.1 leads to a better recording per- The authors would like to thank the Vienna Sci- ence and Technology Fund (WWTF) under grant No. MA14-044, the Advanced Storage Technology Consor- tium (ASTC), and the Austrian Science Fund (FWF) under grant No. I2214-N20 for financial support. The computational results presented have been achieved us- ing the Vienna Scientific Cluster (VSC). 1Hiroshi Kobayashi, Motoharu Tanaka, Hajime Machida, Takashi Yano, and Uee Myong Hwang. Thermomagnetic recording. Google Patents, August 1984. 2C. Mee and G. Fan. A proposed beam-addressable memory. IEEE Transactions on Magnetics, 3(1):72 -- 76, 1967. 3Robert E. Rottmayer, Sharat Batra, Dorothea Buechel, William A. Challener, Julius Hohlfeld, Yukiko Kubota, Lei Li, Bin Lu, Christophe Mihalcea, Keith Mountfield, and others. Heat-assisted magnetic recording. IEEE Transactions on Mag- netics, 42(10):2417 -- 2421, 2006. 4Mark H. Kryder, Edward C. Gage, Terry W. McDaniel, William A. Challener, Robert E. Rottmayer, Ganping Ju, Yiao- Tee Hsia, and M. Fatih Erden. Heat assisted magnetic recording. Proceedings of the IEEE, 96(11):1810 -- 1835, 2008. 5Tim Rausch, Ed Gage, and John Dykes. Heat Assisted Magnetic Recording. In Jean-Yves Bigot, Wolfgang Hbner, Theo Rasing, and Roy Chantrell, editors, Ultrafast Magnetism I, Springer Pro- ceedings in Physics, pages 200 -- 202. Springer International Pub- lishing, 2015. 6G. W. Lewicki and others. Thermomagnetic recording and magneto-optic playback system. Google Patents, December 1971. 7L. Burns Jr Leslie and others. Magnetic recording system. Google Patents, December 1959. 8R. F. L. Evans, Roy W. Chantrell, Ulrich Nowak, Andreas Ly- beratos, and H.-J. Richter. Thermally induced error: Den- sity limit for magnetic data storage. Applied Physics Letters, 100(10):102402, 2012. 9Christoph Vogler, Claas Abert, Florian Bruckner, Dieter Suess, and Dirk Praetorius. Basic noise mechanisms of heat-assisted- 7 37(4):1749 -- 1754, July 2001. 30J. O. Rantschler, R. D. McMichael, A. Castillo, A. J. Shapiro, W. F. Egelhoff, B. B. Maranville, D. Pulugurtha, A. P. Chen, and L. M. Connors. Effect of 3d, 4d, and 5d transition metal doping on damping in permalloy thin films. Journal of Applied Physics, 101(3):033911, February 2007. 31Xiaobin Wang, Bogdan Valcu, and Nan-Hsiung Yeh. Transi- tion width limit in magnetic recording. Applied Physics Letters, 94(20):202508, 2009. 32Gaspare Varvaro and Francesca Casoli. Ultra-High-Density Mag- netic Recording: Storage Materials and Media Designs. CRC Press, March 2016. 33Florian Slanovc, Christoph Vogler, Olivia Muthsam, and Dieter Suess. Systematic parameterization of heat-assisted magnetic recording switching probabilities and the consequences for the resulting snr. arXiv preprint arXiv:1907.03884, 2019. 34F. B. Hagedorn. Analysis of ExchangeCoupled Magnetic Thin Films. Journal of Applied Physics, 41(6):2491 -- 2502, May 1970. 35D. Suess. Multilayer exchange spring media for magnetic record- ing. Applied Physics Letters, 89(11):113105, September 2006. 36Christoph Vogler, Claas Abert, Florian Bruckner, and Dieter Landau-Lifshitz-Bloch equation for exchange-coupled Suess. grains. Physical Review B, 90(21):214431, December 2014. 37S. Hernndez, P. Lu, S. Granz, P. Krivosik, P. Huang, W. Eppler, T. Rausch, and E. Gage. Using Ensemble Waveform Analysis to Compare Heat Assisted Magnetic Recording Characteristics of Modeled and Measured Signals. IEEE Transactions on Magnet- ics, 53(2):1 -- 6, February 2017. magnetic recording. Journal of Applied Physics, 120(15):153901, 2016. 10H. J. Richter, A. Y. Dobin, R. T. Lynch, D. Weller, R. M. Brockie, O. Heinonen, K. Z. Gao, J. Xue, R. J. M. v. d. Veerdonk, P. Asselin, and M. F. Erden. Recording potential of bit-patterned media. Applied Physics Letters, 88(22):222512, May 2006. 11H. J. Richter, A. Y. Dobin, O. Heinonen, K. Z. Gao, R. J. M. v d Veerdonk, R. T. Lynch, J. Xue, D. Weller, P. Asselin, M. F. Erden, and R. M. Brockie. Recording on Bit-Patterned Media at Densities of 1 Tb/in2 and Beyond. IEEE Transactions on Magnetics, 42(10):2255 -- 2260, October 2006. 12Bing K. Yen, Jim Hennessey, Eric Freeman, Kim Yang Lee, David S. Kuo, and Mark Ostrowski. Bit patterned media, August 2013. 13Dieter Suess, Christoph Vogler, Claas Abert, Florian Bruckner, Roman Windl, Leoni Breth, and J. Fidler. Fundamental limits in heat-assisted magnetic recording and methods to overcome it with exchange spring structures. Journal of Applied Physics, 117(16):163913, 2015. 14Dieter Suess. Micromagnetics of exchange spring media: Opti- mization and limits. Journal of magnetism and magnetic mate- rials, 308(2):183 -- 197, 2007. 15Dieter Suess, Thomas Schrefl, S. Fhler, Markus Kirschner, Gino Hrkac, Florian Dorfbauer, and Josef Fidler. Exchange spring media for perpendicular recording. Applied Physics Letters, 87(1):012504, 2005. 16R. H. Victora and X. Shen. Exchange coupled composite media IEEE Transactions on for perpendicular magnetic recording. Magnetics, 41(10):2828 -- 2833, October 2005. 17Jian-Ping Wang, Weikang Shen, and Jianmin Bai. Exchange coupled composite media for perpendicular magnetic recording. IEEE transactions on magnetics, 41(10):3181 -- 3186, 2005. 18Kevin Robert Coffey, Jan-Ulrich Thiele, and Dieter Klaus Weller. Thermal springmagnetic recording media for writing using mag- netic and thermal gradients. Google Patents, April 2005. 19Dieter Suess and Thomas Schrefl. Breaking the thermally in- duced write error in heat assisted recording by using low and high Tc materials. Applied Physics Letters, 102(16):162405, 2013. 20Christoph Vogler, Claas Abert, Florian Bruckner, Dieter Suess, and Dirk Praetorius. Areal density optimizations for heat- assisted magnetic recording of high-density media. Journal of Applied Physics, 119(22):223903, 2016. 21O. Muthsam, C. Vogler, and D. Suess. Noise reduction in heat- assisted magnetic recording of bit-patterned media by optimizing a high/low Tc bilayer structure. Journal of Applied Physics, 122(21):213903, 2017. 22ASTC IDEMA. http://idema.org/?cat=10. 23Richard FL Evans, Weijia J. Fan, Phanwadee Chureemart, Thomas A. Ostler, Matthew OA Ellis, and Roy W. Chantrell. Atomistic spin model simulations of magnetic nanomaterials. Journal of Physics: Condensed Matter, 26(10):103202, 2014. 24Oleg N Mryasov, Ulrich Nowak, K Yu Guslienko, and Roy W Chantrell. Temperature-dependent magnetic properties of fept: Effective spin hamiltonian model. EPL (Europhysics Letters), 69(5):805, 2005. 25O Hovorka, S Devos, Q Coopman, WJ Fan, CJ Aas, RFL Evans, Xi Chen, G Ju, and RW Chantrell. The curie temperature dis- tribution of fept granular magnetic recording media. Applied Physics Letters, 101(5):052406, 2012. 26W. Zhang, S. Jiang, P. K. J. Wong, L. Sun, Y. K. Wang, K. Wang, M. P. de Jong, W. G. van der Wiel, G. van der Laan, and Y. Zhai. Engineering Gilbert damping by dilute Gd doping in soft mag- netic Fe thin films. Journal of Applied Physics, 115(17):17A308, May 2014. 27S. Ingvarsson, Gang Xiao, S. S. P. Parkin, and R. H. Koch. Tun- able magnetization damping in transition metal ternary alloys. Applied Physics Letters, 85(21):4995 -- 4997, November 2004. 28J. Fassbender, J. von Borany, A. Mcklich, K. Potzger, W. Mller, J. McCord, L. Schultz, and R. Mattheis. Structural and magnetic modifications of Cr-implanted Permalloy. Physical Review B, 73(18), May 2006. 29W. Bailey, P. Kabos, F. Mancoff, and S. Russek. Control of mag- netization dynamics in Ni/sub 81/Fe/sub 19/ thin films through the use of rare-earth dopants. IEEE Transactions on Magnetics,
1910.08148	1	1910	2019-10-17T20:29:56	Phonon Transport Controlled by Ferromagnetic Resonance	[ "physics.app-ph", "cond-mat.mes-hall" ]	The resonant coupling of phonons and magnons is important for the interconversion of phononic and spin degrees of freedom. We studied the phonon transmission in LiNbO3 manipulated by the dynamic magnetization in a Ni thin film. It was observed that the phonons could be absorbed strongly through resonant magnon-phonon coupling, which was realized by optimizing the interfacial coupling between Ni and LiNbO3. The line shapes of phonon transmission were further investigated considering the magnon-phonon interconversion in the elastically driven ferromagnetic resonance process. The results promote unique routes for phonon manipulation and detection in the presence of magnetization dynamics.	physics.app-ph	physics	Phonon Transport Controlled by Ferromagnetic Resonance Chenbo Zhao1,2, Yi Li1, Zhizhi Zhang1, Michael Vogel1, John E. Pearson1, Jianbo Wang2, Wei Zhang3,1, Valentine Novosad1, Qingfang Liu2, and Axel Hoffmann1,4 1 Materials Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA 2 Key Laboratory of Magnetism and Magnetic Materials of the Ministry of Education, Lanzhou University, Lanzhou 730000, People's Republic of China 3 Department of Physics, Oakland University, Rochester, MI 48309, USA 4 Department of Materials Science and Engineering, University of Illinois at Urbana-Champaign, Illinois, IL 61801, USA ABSTRACT The resonant coupling of phonons and magnons is important for the interconversion of phononic and spin degrees of freedom. We studied the phonon transmission in LiNbO3 manipulated by the dynamic magnetization in a Ni thin film. It was observed that the phonons could be absorbed strongly through resonant magnon-phonon coupling, which was realized by optimizing the interfacial coupling between Ni and LiNbO3. The line shapes of phonon transmission were further the magnon-phonon interconversion in the elastically driven ferromagnetic resonance process. The results promote unique routes for phonon manipulation and detection in the presence of magnetization dynamics. investigated considering INTRODUCTION Elastically driven ferromagnetic resonance (FMR) is at the core of combining straintronics and spintronics [1-5], which is drawing much attention due to both interesting fundamental physics and potential applications. This includes among others, elastically driven spin pumping [6,7], phonon driven inverse Edelstein effect [8] and field-free magnetization switching [9]. Recently, several studies on magnons-phonons interconversions have emerged [10-14]. In particular, theoretical models were developed in order to provide a microscopic understanding of how magnon-phonon interaction influences the damping and transport of magnons [13,15]. The mechanisms for the coupling between phonons and magnons in ferromagnetic materials include, magnetostriction [1-9,16] and spin-rotation coupling [14,17,18]. The latter mechanism is a manifestation of the Einstein -- de Haas and Barnett effects corresponding to the transfer of the angular momentum between spin and mechanical degrees of freedom [19-21]. In other words, the transfer between magnon- and phonon-angular momentum can be manipulated by magnetization dynamics [15,19,21-24]. These earlier research works shed light on the opportunity for phonon manipulation [25] and detection using magnetization dynamics. However, a remaining challenge is to accurately measure the changes of the phonon systems due to the magnon-phonon interaction [4]. Therefore, we focus here on how the propagation of phonons can be modulated via ferromagnetic resonance. In the present work, we have investigated the dependence of the phonon transport on magnetization dynamics in Ni/LiNbO3 hybrid heterostructures. We Based on surface acoustic wave (SAW) magneto-transmission measurements, the interaction of SAWs and ferromagnetic thin films has been studied experimentally by several groups [1,2,5,8,26]. However, important aspects of the interaction mechanism still remain to be resolved. An important parameter to characterize the interaction mechanism between the SAWs and ferromagnetic thin films is the linewidth of the transmission power. In general, the linewidth of the transmission power obtained from phonon driven FMR is much larger than that in cavity FMR [1,2,26]. In order to fit the broadening of the linewidth of the transmission power obtained from phonon driven FMR, the posteriori damping constant used is several times larger than the damping values obtained from cavity FMR experiments [1,2]. At the same time, the significant broadening of the transmission power line shape contains important information about the magnon and phonon coupling [4], e.g., angular momentum transformation, during phonon driven FMR. Therefore, the phonon transport properties in the presence of phonon driven FMR also provide an effective approach for exploring the angular momentum transformation [11,27] via the coupled phonons and magnons. observe the simultaneous existence of dips and peaks in the transmission power of the surface acoustic waves (SAW) device, which we interpret by considering both phonon attenuation and generation [28,29]. By fitting the theory [28,29] to the phonon transmission power, the value of the FMR field, resonance linewidth and the exchange stiffness parameter of the magnon system can be estimated. The obtained resonant linewidth of the SAW driven FMR is very similar to those from waveguide FMR experiment. This suggests that the phonon driven FMR linewidth broadening is mainly due to magnon-phonon interaction, instead of the nonuniform excitation fields induced by SAW. Our findings thus generate new opportunities for phonon control and detection of magnetization dynamics. EXPERIMENT AND SET UP Figure 1(a) shows a schematic illustration of the magnon-phonon interaction in our SAW driven FMR measurement. In these devices, we excite, at microwave frequencies, the SAW using interdigitated transducers (IDT) on a 127.86o Y-X cut LiNbO3 substrate with the thickness of 500 μm. The SAW then drives subsequently FMR in the Ni film. The optical images of the Ni/LiNbO3 hybrid device with a 50-nm thick Ni rectangular film is shown in Fig. 1(b) and (c). The IDTs of 50-nm thick Al were fabricated by using maskless photolithography and electron-beam evaporation. The finger width and the of 12 μm. The Ni film was deposited via DC sputtering using 10 W power, and an Ar pressure of 3×10-3 Torr. The length of the delay line L was set at L = 438 μm, and the Ni film was deposited in the center of delay line. In order to observed phonon driven FMR, a DC magnetic field H is applied in a direction spacing of the IDT were both 3 μm, launching SAWs with a wavelength (cid:2019)(cid:3020)(cid:3002)(cid:3024) with an angle θ with respect to the SAW wavevector (cid:1863)(cid:3020)(cid:3002)(cid:3024) set to θ(cid:3404)60º and θ(cid:3404)90º, respectively. During phonon driven FMR measurements H is varied (cid:1874)(cid:3020)(cid:3002)(cid:3024)(cid:3404) 3912 m/s, according to the following relation between the velocity (cid:1874)(cid:3020)(cid:3002)(cid:3024) and frequency f0 of SAW: (cid:1874)(cid:3020)(cid:3002)(cid:3024)(cid:3404)(cid:2019)(cid:3020)(cid:3002)(cid:3024)(cid:1858)(cid:2868). The measured (cid:1874)(cid:3020)(cid:3002)(cid:3024) is from -1.1 to 1.1 kOe. As shown in Fig. 1(d), the reflection parameters of the device exhibit one peak at f0= 325 MHz, measured using a vector network analyzer (VNA). The velocity of this SAW mode can be determined as typical for the velocity of a Rayleigh SAW [30]. To separate the SAW signals and the electromagnetic wave, the time-gating function of VNA was used [1,26]. Figure 1(e) shows the transmission parameters of SAW signal for the device with gating on (centered at 145 ns and spanning over 45 ns) [1,26]. The SAW odd harmonics from 3th to 13th have been launched and the highest harmonics is located at 4.23 GHz. Because the 9th and 13th harmonics are weak, this work focuses on the 3th, 5th, 7th and 11th harmonics to investigate the phonon transport during the coupling of magnon and phonon relevant to the Rayleigh SAW. Some of the device with Ni films were annealed at 400 oC for 30 minutes in vacuum, and the as-grown and annealed devices both were measured to study the phonon transport properties [see Fig. 1(a)]. Figure 1 (a) shows schematic illustration of the magnon-phonon interaction during SAW driven FMR; (b) and (c) show optical images of the Ni/LiNbO3 hybrid device with 50-nm thick Ni rectangular film; (d) shows the reflection parameters for the device with H = 1000 Oe applied at θ = 60º; (e) shows the transmission parameters of SAW signal with gating on (centered at 145 ns and spanning over 45 ns) H applied at θ = 60º. RESULTS AND DISSCUSSION The SAW is generated and detected by electromagnetic wave using a pair of spatially separated IDTs on the LiNbO3 crystal. Coupling with the magnon system (Ni films) is achieved during the SAW propagation in between the two IDTs. Phonon-driven FMR can be characterized by the transmission parameter S21 using the VNA with time gating function[1,26]. To enhance the signal-to-noise ratio, the transmission parameter S21 of SAW was converted into transmission power P. Figure 2(a-f) show the colormap of magnetic field H dependence (H is applied at θ = 60º) of the normalized transmission power P of the SAW signal: (a-c) for an as-grown device at 5th, 7th and 11th harmonics, respectively; (d-f) for an annealed device at 5th, 7th and 11th harmonics, respectively. The color codes represent intensity of transmission power of SAW, with red indicating maximum transmission. The dotted green lines represent the Kittel formula, which is based on waveguide FMR measurements for an extended Ni film grown on LiNbO3. The colormap data shows that the intensity of SAW exhibits a minimum near zero field for the as-grown device, but an obvious attenuation near the FMR field is observed for the annealed devices. This shows that the annealing treatment improves the resonant coupling between the LiNbO3 and the ferromagnetic Ni material. The enhancement of the resonant magnon-phonon coupling can be attributed to improving the interface of the magnon-phonon system [31] via the thermal annealing treatment. This magnon-phonon coupling, as indicated by the dispersion crossing, is enhanced due to the strong absorption of phonons at f = 2.24 GHz and 3.56 GHz for the annealed device. The attenuation of the SAW transmission power near zero field [Figs. 2(a-c)] and the FMR field [Figs. 2(d-f)] can be attributed to the magnetization switching [1,32] and the FMR absorption [1-5], respectively. the zero field region due to the magnetization switching, which has been reported before [1,32]. Notably, peaks for transmission power near the FMR field are also observed [Fig. 3(a)]. These peaks indicate that the SAW The plots of the magnetic field dependence of the SAW signal peak position are shown in Fig. 3 and illustrate the phonon transport properties. Figure 3(a) and (c) show the magnetic field (H) dependence of the normalized transmission power for the as-grown device at θ(cid:3404) 60º and 90º, respectively. When H is applied at θ(cid:3404) 60º, the transmission power shows large dips near transmission intensity increases near the FMR field. When H is applied at θ(cid:3404) transmission power for the annealed device at θ(cid:3404) 60º and 90º, respectively. transmission power spectra are both observed at θ(cid:3404) 60º [Fig. 3(b)] simultaneously, but they are not present when θ(cid:3404) 90º [Fig. 3(d)]. This 90º, only the magnetization switching dips remain [Fig. 3(c)]. This shows that for θ = 60º ferromagnetic resonance is excited. However, there is only the uniform mode that participates in the coupling for θ = 90º. We find that the annealing treatment for the device can markedly tune the phonon transport. Figure 3(b) and (d) show the H dependence of the The dips at the FMR field and the peaks above the FMR field in the suggests that both attenuation and amplification of the phonon transport is possible due to the magnon-phonon interaction. The H position of the peaks [Fig. 3(a)] and dips [Fig. 3(b)] for as-grown and annealed devices are plotted in Figs. 3(e) and (f), respectively, as well as the expected behavior based on the Kittel formula. The H positions are in good agreement with the Kittel formula, which indicate that the phonons transport is manipulated by magnetization dynamics during FMR. Figure 2: Colormap of magnetic field dependence of the normalized transmission power of SAW signal for the as-grown and the annealed devices at different SAW harmonics with H applied at θ = 60º: (a-c) for the as-grown device at the 5th, 7th and 11th harmonic; (d-f) for the annealed device at the 5th, 7th and 11th harmonic. The dotted lines represent the Kittel formula, which come from waveguide FMR experiment for the full Ni film. The color codes represent intensity of transmission power of SAW, with red indicating maximal transmission. To model the phonons transport, we take into consideration both phonon attenuation and generation [28,29] manipulated by the magnetic dynamics, during the SAW-driven FMR. Angular momentum interconversion between magnon and phonons is related to the Einstein -- de Haas effect [14,18] and the Barnett effect [21]. Due to the magnon-phonon interconversion during FMR, phonons will redistribute the angular momentum and energy between magnons and phonons [15,17], and thus macroscopic mechanical rotation can be strengthened, leading to an enhancement of the transmission of the SAW. The change of the SAW transmitted power (E) related to the phonon attenuation and generation during FMR can be rewritten as [28,29]: (cid:1831)(cid:3404)(cid:1831)(cid:2868)(cid:3049)(cid:3295)(cid:3030) (cid:3003)(cid:3118)(cid:3118)(cid:3004)(cid:3120)(cid:3120)(cid:4698)(cid:3035)(cid:3280)(cid:3281)(cid:3281)(cid:3132) (cid:4698) exp(cid:4672)(cid:2879)(cid:3086)(cid:3116)(cid:3009)(cid:2870) (cid:4673)(cid:4675)(1), ∆(cid:3009)(cid:3118)(1(cid:3397)(cid:1868)(cid:2870))(cid:2879)(cid:3117)(cid:3118)(cid:4670)1(cid:3397)((cid:1868)(cid:3397)(cid:2010))(cid:2870)(cid:4671)(cid:2879)(cid:3117)(cid:3118)exp(cid:3428) (cid:2869)(cid:2878)((cid:3043)(cid:2878)(cid:3081))(cid:3118)(cid:3432)(cid:4674)1(cid:3397) (cid:2879)(cid:3334)(cid:3118) where, (cid:1868)(cid:3404)(cid:3009)(cid:2879)(cid:3009)(cid:3293)∆(cid:3009) (2), (cid:2015)(cid:3404) (cid:2870)(cid:3095)(cid:3033)(cid:3003)(cid:3118)(cid:3118)(cid:3013) (cid:3004)(cid:3120)(cid:3120)(cid:3049)(cid:3295)(cid:3014)(cid:3294)∆(cid:3009)(cid:3091)(cid:3116) (3), (cid:2010)(cid:3404)(cid:2870)(cid:3002)(cid:3096)((cid:2870)(cid:3095)(cid:3033))(cid:3118) (cid:3004)(cid:3120)(cid:3120)(cid:3014)(cid:3294)∆(cid:3009)(cid:3091)(cid:3116) (4), and the H dependence for the line shape of the transmission power is determined by the parameters p, η and β. Equation (1) corresponds to the conventional microwave transmission through a Ni film [28,29], but employs instead of an externally applied real microwave magnetic field the effective ac magnetic field (cid:1860)(cid:3032)(cid:3033)(cid:3033)(cid:2884) due to the magnon-phonon interaction defined in Eq. (5) below. The parameters in Eqs. (1-4) are: (cid:1874)(cid:3047) = 3912 m/s is the Rayleigh acoustic wave velocity, (cid:1855) = 3×108 m/s is the speed of light, B2 = 8.7×106 N/m2 modulus of Ni, ∆(cid:1834) is the linewidth of the FMR, (cid:1838) = 12 μm is characteristic length related to the SAW wavelength, H is the external magnetic field, (cid:1834)(cid:3045) is the FMR field, (cid:1839)(cid:3046) = 4.7×105 A/m is the saturation magnetization, (cid:1827) is the exchange stiffness parameter, (cid:2025) = 8900 kg/m3 is the density of Ni, (cid:1858) the frequency of the SAW and (cid:2020)(cid:2868) is the permeability of vacuum. The last term in is the magnetoelastic parameter of Ni, C44 = 1.22×1011 N/m2 is the elastic (1) formula the phonon thermo-conductivity [28,33,34]. is scattering due to the viscosity and the magnetization [1,2,5]: where b1 and b2 are the longitudinal-type magnetoelastic coupling constants, b6 the transmission power [Figs. 3(a) and (b); Figs. 3(c) and (d)] exhibits the fingerprint for the Rayleigh SAW driven FMR [1,5], which is relevant to the The field angle dependence (H at θ(cid:3404) 60º and 90º) for the line shape of components (cid:3627)(cid:1860)(cid:3032)(cid:3033)(cid:3033)(cid:2884) (cid:3627) of the effective ac magnetic field (cid:1860)(cid:3032)(cid:3033)(cid:3033) perpendicular to (cid:3627)(cid:1860)(cid:3032)(cid:3033)(cid:3033)(cid:2884) (cid:3627)(cid:3404)(cid:3627)(cid:1860)(cid:3051)(cid:1871)(cid:1861)(cid:1866)(cid:2016)(cid:3398)(cid:1860)(cid:3052)(cid:1871)(cid:1861)(cid:1866)(cid:2016)(cid:3398)(cid:1861)(cid:1860)(cid:3053)(cid:3627)(cid:3404)2((cid:1854)(cid:2869)(cid:3398)(cid:1854)(cid:2870))(cid:2013)(cid:3051)(cid:3051)(cid:2868)cos(cid:2016)sin(cid:2016)(cid:3397) 4(cid:1854)(cid:2874)(cid:2013)(cid:3053)(cid:3051)(cid:2868)cos(cid:2016) (5), is the shear-type magnetoelastic coupling constant, (cid:2013)(cid:3051)(cid:3051)(cid:2868)and (cid:2013)(cid:3053)(cid:3051)(cid:2868) are pure (cid:3627)(cid:1860)(cid:3032)(cid:3033)(cid:3033)(cid:2884) (cid:3627) is zero at θ(cid:3404)90º. Thus, the SAW cannot excite the FMR in that and generation observed for θ(cid:3404)60º [Figs. 3(a) and (b)] are related to the FMR and vanish at θ(cid:3404)90º [Figs. 3(c) and (d)]. Unlike the asymmetry of the dips for magnetoelastic coupling constant ((cid:1854)(cid:2869)(cid:3398)(cid:1854)(cid:2870)) [5]. Therefore, when the values of (cid:1854)(cid:2869) and (cid:1854)(cid:2870) are close to each other, the asymmetry becomes less apparent. This opposite H orientations in previous results [1,5,8], the dips are nearly symmetry in Fig. 3(b). The asymmetry has been found being proportional to the Rayleigh SAW strain components. As can be seen from Eq. (5), the value of configuration. Therefore, both the dips and peaks due to the phonon attenuation also indicates that the shear-type magnetoelastic coupling (b6) is dominant over the longitudinal-type magnetoelastic couplings (b1 and b2) for our device. Figure 3: (a-d) Magnetic field H dependence of the transmission power [the symbols are experimental data, and the solid lines are fit to theory using Eq. (1)]: (a) as-grown at θ(cid:3404)60º; (b) annealed device at θ(cid:3404)60º; (c) as-grown at θ(cid:3404)90º; (d) annealed device at θ(cid:3404)90º. (e) shows the H position of the transmission power peaks at angle θ(cid:3404)60º for the as-grown sample; (f) shows the H position of transmission power dips at angle θ(cid:3404)60º for the annealed device. The Kittel formula is determined from waveguide FMR experiments for full Ni film. Figure 4: (a-c) Theoretical calculation results of the magnetic field H dependence of the transmission power at different frequencies using Eq. (1): (a) With different (cid:2015) at f=3.56 GHz; (b) and (c) with values of (cid:2015)=3.1 and (cid:2015) = 0.3 at different frequencies. The transmission fitting of (cid:2015) and ∆(cid:1834) using ∆(cid:1834)(cid:2015)=(cid:1863)(cid:1858) (k is a constant), where (cid:2015) and ∆(cid:1834) are obtained from parameters (cid:2015), (cid:2010) and ∆(cid:1834) are shown in the table in Fig. 4(d). Since only ∆(cid:1834) and (cid:2015) change with f, the expression of (cid:2015) can be simplified as ∆(cid:1834)(cid:2015)=(cid:1863)(cid:1858) (k is a constant), resulting in a good fit for (cid:2015) and ∆(cid:1834), as shown in Fig. 4(d). This very well. By putting the fitted value of (cid:2010) into Eq. (4), the exchange stiffness The experimental results for the annealed device are shown in Fig. 3(b). The line shape of the transmission power can be fitted using Eq. (1), and the fit power are relative values and the cures are shifted with respective to each other. (d) Linear theory fits to the line shape of the transmission power at different frequency f for the annealed device using Eq. (1), as shown in Fig. 3(b). suggests that the model describes the phonon transport in these SAW devices parameter of Ni film is estimated as A = 0.54×10-6 ergs/cm (f= 3.56 GHz) and 0.64×10-6 ergs/cm (f= 2.24 GHz). These obtained values for A are slightly smaller than A = (0.76±0.03)×10-6 ergs/cm obtained from elastic small-angle neutron scattering for nanocrystalline Ni film [35]. At a lower frequency, f = 0.97 GHz, the anomalous value of η and β obtained from theory fits can be attributed to the nonuniform magnetization distribution, which also results in large inhomogeneities of the magnetization precession. the magnetization switching. According to from an with fits of the data to Eq. (1). We neglect the phonon scattering due to the calculations using Eq. (1), i.e., the H dependence of the transmission power transmission power is shown using relative values and the cures are shifted However, for the transmission power of the as-grown device [Fig. 3(a)], the experimental data cannot be fitted using Eq. (1) due to the strong attenuation given by the materials-specific parameters for Ni and the FMR parameters at f = 3.56 GHz determined independent waveguide-FMR experiments, (cid:2015) is calculated to be (cid:2015) = 3.1 using Eq. (3). This calculated value of (cid:2015) is consistent viscosity and thermo-conductivity and set (cid:2015)(cid:2868) =0 [28]. The theoretical with different values of (cid:2015) at f = 3.56 GHz, are shown in Fig. 4(a). The with respective to each other. The calculations show that the dips at H = (cid:1834)(cid:3045) are increasingly obvious with the increase of (cid:2015), which means the phonons are the phonon generation at (cid:1834)(cid:3404)(cid:1834)(cid:3045)(cid:3397)∆(cid:1834)(cid:3493)2(cid:2015)(cid:3398)1 ((cid:2015)(cid:3408)1/2) become more obvious with the decrease of (cid:2015). When (cid:2015)(cid:3409)1/2, only the peaks at H = (cid:1834)(cid:3045) are dependence of the transmission power with values of (cid:2015) = 3.1 and (cid:2015) = 0.3 at (cid:2015)(cid:3408)1/2] and as-grown device [Fig. 3(a), (cid:2015)(cid:3409)1/2, except the dips near zero field due to magnetization switching], respectively. The changes in (cid:2015) related Furthermore, the linewidth ∆(cid:1834) of the transmission power can be obtained to be ∆(cid:1834) = 325 Oe and 500 Oe at f = 2.24 GHz and 3.56 GHz from to magnetoelastic coupling [Fig. 3(a) and (b)], can be attributed to improvement the interface of the magnon-phonon [31] system by the annealing treatment. different frequencies, respectively. It can be seen that the line shape of the transmission power has a good agreement with the annealed [Fig. 3(b), observed. Figures 4(b) and (c) show theoretical calculations of the H strongly attenuated at the ferromagnetic resonance field. The peaks related to the transmission power dips plotted in Fig. 3(b), which are approximately 1.7 times larger than the values ∆(cid:1834) = 185.3 Oe and 290.6 Oe obtained from results (∆(cid:1834) = 186 Oe and 292 Oe) using Eq. (1). Therefore, compared to the waveguide FMR experiments. These values are consistent to the theory fitting to the magnon-phonon FMR measurements, any broadening effect coming from nonuniform excitation fields induced by SAW [1,2,26] can be ignored. The increase of linewidth can mainly be attributed interaction, e.g., angular momentum transformation [15,23]. These results show that the coupling between elastic and magnetic degrees of freedom open additional channels for information interconversion between phononic and magnonic components. CONCLUSION In conclusion, phonon transport properties during the phonon driven ferromagnetic resonance has been investigated. The ferromagnetic resonance is driven acoustically, since no external rf magnetic field is applied to the ferromagnet. Rather, a purely internal rf magnetic field arises due to magnetoelastic coupling between the surface acoustic wave elastic strain field and the ferromagnet. Annealing of the sample results in increase interfacial magnon-phonon coupling, and thus enables tuning of the line shape of transmission power. Considering both the phonon attenuation and generation simultaneously during the phonon driven FMR, the phonon transport properties and line shape of transmission power can be explained in a quantitative fashion. By analyzing the phonon transmission power, the magnetization dynamics of the magnon system can be detected. We also demonstrate that the broadening effect of the transmission power line shape during phonon driven FMR can be mainly attributed to the interaction of magnons and phonons, instead of the nonuniform excitation fields induced by surface acoustic wave. ACKNOWLEDGEMENTS This work was performed at the Argonne National Laboratory and supported by the Department of Energy, Office of Science, Materials Science and Engineering Division. The use of the Centre for Nanoscale Materials was supported by the US. Department of Energy (DOE), Office of Sciences, Basic Energy Sciences (BES), under Contract No. DE-AC02-06CH11357. Chenbo Zhao acknowledges additional financial support from the China Scholarship Council (no. 201806180105) for a research stay at Argonne. The author thanks José Holanda for uesfull disscussions. REFERENCE [1] M. Weiler, L. Dreher, C. Heeg, H. Huebl, R. Gross, M. S. Brandt, and S. T. B. Goennenwein, Phys Rev Lett 106, 117601 (2011). [2] L. Dreher, M. Weiler, M. Pernpeintner, H. Huebl, R. Gross, M. S. Brandt, and S. T. B. Goennenwein, Phys Rev B 86, 134415 (2012). [3] L. Thevenard, C. Gourdon, J. Y. Prieur, H. J. von Bardeleben, S. Vincent, L. Becerra, L. Largeau, and J. Y. Duquesne, Phys Rev B 90, 094401 (2014). [4] P. G. Gowtham, D. Labanowski, and S. Salahuddin, Phys Rev B 94, 014436 (2016). [5] R. Sasaki, Y. Nii, Y. Iguchi, and Y. Onose, Phys Rev B 95, 020407 (2017). [6] M. Weiler, H. Huebl, F. S. Goerg, F. D. Czeschka, R. Gross, and S. T. Goennenwein, Phys Rev Lett 108, 176601 (2012). [7] A. V. Azovtsev and N. A. Pertsev, Phys Rev B 94, 184401 (2016). [8] M. Xu, J. Puebla, F. Auvray, B. Rana, K. Kondou, and Y. Otani, Phys Rev B 97, 180301 (2018). [9] I. S. Camara, J. Y. Duquesne, A. Lemaître, C. Gourdon, and L. Thevenard, Phys Rev Appl 11, 014045 (2019). [10] S. Bhuktare, A. Bose, H. Singh, and A. A. Tulapurkar, Sci Rep 7, 840 (2017). [11] J. Holanda, D. S. Maior, A. Azevedo, and S. M. Rezende, Nat Phys 14, 500 (2018). [12] S. Bhuktare, A. S. Shukla, H. Singh, A. Bose, and A. A. Tulapurkar, Appl Phys Lett 114 (2019). [13] S. Streib, N. Vidal-Silva, K. Shen, and G. E. W. Bauer, Phys Rev B 99, 184442 (2019). [14] M. Matsuo, J. i. Ieda, K. Harii, E. Saitoh, and S. Maekawa, Phys Rev B 87, 180402 (2013). [15] S. Streib, H. Keshtgar, and G. E. W. Bauer, Phys Rev Lett 121, 027202 (2018). [16] V. Sampath, N. D'Souza, D. Bhattacharya, G. M. Atkinson, S. Bandyopadhyay, and J. Atulasimha, Nano Lett 16, 5681 (2016). [17] E. M. Chudnovsky and R. Jaafar, Phys Rev Appl 5, 031002 (2016). [18] D. Kobayashi, T. Yoshikawa, M. Matsuo, R. Iguchi, S. Maekawa, E. Saitoh, and Y. Nozaki, Phys Rev Lett 119, 077202 (2017). [19] D. A. Garanin and E. M. Chudnovsky, Phys Rev B 92, 024421 (2015). [20] E. M. Chudnovsky, D. A. Garanin, and R. Schilling, Phys Rev B 72, 094426 (2005). [21] S. J. Barnett, Phys Rev 6, 239 (1915). [22] J. H. Mentink, M. I. Katsnelson, and M. Lemeshko, Phys Rev B 99, 064428 (2019). [23] L. Zhang and Q. Niu, Phys Rev Lett 112, 085503 (2014). [24] E. Leader, Physics Letters B 756, 303 (2016). [25] W. Fu, Z. Shen, Y. Xu, C. L. Zou, R. Cheng, X. Han, and H. X. Tang, Nat Commun 10, 2743 (2019). [26] P. G. Gowtham, T. Moriyama, D. C. Ralph, and R. A. Buhrman, J Appl Phys 118 (2015). [27] Y. Long, J. Ren, and H. Chen, Proc Natl Acad Sci U S A 115, 9951 (2018). [28] I. A. Privorotskii, IEEE Trans Magn 16, 3 (1980). [29] I. A. Privorotskii, R. A. B. Devine, and G. C. Alexandrakis, J Appl Phys 50 (1979). [30] T. Makkonen, V. P. Plessky, W. Steichen, and M. M. Salomaa, Appl Phys Lett 82, 3351 (2003). [31] E.-J. Guo, J. Cramer, A. Kehlberger, C. A. Ferguson, D. A. MacLaren, G. Jakob, and M. Kläui, Phys Rev X 6, 031012 (2016). [32] I. a. Feng, M. Tachiki, C. Krischer, and M. Levy, J Appl Phys 53, 177 (1982). [33] I. P. Morton and H. M. Rosenberg, Phy Rev Lett 8, 200 (1962). [34] R. O. Pohl, Phys Rev Lett 8, 481 (1962). [35] A. Michels, J. Weissmüller, A. Wiedenmann, and J. G. Barker, J Appl Phys 87, 5953 (2000).
1912.08299	1	1912	2019-12-17T22:44:10	Millikelvin-compatible apparatus for studies of quantum materials under uniaxial stress	[ "physics.app-ph", "cond-mat.str-el" ]	Various new phenomena emerge in quantum materials under elastic deformations, such as hydrostatic or uniaxial stresses. In particular, using uniaxial strain or stress can help to tune or uncover specific structural or electronic orders in materials with multiple coexisting phases. Those phases may be associated with a quantum phase transition requiring a millikelvin environment combined with multiple experimental probes. Here, we describe our unique apparatus, which allows in situ tuning of strain in large samples inside a dilution refrigerator while the samples are monitored via an optical microscope. We describe the engineering details and show some typical results of characterizing superconducting strontium titanate under stress. This letter should serve as a practical reference for experts in ultra-low temperature experimental physics involving uniaxial stresses or strains.	physics.app-ph	physics	Millikelvin-compatible apparatus for studies of quantum materials under uniaxial stress Donovan Davino1, Jacob Franklin1 & Ilya Sochnikov1,2, a) *These authors contributed equally to this work. AFFILIATIONS 1Physics Department, University of Connecticut, Storrs Connecticut, USA 06269 2Institute of Material Science, University of Connecticut, Storrs Connecticut, USA 06269 a) Author to whom correspondence should be addressed. Electronic Mail: [email protected] Abstract Various new phenomena emerge in quantum materials under elastic deformations, such as hydrostatic or uniaxial stresses. In particular, using uniaxial strain or stress can help to tune or uncover specific structural or electronic orders in materials with multiple coexisting phases. Those phases may be associated with a quantum phase transition requiring a millikelvin environment combined with multiple experimental probes. Here, we describe our unique apparatus, which allows in situ tuning of strain in large samples inside a dilution refrigerator while the samples are monitored via an optical microscope. We describe the engineering details and show some typical results of characterizing superconducting strontium titanate under stress. This letter should serve as a practical reference for experts in ultra-low temperature experimental physics involving uniaxial stresses or strains. Introduction In a wide variety of quantum materials, strain plays an important role. For example, in strontium titanate, which is a low critical temperature (Tc) superconductor, questions about our fundamental understanding of electron lattice coupling and ways to enhance Tc are at the forefront of the modern quest to understand quantum phase transitions.[1] -- [7] In high temperature superconductors, pressure and strain may be the keys to potentially even higher temperature superconductivity.[8] For example, recent observations of the high-pressure-induced (compressive strain) record-breaking superconductivity in sulfur hydride at about 200 K are a vivid demonstration that strain and pressure effects are a promising control parameter to make significant advances in tuning superconducting transitions.[9] In magnetic systems, such as quantum liquid candidates and quantum spin-ices, strain and lattice effects may be the key in realizing new exotic quasiparticles. [10], [11] In topological quantum materials, the strain might be key to realizing dissipationless protected states that may be useful for future quantum computing approaches. [12] -- [15] Since many of these effects happen at millikelvin temperatures, it is important to develop strain cells that work reliably at such low temperatures in dilution refrigerators. This work presents such a millikelvin 1 apparatus with examples of strain effects detected via optical, magnetic and electronic means in compressed superconducting strontium titanate. Strain cell Figure 1 illustrates our thermally compensated copper strain cell. The cell has a copper lever that compresses or stretches the sample. The lever is controlled by a lead screw that is rotated by a high-torque, low-speed geared DC motor. The lead screw driving rod is fed through the entire height of the Bluefors LD-250 dilution refrigerator (Figure 2), with thermalization points at every cold plate, lead-bronze bearings at some low-temperature stages, and a high-vacuum ferrofluid rotational feedthrough at the room temperature stage. The strain in the samples is measured with a miniature gauge directly attached to the sample. [2], [6], [7] The cell was designed to accommodate the changes in strontium titanate sample size upon cooling with accuracy of about 0.01% in residual strain. We verified the accuracy by comparing the gauge readings on clamped samples to the readings of the strain gauges mounted on free- standing samples (data not shown). We typically mounted the samples with attached gauges in the copper strain cell using Stycast epoxy, leaving ~5-7 mm of the sample length for the probing of superconducting properties and optical imaging. Therefore, as-cooled samples typically had very close to zero strain. [2] Thus, this design provides small strains and minimizes dislocation defects in the sample. This performance metric of our strain device is comparable to that of commercial and non- commercial piezo-based devices, [16] but our copper clamp is essential for efficient thermalization at temperatures in a dilution refrigerator. We chose to develop an all-metal strain rig (lead screw clamping) due to better control of tensile strain and the ability to apply very large forces (>400 N). These forces translate into more than 1 GPa of stress on samples. The disadvantage which results from the larger size of our device is that it causes increased cooling times. However, this disadvantage is insubstantial, and the larger size also results in the advantage that our clamp allows us to attach resistive gauges directly to large samples for in situ monitoring of strain, as well as other probes such as optics, magnetic susceptibility and scanning SQUID (scanning SQUID data not shown). Even though we preferred to work with 4-10 mm long crystals when those were available, our clamp can accommodate samples that are about 1-20 mm long. The cross section of the samples should be determined according to the maximal desired strain or stress. The simulation (finite element analysis in Fusion 360) shown in Figure 1 was used to analyze the expected experimental conditions of the strain cell. The real-world strain cell was made from H04 Copper with a yield strength of about 240 Megapascal. The cell is then pre-loaded so that it can stretch or compress the sample through force applied to the lever arm by the lead screw. The simulation shows that at a preloaded angle of 1.4 degrees (measured from the lever arm to the base) 400N should be applied to displace the sample location on the lever arm by .75mm. At this point the end of the arm has been displaced 2.2 mm making it an arc symmetrical relative to the sample. One benefit of preloading the sample like this is that it allows for a more uniform relationship when force is applied to the lever arm compared to starting the lever arm completely parallel to the beam. Additionally, because 400 N has already been applied to the lever arm, when force is taken away the sample can then be stretched (tensile stress). The simulation shows that at 2 400N, a safety factor of 2 is reached in the copper parts (plastic deformation occurs below 1). This shows that there is still some room to add more force to the sample which allows for further compression while still being able to compress the sample. Figure 1. The copper strain compressive and tensile cell. (a) The geometry of the clamp, combined with the copper and stainless-steel parts, compensates for the various thermal shrinkages of STO 3 samples and the adjacent copper parts. The main material is oxygen-free copper for use in a dilution refrigerator. A simple cell design is critical to the success of the measurements of samples that are essentially free of residual strain when cooled to temperatures of 10s of mK. (b) Finite element stress analysis of the cell in preloaded state in units of MPa. The lever is parallel to the baseplate in the preloaded state allowing for uniform stress in the sample. (c) A view of the clamp from under and behind it to assist in the visualization of its geometry. The cell is able to provide >400 N of compressive and tensile force. No commercial cells can achieve these values at such low temperatures to the best of our knowledge. Optical monitoring of tetragonal domains in a dilution refrigerator In addition to the cell, we have optical, magnetic and electrical probes. The optical microscope is the most unique among the three and is described here. We constructed a polarizing optical microscope in our BF-250 dilution refrigerator using mostly conventional optical components. We took care to mount the lenses and the mirrors in as strain- free a manner as possible. Some lenses simply rested on stopping rings inside aluminum lens tubes to avoid cracking due to thermal contraction. We degreased and removed plastic parts from Thorlabs (New Jersey, USA) and Newport (California, USA) optomechanical components to make them more compatible with vacuum and cryogenics. Otherwise, the microscope has a conventional geometry (Figure 2): with an objective and ocular lens and a beam splitter for illumination in between. An LED source (Thorlabs MCWHF2) is located outside the cryostat. A silica HV vacuum-compatible optical fiber 800 m in diameter (Thorlabs MV63L1) carries the light. A polymer film with no glass layers polarizes the light and two additional fused silica lenses refocus it while passing it to a 50:50 beam splitter (Thorlabs BSW10) in a Köhler illumination scheme. We installed a series of infrared-opaque windows at each cold plate of the cryostat. The light focuses on the sample with a standard single AR-A 1'' diameter 250 mm focal length fused silica objective lens (with two adjustable rerouting mirrors), reflects back from the sample, passes through the same beam splitter, a standard AR-A fused silica 1" diameter 750 mm focal length condenser lens, passes through another standard polymer film polarizer, and then through a standard UHV fused silica AR-A viewport to a AmScope MU1403B 14MP High-Speed USB 3.0 CMOS camera. The camera acquired the images, while additional optomechanical components aligned the camera outside the cryostat. The total optical path is over one meter long, requiring very fine manual alignment. We obtained a spatial resolution of ~5-8 m from the microscope. Representative optical images of strontium titanate at a range of strains is shown in Figure 3. When the samples undergo the antiferrodistortive phase transition around 110 K, large domain regions form. We determined that samples can be easily detwinned under stress in the (110) direction. Figure 3 shows 0.5x1.5x10 mm3 multi-domain vs. detwinned states of the sample. Further compression resulted in suppression of the critical temperature in this strontium titanate sample, as detected in the resistive and magnetic transition (Figure 3), all measured simultaneously. The changes in superconductivity are interpreted as being due to influence of strain on the phonon modes and the proximity to the ferroelectric quantum phase transition.[2], [6] Additional data from 4 other samples and reproducibility of the results related to the sample mounting procedure are discussed elsewhere.[2], [6] Figure 2. Polarized optical microscopy in the Bluefors dilution refrigerator to image the strained samples in situ. (1) 50 K stage with an optical IR opaque window. (2) 4K stage with copper thermalization touching the lead screw. (3) Illumination beam splitter at 1 K stage. (4) 1 K stage with a thermalized brass bearing. (5) A flexible connection in the lead screw. (6) Part of the lead screw. (7) Reinforcement copper plate. (8) The objective lens in adjustable lens tube. (9) Two 900 adjustable mirrors. (10) The strain clamp with the sample illuminated. Even with all the probes attached the base temperature of <25 mK can be reached (illumination off). 5 Figure 3. Simultaneously acquired optical images (a, b) as well as magnetic and resistive signatures (c) of the superconducting transition in an x=0.014 sample oriented with (110)c face normal to the stress direction. The long 6-mm side (and the visible edge) of the sample was parallel to [110]c. The sample was 0.5 mm thick along [001]c and 2 mm wide along [110]c. Samples detwinned under (110)c compression with the c-axis aligning out of the images plane in the [001]c direction. (a) shows an as-cooled sample's surface at 40 K. The visible edge is parallel to the (110) stress. (b) shows a detwinned sample at -0.015% [110]c strain. Imaged area: 1.5x1.5 mm2. The superconducting transition is strongly suppressed, and the microscopy allows us to determine the orientation of the antiferrodistortive axis which is essential to understand the effect. Summary and Conclusions The apparatus presented here opens a new page in investigations of the interplay between strain and quantum orders. Our typical strontium titanate data set shows a reliable testing platform for 6 behaviors across various tuning parameters and signatures of quantum orders. Several orders -- superconducting, structural, polar, and magnetic can be detected. Further work includes installation of a scanning Superconducting Quantum Interference Device (SQUID) microscope at the strain clamp. [17], [18] References [1] C. Collignon, B. Fauqué, A. Cavanna, U. Gennser, D. Mailly, and K. Behnia, "Superfluid density and carrier concentration across a superconducting dome: The case of strontium titanate," Phys. Rev. B, vol. 96, no. 22, p. 224506, Dec. 2017. [2] C. Herrera, J. Cerbin, K. Dunnett, A. V. Balatsky, and I. Sochnikov, "Strain-engineered interaction of quantum polar and superconducting phases," arXiv:1808.03739 [cond-mat], Aug. 2018. [3] K. Ahadi, L. Galletti, Y. Li, S. Salmani-Rezaie, W. Wu, and S. Stemmer, "Enhancing superconductivity in SrTiO3 films with strain," Sci Adv, vol. 5, no. 4, p. eaaw0120, Apr. 2019. [4] A. Stucky et al., "Isotope effect in superconducting n-doped SrTiO3," Sci. Rep., vol. 6, p. 37582, Nov. 2016. [5] Y. Tomioka, N. Shirakawa, K. Shibuya, and I. H. Inoue, "Enhanced superconductivity close to a non-magnetic quantum critical point in electron-doped strontium titanate," Nature Comm., vol. 10, no. 1, p. 738, Feb. 2019. [6] C. Herrera and I. Sochnikov, "A Local Maximum in the Superconducting Transition Temperature of Nb-Doped Strontium Titanate Under Uniaxial Compressive Stress," J. Supercond. Nov. Magn., Aug. 2019. [7] C. Herrera and I. Sochnikov, "Precision Measurements of the AC Field Dependence of the Superconducting Transition in Strontium Titanate," J. Supercond. Nov. Magn., Sep. 2019. [8] A. Gozar et al., "High-temperature interface superconductivity between metallic and insulating copper oxides," Nature, vol. 455, no. 7214, pp. 782 -- 785, Oct. 2008. [9] A. P. Drozdov, M. I. Eremets, I. A. Troyan, V. Ksenofontov, and S. I. Shylin, "Conventional superconductivity at 203 Kelvin at high pressures in the sulfur hydride system," Nature, vol. 525, p. 73, Aug. 2015. [10] R. Yadav, S. Rachel, L. Hozoi, J. van den Brink, and G. Jackeli, "Strain- and pressure-tuned magnetic interactions in honeycomb Kitaev materials," Phys. Rev. B, vol. 98, no. 12, p. 121107, Sep. 2018. [11] M. Maesato, Y. Shimizu, T. Ishikawa, G. Saito, K. Miyagawa, and K. Kanoda, "Spin-liquid behavior and superconductivity in K-(BEDT-TTF)2X: The role of uniaxial strain," J. Phys. IV France, vol. 114, pp. 227 -- 231, Apr. 2004. [12] M. König et al., "Quantum Spin Hall Insulator State in HgTe Quantum Wells," Science, vol. 318, no. 5851, p. 766, Nov. 2007. [13] Z. Tajkov, D. Visontai, L. Oroszlány, and J. Koltai, "Uniaxial strain induced topological phase transition in bismuth -- tellurohalide -- graphene heterostructures," Nanoscale, vol. 11, no. 26, pp. 12704 -- 12711, 2019. [14] H. Aramberri and M. C. Muñoz, "Strain effects in topological insulators: Topological order and the emergence of switchable topological interface states in Sb2Te3/Bi2Te3 heterojunctions," Phys. Rev. B, vol. 95, no. 20, p. 205422, May 2017. 7 [15] I. Sochnikov et al., "Nonsinusoidal Current-Phase Relationship in Josephson Junctions from the 3D Topological Insulator HgTe," Phys. Rev. Lett., vol. 114, no. 6, p. 066801, Feb. 2015. [16] C. W. Hicks, M. E. Barber, S. D. Edkins, D. O. Brodsky, and A. P. Mackenzie, "Piezoelectric-based apparatus for strain tuning," Review of Scientific Instruments, vol. 85, no. 6, p. 065003, Jun. 2014. [17] P. G. Björnsson, B. W. Gardner, J. R. Kirtley, and K. A. Moler, "Scanning superconducting quantum interference device microscope in a dilution refrigerator," Review of Scientific Instruments, vol. 72, no. 11, pp. 4153 -- 4158, Nov. 2001. [18] I. Sochnikov et al., "Direct Measurement of Current-Phase Relations in Superconductor/Topological Insulator/Superconductor Junctions," Nano Lett., vol. 13, no. 7, pp. 3086 -- 3092, Jul. 2013. 8
1708.05411	1	1708	2017-08-08T15:19:16	All-optical lithography process for contacting atomically-precise devices	[ "physics.app-ph", "cond-mat.mes-hall" ]	We describe an all-optical lithography process that can be used to make electrical contact to atomic-precision donor devices made in silicon using scanning tunneling microscopy (STM). This is accomplished by implementing a cleaning procedure in the STM that allows the integration of metal alignment marks and ion-implanted contacts at the wafer level. Low-temperature transport measurements of a patterned device establish the viability of the process.	physics.app-ph	physics	All-optical lithography process for contacting atomically-precise devices D. R. Ward,1 M. T. Marshall,1 D. M. Campbell,1 T. M. Lu,1 J. C. Koepke,1 D. A. Scrymgeour,1 E. Bussmann,1 and S. Misra1, a) Sandia National Laboratories, New Mexico 87185 (Dated: 25 March 2018) We describe an all-optical lithography process that can be used to make electrical contact to atomic-precision donor devices made in silicon using scanning tunneling microscopy (STM). This is accomplished by implement- ing a cleaning procedure in the STM that allows the integration of metal alignment marks and ion-implanted contacts at the wafer level. Low-temperature transport measurements of a patterned device establish the viability of the process. Keywords: Methods of micro- and nanofabrication and processing The ability to fabricate devices with atomic precision holds promise for revealing the key physics underlying everything from quantum bits1,2 to ultra-scaled digi- tal circuits3–7. A common atomic-precision fabrication (APFab) pathway uses a scanning tunneling microscope (STM) to create lithographic patterns on a hydrogen- passivated Si(100) surface8. Phosphine gas introduced into the vacuum system selectively adsorbs on sites where Si dangling bonds have been re-exposed by patterning9, yielding atomically precise, planar structures made of P donors. Unlike electron beam lithography (EBL), which can pattern hydrogen with a resolution of around 100 nm and is unable to image the pattern10, the STM is an ideal instrument for this process because it can both pattern and image the hydrogen resist with atomic precision11. However, STMs are typically capable of patterning de- vices only up to 10 µm by 10 µm in size, which are too small to directly contact. A post-patterning microfabri- cation process, consisting of etching via holes in an encap- sulating Si overlayer and then depositing metal in direct contact with the planar donor layer, is used to make elec- trical contact to the devices. Even the largest features made with the STM are small enough that this contacting process relies on EBL for patterning and 200 nm- scale processing. At this scale, making good electrical contact between a deposited metal and an atomically-thin one- dimensional line of donors at the edge of an etched hole is challenging, and even successful EBL process flows in this application are rate-limiting. In this paper, we detail an all-optical lithography con- tacting process that reduces the time of fabricating an atomic-precision device by an order of magnitude. This is made possible by the integration of both ion-implanted contacts and metal alignment marks in the starting ma- terial, which bridge the scale between the largest regions accessible by STM and the smallest length scale acces- sible by low-cost photolithography. Specifically, the ion- implanted contacts neck down to a small enough area that the STM can place the APFab device in direct con- tact with them, and extend out to a region large enough a)Electronic mail: [email protected] FIG. 1. (a) STM sample cleaning procedure, including oxy- gen plasma clean (left), atomic hydrogen clean (center), and oxide removal anneal (right). (b) STM topographs, taken at a tunnel junction setpoint of -2.5 V and 200 pA, show a clean Si(100) surface. that multiple photolithography steps done to a precision of 2 µm can connect the APFab device to 200 µm sized metal pads. Directly fabricating APFab donor structures on top of ion implanted Si simplifies the burden on micro- fabrication to that of making contact between deposited metal and an ion implanted region, which can be done with near perfect yield. Moreover, this entire contact- ing process can be executed in a single day using tools available in most clean-rooms, and can be run on mul- tiple chips in parallel. Moving forward, the increased throughput of our reliable all-optical contacting process promises to dramatically reduce the cost of making new discoveries using APFab devices. The standard process for cleaning Si(100) samples for STM patterning12 precludes the integration of metal alignment marks and ion implanted contacts, since the caustic chemical pre-cleaning ex situ and in vacuo an- nealing to 1200°C would destroy them. As illustrated in Figure 1a, we have adopted a modified cleaning proce- dure and applied it to APFab. All the Si(100) samples used in this work were first cleaned ex situ in an ultra- sonic bath of acetone and isopropyl alcohol to remove Si(100)Si(100)SiOResidual CarbonHResidual CarbonSi(100)(a)(b)10 nm1 µm2SiO2600 C800 CSiOO0.80210ÅÅoo2 This cleaning procedure enables the integration of ion- implanted contacts and metal alignment marks in the starting material, at the wafer level, as illustrated in Fig- ure 2a. Alignment marks are first etched in the mate- rial, a p-type Si(100) wafer having a volume resistivity of 10-20 Ω cm and covered in 10 nm of sacrificial oxide (not shown). A photoresist mask is then used to per- form a selective implant of As ions at 40 keV and an areal density of 3x1015 ions/cm2. These ion-implanted contacts start from a 40 µm by 40 µm field, large enough for aligning subsequent photolithography steps, and neck down to a 8 µm by 8 µm field, small enough for the STM to contact directly. The resist is then stripped and the sacrificial oxide removed using HF, exposing a pristine surface. Following RCA1, RCA2 , and HF cleans14, a 10 nm steam oxide is grown at 850°C to protect the sam- ple. Due to damage from the ion implantation, oxide grows at roughly 6 times the rate in implanted regions compared to pristine ones. The oxidation process is fol- lowed by a 15 minute anneal at 850°C in nitrogen. A final photoresist mask is used to deposit tungsten align- ment markers (Figure 2b). Critically for APFab, this process produces atomically clean surfaces once the re- sultant chips are subjected to the sample clean outlined above (Figure 1b). By avoiding any acid-based cleaning, metal alignment marks are not damaged. Limiting the flash to 800°C prevents any significant diffusion of the ion-implanted contacts. While ion-implanted contacts have been implemented before15, the process flow relied on interdigitated implanted contacts across large regions of the chip, and is thus limited to simple two-terminal devices. No such limitations exist for the process flow we have implemented. Flashing the sample to 800°C for 5 minutes is suffi- cient to remove the 10 nm of surface oxide, but leaves the implanted contacts buried in oxide. Since STM cannot tunnel into a thick insulator, the sample must be flashed for a longer period of time to expose enough doped Si to connect directly with hydrogen lithography. Figure 2c shows a topographic image of a sample after it is flashed to 800°C for 15 minutes. Rather than a uniform reduc- tion of all the thick oxide in the implanted region, the partial removal of the thick oxide proceeds from the edge of the implanted region. This leaves a 30-60 nm deep trench that is easily identified by STM. Leveraging the high contrast metal markers and a high-resolution optical camera, we can align the tip precisely to the implanted contacts with sub 2 µm precision. To determine whether the implanted dopants diffuse out of the implanted re- gion, we simultaneously acquired topographic and spec- troscopic data using the STM in Figure 2d. This data indicates that there is a region inside the trench that has an enhanced tunneling density of states, corresponding to a high concentration of activated donors16. Moreover, this enhanced density of states is sharply confined to the trench, indicating that As dopants have not diffused out from the implant region. Because the STM tip can be aligned to the high con- FIG. 2. (a) Side view of the process flow for integrating ion- implanted contacts and metal marks into the starting wafer. Here, Si is shown in blue, oxide in teal, implanted regions in green, and metal markers in brown. (b) Illustration of device fabrication process covering the 170 x 170 µm device area of the die, including the ion-implanted region (green) and the high contrast tungsten alignment markers (brown). (c) Atomic force microscope topographic image of the cen- ter of the implanted region of a chip after the sample prepa- ration routine outlined in Figure 1. STM data in (d) was taken in the white box before sample was removed from vac- uum. The raised plateaus at the edge of the field of view are the remnant oxide from the middle of the ion implant. (d) False color overlay of STM differential conductance on si- multaneously acquired topographic data of the ion-implanted region. Topographic data was taken with a tunnel junction setpoint of -1.5 V and 800 pA; the differential conductivity was recorded simultaneously using a lockin-amplifier at a fre- quency of 511 Hz and an ac excitation of 50 mV. The speck- led green areas in between the implants are an artifact of noise- they do not appear in both the forward-scanned and backward-scanned frames of the data (not shown). remnant photoresist. An oxygen plasma clean at 100 W of power for 20 minutes removes most of the remaining hydrocarbon debris from the surface. After inserting the samples into the STM vacuum chamber, we degas the samples by heating them successively to 450°C for 20 minutes and then 600°C for 40 minutes. Subsequently exposing the samples to atomic hydrogen removes the remaining trace carbon on the surface, based on the pro- cess described in Ref. 13. For this, a tungsten filament at 1700°C in a background pressure of 5x10−6 torr hydrogen gas generates atomic hydrogen while the sample is heated to 600°C for 20 minutes. Finally, we heat the sample to a modest 800°C to remove the surface oxide. The resultant surfaces are atomically clean on small length scales and show no contamination on larger length scales (Figure 1b). 050100036912036X(µm)Y(µm)Z (nm)134.6dI/dV (nS)(d)6 µm0Z (nm)200(c)(b)(a)W markerOxidizeImplantCleanDeposit3 FIG. 3. (a) APFab process flow, including hydrogen termination (green), phosphine exposure (red) and silicon capping. (b) STM image, taken with a tunnel junction setpoint of -3 V and 200 pA, of a ∼2000 nm by 500 nm depassivated region. A Sobel filter has been used to enhance edge contrast. The green arrow points to a step edge, and yellow arrows point out the depassivated region connecting contacts 3 and 4 (right-side, top and bottom in c) (c and d) Plan and cross-sectional view of the process used to make contact to the APFab device. Here, Si is shown in blue, oxide in teal, implanted regions in green, metal markers in brown, the P APFab device in red, and aluminum contacts in grey. We wire bond directly to the aluminum contacts. trast metal marks using a long-focus optical microscope, hydrogen lithography can be used to make direct con- tact to the implanted contacts, which neck into a square that is 8 µm on a side. The samples are first terminated with atomic hydrogen, which serves as the monolayer re- sist for this process (Figure 3a). The hydrogen termina- tion is accomplished using a tungsten filament and filling the chamber to a pressure of 2x10−6 torr, while holding the sample at 300°C. This hydrogen resist can both be imaged by STM at low junction biases (1-3 V), or re- moved at higher junction biases with either atomic pre- cision (∼3-5 V, 10 nm/s tip speed) or more coarsely (∼7- 10 V, 200 nm/s tip speed)8,11,17. In Figure 3b, we have patterned a 500 nm wide wire between two implanted contacts. The resultant exposed dangling bonds selec- tively adsorb phosphine when it is introduced into the chamber9. We apply a total dose of 15 L at a chamber pressure of 2x10−8 torr. A thermally activated surface decomposition reaction of the phosphine, at a tempera- ture that leaves the hydrogen resist intact, results in P donors incorporated into the lattice at a density rang- ing between 17%18 for the smallest windows (1 donor in a 3 dimer window) and 25% for large areas19. For the device in Figure 3b, this will result in a P nanowire con- necting the two As implanted regions. To preserve the atomically-precise donor-based device, it is then encapsu- lated in Si deposited at a rate of 1 nm/s to a thickness of 30 nm, while holding the sample at 350°C. From sample preparation to encapsulation, fabricating an atomic pre- cision device that fans out to contact pads in a roughly 8 µm by 8 µm region takes about 12 hours, with the two most time-intensive parts being large STM scans to lo- cate the implants, and patterning of the coarse features, which has been computer-automated20. The process of making electrical connection to the APFab device is now simplified as compared to EBL methods, requiring only optical lithography and standard clean-room microfabrication to put metal in direct con- tact with the eight ion-implanted contacts in a 40 µm by 40 µm area (Figure 3d). Etching down to the implanted Si is complicated by the material stack in that part of the sample, which starts with the Si capping layer, fol- lowed by oxide which was incompletely removed during the sample flashing process, and finally by the doped Si. Contacts are made by patterning 2 µm diameter vias with optical lithography followed by a reactive ion etch of the Si capping layer using CF4 at 25°C. Next, the leads are patterned for a lift-off metal deposition. Immediately before metal deposition a relatively long, 90 s etch in 1:6 BOE (buffered oxide etch) is used to remove the re- maining oxide in the vias over the ion-implanted regions. After depositing 150 nm of Al, by electron beam deposi- tion, a standard metal lift-off process is used to complete the eight contacts that fan out into bond pads, shown schematically in Figure 3c. Electrical transport data on the simple nanowire de- vice shown in Figure 3b establishes the validity of this approach to making APFab devices (Figure 4). These measurements must be carried out below ∼50 K to freeze out the carriers in the substrate. At 4 K the only resis- tive elements in the path of the current are cables, the contacts to the 2D device layer, and the APFab nanowire itself. The DC transport through the nanowire is Ohmic down to tens of micro-Volts, and gives a resistance of 5.6 kΩ. Accounting for the resistance of the nanowire (4 squares of P doped Si, whose resistivity is typically (b)(d)(a)Si(100)300 CH terminate350 CPH dose33Pattern HIncorporate PSi capHPHSi500 nm(c)12345678Pattern P layerCap with SiDeposit Al contactsEtch Si and oxideoo525 Ω/(cid:50)), the total contact resistance to the P device layer comes to 1.75 kΩ per contact, which includes the metal-implant interface, the resistivity of the implanted region itself, and the implant-P interface. Most impor- tantly, we have found this method to produce a high yield of successful contacts; all eight contacts across three dif- ferent chips have been shown to work in the same manner as Figure 4. The high yield is attributed to the fact that implanted contacts represent an extended surface that deposited metal can come into contact with, including the bottom surface of the via holes. This is in contrast to trying to directly contact the two-dimensional P layer itself, in which case the metal needs to make a line con- tact at some location up the side-wall of the via hole. We also examine, in Figure 4b, the transport between contacts that are not connected by a patterned APFab structure. These show a miniscule amount of leakage between isolated contact pairs- less than 0.1 nA at 2 V of bias. Two control samples- one which saw the same thermal processing as our APFab device but no phos- phine dose, and a second one which was not subjected to any thermal processing- show similar levels of leakage current to one another. This indicates that the thermal budget of our process does not lead to enough As implant diffusion to be measurable. This also suggests that the additional leakage current between isolated contacts in the patterned sample originates largely from phosphine adsorbing through imperfections in the hydrogen resist. Both the Ohmic conduction through the nanowire, and the small leakage between unconnected pairs of contacts, compare well to an earlier effort which realized metal sili- cide contacts, but reported nonlinear I-V curves through an Ag nanowire with much larger leakage between un- connected contacts.21 In conclusion, we have presented a new all-optical method for contacting APFab devices that takes a single day to execute, easily allows multiple chips to be pro- cessed in parallel, and achieves a high yield of successful contacts. This was made possible by adopting a pro- cess for cleaning Si(100) in an STM that is sufficiently low-temperature to allow for the integration of tungsten metal alignment markers and ion implanted contacts in the starting material. The fast processing time and high yield are expected to dramatically reduce the costs of developing new innovations with APFab. The Digital Electronics at the Atomic Limit project is supported by the Laboratory Directed Research and Development Program at Sandia National Laboratories, and was performed, in part, at the Center for Integrated Nanotechnologies, an Office of Science User Facility oper- ated for the U.S. Department of Energy (DOE) Office of Science. Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technol- ogy and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA-0003525. 4 FIG. 4. (a) Current-voltage curve taken of the nanowire pat- terned in Figure 3b between contacts 3 and 4. (b) Current- voltage curves taken between all other contact pairs on the same chip (blue), which had no P nanostructure connecting them. Also shown are current-voltage curves from a chip pro- cessed in the same way, but not dosed with phosphine (red), and a third chip that not subjected to any in-vacuum pro- cessing at all (black). All data were taken at 4 K by biasing a single contact and collecting the current on a neighboring con- tact while leaving the other contacts electrically floated. The absolute value of the current is plotted for negative biases. 1H. Buch, S. Mahapatra, R. Rahman, A. Morello, and M. Y. Simmons, Nature Comm. 4, 2017 (2013). 2T. F. Watson, B. Weber, M. G. House, H. Buch, and M. Y. Simmons, Phys. Rev. Lett. 115, 166806 (2015). 3B. Weber, S. Mahapatra, H. Ryu, S. L. an A. Fuhrer, T. C. G. Reusch, D. L. Thompson, W. C. T. Lee, G. Klimeck, L. C. L. Hollenberg, and M. Y. Simmons, Science 335, 64 (2012). 4M. Fueschle, J. A. Miwa, S. Mahapatra, H. Ryo, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nature Nano. 7, 242 (2012). 5M. Rudolph, S. M. Carr, G. Subramania, G. T. Eyck, J. Dominguez, T. Pluym, M. P. Lilly, M. S. Carroll, and E. Buss- mann, Appl. Phys. Lett 105, 163110 (2014). 6N. Pascher, S. Hennel, S. Mueller, and A. Fuhrer, New J. of Phys. 18, 083001 (2016). 7G. Cheng, P. F. Siles, F. Bi, C. Cen, D. F. Bogorin, C. W. Bark, C. M. Folkman, J.-W. Park, C.-B. Eom, G. Medeiros-Ribiero, and J. Levy, Nature Nano. 6, 343 (2011). 8J. W. Lyding, T.-C. Shen, J. S. Hubacek, J. R. Tucker, and G. C. Abeln, Appl. Phys. Lett. 64, 2010 (1994). 9J. L. O'Brien, S. R. Chofield, M. Y. Simmons, R. G. Clark, A. S. Dzurak, N. J. Curson, N. S. McAlpine, M. E. Hawley, and G. W. Brown, Phys. Rev. B 64, 161401 (2001). 10T. Hallam, M. J. Butcher, K. E. J. Goh, F. J. Reuss, and M. Y. Simmons, J. Appl. Phys. 102, 034308 (2007). 11T.-C. Shen, C. Wang, G. C. Abeln, J. R. Tucker, J. W. Lyding, P. Avouris, and R. E. Walkup, Science 268, 1590 (1995). 12M. B. W. B. S. Swartzentruber, Y.-W. Mo and M. G. Lagally, J. Vac. Sci. Technol. A 7, 2901 (1989). 13A. Assmuth, T. Stimpel-Lindner, O. Senftleben, A. Bayerstadler, T. Siluma, H. Baumgartner, and I. Eisele, Appl. Surf. Sci. 253, 8389 (2007). 14W. Kern and D. A. Pauotien, RCA Rev. 31, 187 (1970). 15J. C. Kim, J. S. Kline, and J. R. Tucker, Appl. Surf. Sci. 239, 335 (2005). 16P. Studer, S. R. Schofield, C. F. Hirjibehedin, and N. J. Curson, Appl. Phys. Lett. 102, 012107 (2013). 17J. B. Ballard, T. W. Sisson, J. H. G. Owen, W. R. Owen, E. Fuchs, J. A. J. N. Randall, and J. R. V. Ehr, J. Vac. Sci. Tech. B 31, 06FC01 (2013). 18S. R. Schofield, N. J. Curson, M. Y. Simmons, F. J. Ruess, T. Hal- -4-20241-10Bias (mV)Current (µA)012-2-1-12-11-10Bias (V)log Current (A)(a)(b)R = 5.6 kΩR > 20 GΩ101010lam, L. Oberbeck, and R. G. Clark, Phys. Rev. Lett. 91, 136104 (2003). 19L. Oberbeck, N. J. Curson, M. Y. Simmons, R. Brenner, A. R. Hamilton, S. R. Schofield, and R. G. Clark, Appl. Phys. Lett. 81, 3197 (2002). 20D. R. Ward, S. Misra, D. A. Scrymgeour, R. J. Simonson, M. T. Marshall, J. C. Koepke, E. Bussmann, M. S. Carroll, J. B. Bal- lard, J. H. G. Owen, S. W. Schmucker, E. Fuchs, S. Pryad- kin, and J. N. Randall, Sandia Tech. Rep. SAND2016, 5470 C (2016). 21M. Fujimori, S. Heike, Y. Terada, and T. Hashizume, Nanotech. 15, S333 (2004). 5
1708.09709	2	1708	2017-10-04T12:29:24	Computational polarimetric microwave imaging	[ "physics.app-ph" ]	We propose a polarimetric microwave imaging technique that exploits recent advances in computational imaging. We utilize a frequency-diverse cavity-backed metasurface, allowing us to demonstrate high-resolution polarimetric imaging using a single transceiver and frequency sweep over the operational microwave bandwidth. The frequency-diverse metasurface imager greatly simplifies the system architecture compared with active arrays and other conventional microwave imaging approaches. We further develop the theoretical framework for computational polarimetric imaging and validate the approach experimentally using a multi-modal leaky cavity. The scalar approximation for the interaction between the radiated waves and the target---often applied in microwave computational imaging schemes---is thus extended to retrieve the susceptibility tensors, and hence providing additional information about the targets. Computational polarimetry has relevance for existing systems in the field that extract polarimetric imagery, and particular for ground observation. A growing number of short-range microwave imaging applications can also notably benefit from computational polarimetry, particularly for imaging objects that are difficult to reconstruct when assuming scalar estimations.	physics.app-ph	physics	Computational polarimetric microwave imaging THOMAS FROMENTEZE,1,2,* OKAN YURDUSEVEN,2 MICHAEL BOYARSKY,2 JONAH GOLLUB,2 DANIEL L. MARKS,2 AND DAVID R. SMITH2 1Xlim Research Institute, University of Limoges, 87060 Limoges, France 2Center for Metamaterials and Integrated Plasmonics, Department of Electrical and Computer Engineering, Duke University, Durham, 27708, North Carolina, USA *[email protected] Abstract: We propose a polarimetric microwave imaging technique that exploits recent advances in computational imaging. We utilize a frequency-diverse cavity-backed metasurface, allowing us to demonstrate high-resolution polarimetric imaging using a single transceiver and frequency sweep over the operational microwave bandwidth. The frequency-diverse metasurface imager greatly simplifies the system architecture compared with active arrays and other conventional microwave imaging approaches. We further develop the theoretical framework for computational polarimetric imaging and validate the approach experimentally using a multi-modal leaky cavity. The scalar approximation for the interaction between the radiated waves and the target -- often applied in microwave computational imaging schemes -- is thus extended to retrieve the susceptibility tensors, and hence providing additional information about the targets. Computational polarimetry has relevance for existing systems in the field that extract polarimetric imagery, and particular for ground observation. A growing number of short-range microwave imaging applications can also notably benefit from computational polarimetry, particularly for imaging objects that are difficult to reconstruct when assuming scalar estimations. © 2021 Optical Society of America OCIS codes: (110.5405) Polarimetric imaging; (110.1758) Computational imaging ; (110.3200) Inverse scattering References and links 1. E. C. Fear, X. Li, S. C. Hagness, and M. A. Stuchly, "Confocal microwave imaging for breast cancer detection: Localization of tumors in three dimensions," IEEE Transactions on Biomedical Engineering 49, 812 -- 822 (2002). 2. T. Rubaek, P. M. Meaney, P. Meincke, and K. D. Paulsen, "Nonlinear microwave imaging for breast-cancer screening using gauss -- newton's method and the cgls inversion algorithm," IEEE Transactions on Antennas and Propagation 55, 2320 -- 2331 (2007). 3. T. C. Williams, J. M. Sill, and E. C. Fear, "Breast surface estimation for radar-based breast imaging systems," IEEE Transactions on Biomedical Engineering 55, 1678 -- 1686 (2008). 4. N. K. Nikolova, "Microwave imaging for breast cancer," IEEE microwave magazine 12, 78 -- 94 (2011). 5. G. Bellizzi, O. M. Bucci, and I. Catapano, "Microwave cancer imaging exploiting magnetic nanoparticles as contrast agent," IEEE Transactions on Biomedical Engineering 58, 2528 -- 2536 (2011). 6. A. Abbosh, B. Mohammed, and K. Bialkowski, "Differential microwave imaging of the breast pair," IEEE Antennas and Wireless Propagation Letters 15, 1434 -- 1437 (2016). 7. D. M. Sheen, D. L. McMakin, and T. E. Hall, "Three-dimensional millimeter-wave imaging for concealed weapon detection," IEEE Transactions on microwave theory and techniques 49, 1581 -- 1592 (2001). 8. D. Sheen, D. McMakin, and T. Hall, "Near-field three-dimensional radar imaging techniques and applications," Applied Optics 49, E83 -- E93 (2010). 9. X. Zhuge and A. G. Yarovoy, "A sparse aperture mimo-sar-based uwb imaging system for concealed weapon detection," IEEE Transactions on Geoscience and Remote Sensing 49, 509 -- 518 (2011). 10. S. S. Ahmed, A. Schiessl, and L.-P. Schmidt, "A novel fully electronic active real-time imager based on a planar multistatic sparse array," IEEE Transactions on Microwave Theory and Techniques 59, 3567 -- 3576 (2011). 11. S. S. Ahmed, A. Schiessl, F. Gumbmann, M. Tiebout, S. Methfessel, and L. Schmidt, "Advanced microwave imaging," IEEE microwave magazine 13, 26 -- 43 (2012). 12. Y. Rodriguez-Vaqueiro, Y. A. Lopez, B. Gonzalez-Valdes, J. A. Martinez, F. Las-Heras, and C. M. Rappaport, "On the use of compressed sensing techniques for improving multistatic millimeter-wave portal-based personnel screening," IEEE Transactions on Antennas and Propagation 62, 494 -- 499 (2014). 13. B. Gonzalez-Valdes, Y. Álvarez, Y. Rodriguez-Vaqueiro, A. Arboleya-Arboleya, A. García-Pino, C. M. Rappaport, F. Las-Heras, and J. A. Martinez-Lorenzo, "Millimeter wave imaging architecture for on-the-move whole body imaging," IEEE Transactions on Antennas and Propagation 64, 2328 -- 2338 (2016). 14. E. J. Baranoski, "Through-wall imaging: Historical perspective and future directions," Journal of the Franklin Institute 345, 556 -- 569 (2008). 15. M. Dehmollaian, M. Thiel, and K. Sarabandi, "Through-the-wall imaging using differential sar," IEEE transactions on geoscience and remote sensing 47, 1289 -- 1296 (2009). 16. T. S. Ralston, G. L. Charvat, and J. E. Peabody, "Real-time through-wall imaging using an ultrawideband multiple- input multiple-output (mimo) phased array radar system," in "Phased Array Systems and Technology (ARRAY), 2010 IEEE International Symposium on," (IEEE, 2010), pp. 551 -- 558. 17. Y. Wang and A. E. Fathy, "Advanced system level simulation platform for three-dimensional uwb through-wall imaging sar using time-domain approach," IEEE Transactions on Geoscience and Remote Sensing 50, 1986 -- 2000 (2012). 18. S. Caorsi, A. Massa, M. Pastorino, and M. Donelli, "Improved microwave imaging procedure for nondestructive evaluations of two-dimensional structures," IEEE Transactions on Antennas and Propagation 52, 1386 -- 1397 (2004). 19. M. Benedetti, M. Donelli, A. Martini, M. Pastorino, A. Rosani, and A. Massa, "An innovative microwave-imaging technique for nondestructive evaluation: applications to civil structures monitoring and biological bodies inspection," IEEE Transactions on Instrumentation and Measurement 55, 1878 -- 1884 (2006). 20. S. Kharkovsky and R. Zoughi, "Microwave and millimeter wave nondestructive testing and evaluation-overview and recent advances," IEEE Instrumentation & Measurement Magazine 10, 26 -- 38 (2007). 21. U. Hasar, "Non-destructive testing of hardened cement specimens at microwave frequencies using a simple free-space 22. D. Carsenat and C. Decroze, "Uwb antennas beamforming using passive time-reversal device," IEEE Antennas and method," NDT & E International 42, 550 -- 557 (2009). Wireless Propagation Letters 11, 779 -- 782 (2012). 23. T. Fromenteze, O. Yurduseven, M. F. Imani, J. Gollub, C. Decroze, D. Carsenat, and D. R. Smith, "Computational imaging using a mode-mixing cavity at microwave frequencies," Applied Physics Letters 106, 194104 (2015). 24. T. Fromenteze, E. L. Kpré, D. Carsenat, C. Decroze, and T. Sakamoto, "Single-shot compressive multiple-inputs multiple-outputs radar imaging using a two-port passive device," IEEE Access 4, 1050 -- 1060 (2016). 25. J. Hunt, T. Driscoll, A. Mrozack, G. Lipworth, M. Reynolds, D. Brady, and D. R. Smith, "Metamaterial apertures for computational imaging," Science 339, 310 -- 313 (2013). 26. G. Lipworth, A. Mrozack, J. Hunt, D. L. Marks, T. Driscoll, D. Brady, and D. R. Smith, "Metamaterial apertures for coherent computational imaging on the physical layer," JOSA A 30, 1603 -- 1612 (2013). 27. O. Yurduseven, J. N. Gollub, D. L. Marks, and D. R. Smith, "Frequency-diverse microwave imaging using planar mills-cross cavity apertures," Optics express 24, 8907 -- 8925 (2016). 28. D. L. Marks, J. Gollub, and D. R. Smith, "Spatially resolving antenna arrays using frequency diversity," JOSA A 33, 899 -- 912 (2016). 29. J. Gollub, O. Yurduseven, K. Trofatter, D. Arnitz, M. Imani, T. Sleasman, M. Boyarsky, A. Rose, A. Pedross-Engel, H. Odabasi, T. Zvolensky, G. Lipworth, D. Brady, D. L. Marks, M. S. Reynolds, and D. R. Smith, "Large metasurface aperture for millimeter wave computational imaging at the human-scale," Scientific reports 7, 42650 (2017). 30. O. Yurduseven, D. L. Marks, T. Fromenteze, J. N. Gollub, and D. R. Smith, "Millimeter-wave spotlight imager using dynamic holographic metasurface antennas," Optics Express 25, 18230 -- 18249 (2017). 31. T. Fromenteze, X. Liu, M. Boyarsky, J. Gollub, and D. R. Smith, "Phaseless computational imaging with a radiating metasurface," Optics Express 24, 16760 -- 16776 (2016). 32. W. C. Chew, Waves and fields in inhomogeneous media, vol. 522 (IEEE press, 1995). 33. F. T. Ulaby and C. Elachi, "Radar polarimetry for geoscience applications," Norwood, MA, Artech House, Inc., 1990, 376 p. No individual items are abstracted in this volume. 1 (1990). 34. R. N. Treuhaft and P. R. Siqueira, "Vertical structure of vegetated land surfaces from interferometric and polarimetric radar," Radio Science 35, 141 -- 177 (2000). 35. J.-S. Lee and E. Pottier, Polarimetric radar imaging: from basics to applications (CRC, 2009). 36. D. Velotto, F. Nunziata, M. Migliaccio, and S. Lehner, "Dual-polarimetric terrasar-x sar data for target at sea observation," IEEE Geoscience and Remote Sensing Letters 10, 1114 -- 1118 (2013). 37. R. K. Amineh, A. Khalatpour, and N. K. Nikolova, "Three-dimensional microwave holographic imaging using co-and cross-polarized data," IEEE Transactions on Antennas and Propagation 60, 3526 -- 3531 (2012). 38. R. Salman, I. Willms, L. Reichardt, T. Zwick, and W. Wiesbeck, "On polarization diversity gain in short range uwb-radar object imaging," in "Ultra-Wideband (ICUWB), 2012 IEEE International Conference on," (IEEE, 2012), pp. 402 -- 406. 39. J. Wang, P. Aubry, and A. Yarovoy, "A novel approach to full-polarimetric short-range imaging with copolarized data," IEEE Transactions on Antennas and Propagation 64, 4733 -- 4744 (2016). 40. D. M. Sheen, D. L. McMakin, W. M. Lechelt, and J. W. Griffin, "Circularly polarized millimeter-wave imaging for personnel screening," in "Defense and Security," (International Society for Optics and Photonics, 2005), pp. 117 -- 126. 41. D. M. Sheen, D. L. McMakin, W. M. Lechelt, and J. W. Griffin, "Circularly polarized millimeter-wave imaging for personnel screening," in "Proc. SPIE," , vol. 5789 (2005), vol. 5789, pp. 117 -- 126. 42. A. Abubakar, P. M. Van den Berg, and J. J. Mallorqui, "Imaging of biomedical data using a multiplicative regularized contrast source inversion method," IEEE Transactions on Microwave Theory and Techniques 50, 1761 -- 1771 (2002). 43. P. J. Basser, J. Mattiello, and D. LeBihan, "Mr diffusion tensor spectroscopy and imaging," Biophysical journal 66, 259 -- 267 (1994). 44. D. Le Bihan, J.-F. Mangin, C. Poupon, C. A. Clark, S. Pappata, N. Molko, and H. Chabriat, "Diffusion tensor imaging: concepts and applications," Journal of magnetic resonance imaging 13, 534 -- 546 (2001). 45. S. Mori and J. Zhang, "Principles of diffusion tensor imaging and its applications to basic neuroscience research," Neuron 51, 527 -- 539 (2006). 46. J. D. Jackson, Classical electrodynamics (Wiley, 1999). 47. G. Lipworth, A. Rose, O. Yurduseven, V. R. Gowda, M. F. Imani, H. Odabasi, P. Trofatter, J. Gollub, and D. R. Smith, "Comprehensive simulation platform for a metamaterial imaging system," Applied optics 54, 9343 -- 9353 (2015). 48. T. Fromenteze, C. Decroze, and D. Carsenat, "Passive uwb multiplexing device for a single-port mimo radar," in "Microwave Symposium (IMS)," (2014), pp. 1 -- 4. 49. R. Barrett, M. Berry, T. F. Chan, J. Demmel, J. Donato, J. Dongarra, V. Eijkhout, R. Pozo, C. Romine, and H. Van der Vorst, Templates for the solution of linear systems: building blocks for iterative methods (SIAM, 1994). 50. O. Yurduseven, M. F. Imani, H. Odabasi, J. Gollub, G. Lipworth, A. Rose, and D. R. Smith, "Resolution of the frequency diverse metamaterial aperture imager," Progress In Electromagnetics Research 150, 97 -- 107 (2015). 51. T. Fromenteze, E. L. Kpré, C. Decroze, D. Carsenat, O. Yurduseven, M. Imani, J. Gollub, and D. R. Smith, "Unification of compressed imaging techniques in the microwave range and deconvolution strategy," in "Radar Conference (EuRAD), 2015 European," (IEEE, 2015), pp. 161 -- 164. 52. P.-S. Kildal, K. Rosengren, J. Byun, and J. Lee, "Definition of effective diversity gain and how to measure it in a reverberation chamber," Microwave and Optical Technology Letters 34, 56 -- 59 (2002). 53. D. A. Hill and J. M. Ladbury, "Spatial-correlation functions of fields and energy density in a reverberation chamber," IEEE Transactions on Electromagnetic Compatibility 44, 95 -- 101 (2002). 54. C. Draeger and M. Fink, "One-channel time reversal of elastic waves in a chaotic 2d-silicon cavity," Physical Review Letters 79, 407 (1997). 55. G. Montaldo, D. Palacio, M. Tanter, and M. Fink, "Time reversal kaleidoscope: A smart transducer for three- dimensional ultrasonic imaging," Applied physics letters 84, 3879 -- 3881 (2004). 56. O. Yurduseven, T. Fromenteze, D. L. Marks, J. N. Gollub, and D. R. Smith, "Frequency-diverse computational microwave phaseless imaging," IEEE Antennas and Wireless Propagation Letters (2017). 57. J. J. van Zyl, M. Arii, and Y. Kim, "Model-based decomposition of polarimetric sar covariance matrices constrained for nonnegative eigenvalues," IEEE Transactions on Geoscience and Remote Sensing 49, 3452 -- 3459 (2011). Introduction 1. Recent advances by the microwave community have resulted in the development of innovative imaging modalities for medical diagnosis [1 -- 6], concealed threat detection [7 -- 13], through- wall imaging [14 -- 17], and non-destructive testing [18 -- 21]. Radio-frequency (RF) bands are particularly suited to these applications, since RF waves can penetrate through many materials that are opaque at optical frequencies. Furthermore, since RF waves are non-ionizing, they are considered safe for human exposure at suitably low power levels. The potential for efficient, cost-effective, and high-resolution systems that can achieve fast acquisition rates have recently been demonstrated in computational imaging systems based on cavity-backed [22 -- 24] and metasurface [25 -- 30] apertures. These systems radiate pseudo- orthogonal field distributions in transmission and -- by exploitation of the reciprocity principle -- in reception, to multiplex information and reconstruct an image. The frequency-diverse aperture limits the complexity of the hardware architecture required for real-time high-resolution imaging, obviating the actively controlled components or the need for mechanical motion that is typically required in conventional systems. Recently, frequency-diverse metasurface apertures were adapted for intensity-only measurements, demonstrating that the phase information can be coded in the dispersive radiation of these complex, but passive, antennas and solved for within the inverse problem [31]. This allows further simplification of the associated measurement electronics. However, the application of such computational systems to polarimetric imaging remains unexplored. Polarization is a relevant source of information in radar imaging, allowing determination of the nature of the interaction between in scene targets and the spatial components of the radiated electromagnetic fields [32]. Polarimetric radars were initially and widely implemented in geoscience applications, allowing for the interrogation of relevant parameters unavailable to single-polarized radars (soil classification, soil moisture, snow distribution, etc.) [33 -- 36]. Following the growing interest in short-range imagery, polarization sensitive radars have been adapted for exploiting the vectorial nature of electromagnetic waves. These approaches thus improve the quality of the reconstructed images, enhancing the accuracy of the estimated contrasts and revealing target features invisible to the approaches based on single-polarized measurements [37 -- 39]. The elimination of artifacts due to multipath in concealed weapon detection has also been demonstrated by illuminating targets using a combinations of left and right-hand circular polarizations [40, 41]. Finally, full-polarimetric imaging has found recent applications in medical diagnosis [42]. However, we note that the technical constraints associated with the development of polarimetric radars are exacerbated when striving for high-resolution and real-time systems, therefore making full reconstruction of polarimetric images extremely burdensome. In this paper, we propose to extend the framework of microwave computational imaging to the measurement of polarization information in short-range applications by encoding the susceptibility of the target in the physical layer of the antenna, i.e. as a single frequency-dependent signal. In the Section 2 of this article, the theoretical principle of polarimetric computational imaging is first introduced. A generic polarization-sensitive metasurface is considered for simplifying the architecture required for retrieving vectorial information in the target space. The cavity-backed metasurface developed to this end is then presented in the Section 3. A proof of concept of short-range imaging is then presented for validating this novel principle in the Section 4. Finally Section 5 provides the concluding remarks. 2. Computational polarimetry: the interrogation of a susceptibility tensor from a signal coded in the physical layer We first describe the underlying principle of polarimetric microwave imaging, using the illustrative geometry shown in Fig. 1. Fig. 1. Polarimetric radar system: Two arrays made of dual-polarized antennas are used in transmission and reception for interrogating the susceptibility tensor ¯χ(r) in the target space. To estimate the image, the fields radiated by each antennas must be known -- either measured or derived (e.g. from the propagation of the surface currents with the dyadic Green's functions G(r − rt,r , ν). In this conceptual example, two arrays of dual-polarized antennas are used to measure the interaction between radiated waves and a target. In the configuration shown, the field is generated +¯χ(r)G(r−rt,ν)G(r−rr,ν)TxArrayRxArrayxzyby the transmit array and the scattered fields are detected by the receive array. The electric ¯χ(r) accounts for the interaction between the radiated waves and the susceptibility tensor illuminated scene. In a conventional computational imaging approach, the field radiated and measured by a pair of transmit and receive antennas would interact with the target susceptibility, generating the frequency-dependent signal ¯S(rt, rr, ν). This 2 × 2 matrix contains the interaction between the excited x or z-oriented element in transmission at location rt with the x or z-oriented element in reception at location rr, where x and z are coordinates in the plane of the arrays (both arrays are assumed to be in the same plane). In the first Born approximation, the interaction between the dyadic fields radiated by the x-polarized antenna in transmission, with the z-polarized antenna in reception is: Sx,z(rt, rr, ν) = Ex(rt, r, ν) ¯χ(r) Ez(rr, r, ν)T d3r (1) The vector fields Ex(rt, r, ν) and Ez(rr, r, ν) are radiated by each of the transmit and receive antennas from their respective locations rt and rr. The total measured frequency-dependent signal can be expressed in matrix form for each dual-polarized transmit and receive pair as: r ¯S(rt, rr, ν) = ¯E(rt, r, ν) ¯χ(r) ¯E(rr, r, ν)T d3r (2) Þ Þ r where ¯E is a 2 × 3 matrix representing the three-dimensional vector field radiated by two co-localized transverse polarized antennas. Assuming that these fields are determined either by an experimental measurement or from a numerical analysis, the susceptibility tensor can then be interrogated by solving the inverse problem corresponding to the forward model specified by Eq. (2). Controlling spatially the radiated polarized fields over two apertures is particularly burdensome in applications where real-time and high-resolution constraints are imposed. The RF architectures associated with each array would thus have to include as many active chains as dual-polarized antennas, working simultaneously over an ultra-wide bandwidth. Fast switches would also need to be implemented for selecting the excited polarization of each antenna leading to three-dimensional vector fields. In this paper, the polarimetric imaging approach is simplified by implementing a computational technique (Fig. 2). The two arrays are replaced by metasurfaces, which radiate pseudo-orthogonal patterns in space and frequency. In this way, the large number of active chains implemented for the independent transmission and reception of waves (multiplied by two for a polarization sensitive system) is substantially decreased. Instead a single compressed signal can be measured through the input and output ports and the radiating antennas. The amount of information included in a set of these measurements directly depends on the spatial diversity of the fields emitted and received by the antennas [28], and is now studied considering the potential contribution of polarization diversity. The expression of the fields E(rt,r, r, ν) radiated respectively by the transmit and receive arrays is defined by the expression of the equivalent vector potentials At,r(rt,r, ν), according to the corresponding current distributions Jt,r(rt,r, ν) in the radiating apertures that At,r(rt,r, ν) satisfies the Helmoltz equation [32]: ∇2 At,r(rt,r, ν) + k2 At,r(rt,r, ν) = −µ0 Jt,r(rt,r, ν) (3) The expression of At,r(r, ν) is thus obtained by computing the convolution product between the current distribution Jt,r(rt,r, ν) and a scalar free-space Green's function g(r, ν) in the antenna space rt,r: Fig. 2. Radiating metasurface implemented for the demonstration of microwave computational polarimetry. The current distribution is shown on the aperture and represents an excitation of the transmission port at a single frequency. The metasurface is conceived to obtain pseudo-orthogonal current distributions in the frequency domain. At,r(r, ν) = µ0 The expression of the electric field is finally given by [32]: rt,r Jt,r(rt,r, ν) g(r − rt,r, ν) d3rt,r Þ (cid:2)∇(∇ · At,r(r, ν)) + k2 At,r(r, ν)(cid:3) Þ Jt,r(rt,r, ν) ¯G(r, rt,r, ν) d3rt,r Et,r(r, ν) = − jc k = j2πν µ0 rt,r (4) (5) (6) where k = 2πν/c is the wavenumber and ¯G(ra, rb, ν) is the dyadic Green's function modeling the propagation from ra to rb, written as: ¯G(ra, rb, ν) = {g(ra − rb, ν)} ¯I + (cid:18) (cid:19) (7) ∇∇ k2 The susceptibility tensor ¯χ(r) can be estimated from the interaction between the fields radiated by two metasurfaces [28], leading to the following measured frequency signal: Þ Þ r r s(ν) = = jπ0 ν jπ0 ν Er(r, ν) ¯χ(r) Et(r, ν)T d3r Tr[ ¯χ(r) Er(r, ν) Et(r, ν)T] d3r (8) (9) The symmetry of the dyadic Green's function matrix leads to ¯G(r, rt, ν) = ¯G(r, rt, ν)T . As demonstrated in [28], the outer product Er(r, ν) Et(r, ν)T forms a basis interrogating the reflectivity of the scene. In previous studies of computational imaging using a signal coded by the physical layer, a scalar approximation was used for the susceptibility tensor, allowing for an estimation Jt(rt,ν)Jr(rr,ν)+¯χ(r)G(r−rt,ν)G(r−rr,ν)Jt(rt,ν)Jr(rr,ν)+¯χ(r)G(r−rt,ν)G(r−rr,ν)of a target reflectivity illuminated by several metasurfaces. Here, we extend this framework to a tensorial reconstruction of the susceptibility in the target space, passively encoding the three-dimensional polarization information into a unique frequency signal measured between two ports. This signal is simplified when restricting the study to anisotropic media defined by a diagonalized susceptibility tensor. From a fundamental perspective, we consider electrons within the target medium interacting with the incident electric field as being represented by a simple mechanical model (Fig. 3). Fig. 3. Description of the susceptibility tensor with a mechanical model of the bound electron. The induced polarization corresponds to a motion of these particles reacting to an electric field. This tensor is diagonalized by the compensation of the rotation between the radiated wave's axis (red) and the main axis of the electron (black) confined in its movements by the interaction with the surrounding particles. In the depicted model, the principal axes of the bound electron (x(cid:48), y(cid:48), z(cid:48)) are not aligned with the coordinate axes (x, y, z), the latter defined as references for the transmitted and received electromagnetic fields. Thus, the non-diagonal terms are created by projections of the radiated field over the main axis of the tensor at each location, leading to the following eigendecomposition: ¯χ(r) = ¯R(r) diag(cid:0)ξ(r)(cid:1) ¯R(r)T (10) where ¯R(r) is a rotation matrix and ξ(r) is the diagonalized susceptibility tensor. Assuming the first Born approximation, a forward model is defined that links the measurements to the susceptibility tensor defined at each location of the target space. The principal axes of each tensor can thus be estimated from a diagonalization determining the orientation of the currents on the targets, which is homogenized over the size of the reconstructed voxels. Finally, These estimations can be represented using the weighted eigenvectors or equivalent ellipsoids, showing the anisotropy of the scattering (Fig. 4). Similar studies are notably developed for the estimation of anisotropy in the the field of diffusion tensor imaging [43 -- 45]. Having introduced the essential aspects of the computational imaging, we take a closer look at the radiation of polarized fields by metasurfaces. Indeed, simplifications can be applied when considering antennas radiating in only one direction of the space. The radiated field can thus be deduced by the measurement of the tangential electric field on a synthetic aperture plane with t,r (r, ν) [46]. From this field, we can extract a set of vanishing boundary conditions, denoted Etan equivalent dipoles, as e−y0x0z0yxzFig. 4. Ellipsoids representing the eigenvalues/eigenvectors decomposition of the electric susceptibility tensor estimated for each voxel. The anisotropic scattering is represented by an ellipsoid with axis of unequal lengths, depicting the local orientation of the tensor. (cid:16) Þ Þ (cid:48) r t,r (cid:48) t,r, ν)(cid:17) t,r, ν)(cid:17) (cid:48) (cid:16) y · Etan t,r (r n × Etan t,r (r d2r (cid:48) t,r d2r (cid:48) t,r pt,r(rt,r) = 0 n 1 mt,r(rt,r) = j µπν (cid:48) r t,r where pt,r(rt,r) are the equivalent electric dipoles and mt,r(rt,r) the magnetic ones, computed in (cid:48) transmission and reception over small 2D domains r t,r, effectively defining a grid where each facet is centered at rt,r. The vector n is defined as the normal to the measurement plane and y is a unit vector perpendicular to the plane (optical axis). For the sake of simplicity, the transmit and receive metasurfaces can be defined as co-located and oriented such that n = y for each radiating plane. The electric dipoles pt,r are thus canceled out by the dot product y · Etan t,r = 0, allowing for a decomposition of the field in the target space as the superposition of contributions radiated by magnetic dipoles only, as depicted in Fig. 5. In this example, different distributions of random magnetic dipoles are radiating according to the excitation port and to the frequency. The expression of the radiated electric field in the target space is finally expressed from the contributions of all the magnetic dipoles, summing over the discrete dipole space rt,r [46,47]: (11) (12) (13) (14) Et,r(r, ν) = − µ0πν2 c with the cross-product given as: r × mt,r(rt,r) =  0 z −y [r × mt,r(rt,r)] g(rt,r − r, ν) rt,r −z 0 x  mx 0 mz  =  y −x 0  y mz z mx − x mz x my − y mx The projection property of the cross product allows for a random distribution of magnetic sources contained in a plane to be sufficient for radiating 3D polarized electric near fields that are able to interrogate the full susceptibility tensor. A pseudo-random distribution of sources in the flat metasurface aperture thus radiates a complex field able to probe the medium of interest, λ1~v1λ2~v2λ3~v3Isotropicscatteringλ1~v1λ2~v2λ3~v3AnisotropicscatteringFig. 5. Analysis of the field generated by one dipole of the transmitting (a) and receiving (b) current distributions at the same location r0. ¯χ(r) = ν encoding the susceptibility variation of the target space into a unique frequency-dependent signal. This system is able to passively multiplex the target space information in transmission and reception, extending earlier approaches based on a scalar approximation of the susceptibility tensor in [24,48]. The susceptibility of the target space can be estimated by a pseudo-inversion of the radiated fields in transmission and reception of the form Er(r, ν)+s(ν) Et(r, ν)+T ν jπ0 (15) where Et,r(r, ν)+ is the pseudo-inverse computed for each independent polarization state on the discrete representation of the radiated fields in transmission and reception [49 -- 51], determined t,r (r, ν) can be determined from from Eqs. (12) and (13) and assuming that the tangential field Etan an analytical model (or measured). We stress that these estimations can only be achieved efficiently if these fields present a low level of correlation, spatially, and for each polarization state. Conveniently, we can exploit frequency diversity to encode the measured information onto the ports connected to the metasurface antenna. To be more specific, we propose to develop the expression of the sub-element ¯χ(m,n) of the retrieved susceptibility tensor at the location r0: ν ¯χ(m,n)(r0) = = Þ p,q Er,m(r0, ν)+s(ν) Et,n(r0, ν)+T ν jπ0 ¯χ(p,q)(r0) R(m,n, p,q)(r0) (16) (17) where R(m,n, p,q)(r0) is a function of the degree of correlation in space and polarization of the radiated fields, and is defined as: R(m,n, p,q)(r0) = Er,m(r0, ν)+Er, p(r, ν) Et,q(r, ν)T Et,n(r0, ν)+T d3r (18) An ideal estimation of the full susceptibility tensor at this specific location would thus impose ν r mtiEti(r0,ν)r0•¯χ(r0)mrjErj(r0,ν)r0•¯χ(r0)(a)Transmission(b)ReceptionmtiEti(r0,ν)r0•¯χ(r0)mrjErj(r0,ν)r0•¯χ(r0)(a)Transmission(b)ReceptionmtiEti(r0,ν)r0•¯χ(r0)mrjErj(r0,ν)r0•¯χ(r0)(a)Transmission(b)Receptionto ensure that (cid:40)1, 0 if (m, n) = (p, q) if (m, n) (cid:44) (p, q) R(m,n, p,q)(r0) = (19) In practice, the sum of non-zero cross-terms can thus introduce a speckle noise impacting the accuracy of the estimation. In general, the reconstructions will thus highly depend on the technique implemented for solving the inverse problem formulated in Eq. (15), using direct or iterative strategies of reconstruction. In this article, all the results were computed by implementing an iterative least-square algorithm. 3. Cavity-backed polarization-sensitive metasurface We validate the computational polarimetric imaging technique using the frequency-diverse, chaotic leaky microwave cavity illustrated in Fig. 6. Fig. 6. Radiating metasurface implemented for the validation of the proposed computational polarimetric imaging technique. The radiated near-field is measured on both transverse polarizations with a single-polarized open-ended waveguide probe. The radiated patterns must exhibit a low level of correlation among the radiated field patterns in frequency, space, and polarization -- ensuring an appropriate multiplexing of the polarimetric information gathered from the target space and encoding it into a single frequency vector. These properties are obtained by exploiting the modal diversity in a leaky cavity designed to be several times larger than the operating wavelength. The frequency bandwidth is thus set between 17.5 and 26.5 GHz, corresponding to wavelengths of 1.15 cm to 1.67 cm, for a cavity with inner dimensions of 15.2 × 28.5 × 28.5 cm3. This type of over-sized reverberating cavity finds applications in various domains such as the evaluation of the gain of antennas or for electromagnetic vulnerability measurements [52,53]. A metallic ball is inserted into this air-filled cavity to optimize the modal diversity, to create irregular and convex boundaries and prevent the development of degenerate states [54,55]. As opposed to previous microwave computational imaging demonstrations [23,31] also based on air-filled leaky cavities, this work introduces a means of estimating the vector response of the target space instead of its scalar approximation. In addition to the low level of correlation of the fields in space and frequency that is required in the previous experiments, we also require an additional constraint of pseudo-orthogonality in polarization. The various correlations in frequency, space and polarization can be probed by analyzing the near field scans 1,2(rt,r, ν) obtained by exciting the Ports 1 and 2 connected to the cavity successively (Fig. 7). of Etan For the measurements, a commercial near-field scanner was used, NSI 200V-3x3 [23, 27]. ρ1(ν)ρ2(ν)15cm15cm28.5cm28.5cm15.2cmNear-fieldscanpathFig. 7. Near-field scans measured by a raster scan back-propagated to the aperture of the metasurface. The real part of the fields are represented for each excitation port and two consecutive frequencies just above 25 GHz, spaced by only 2.5 MHz. The measured field distribution is equivalent to the one depicted in Fig. 5, since the tangential fields and the magnetic dipoles are directly related (Eq. 12). One can observe that both the location of the feeding port and the frequency, even when the shift represents a tiny fraction of the operating frequency, have a noticeable impact on the near field distribution in the aperture of the metasurface. Before the introduction of any imaging experiment, the correlation between the radiated fields is studied. This analysis is initiated by considering only the field radiated on the x-polarization when feeding port 1. Near field scans are recorded over 4001 frequency points between 17.5 GHz and 26.5 GHz, leading to a frequency sampling interval of 2.25 MHz. Pearson's formula is used for the computation of the correlation, computed between two scalar zero-mean fields E1(rt, ν) and E2(rt, ν) as: C1,2(rt, rt) = E[(E1)]E[(E2)] (20) where E[Ei] and σi are the expectation and the standard deviation of Ei, respectively, computed over the frequency dimension for each spatial coordinate r. The autocorrelation of the first scan is depicted in Fig. 8. σ1σ2 -0.0500.05x (m)-0.0500.05z (m)Port 1 - f = 25.00150 GHz-0.0500.05x (m)-0.0500.05z (m)Port 2 - f = 25.00150 GHz-0.0500.05x (m)-0.0500.05z (m)Port 1 - f = 25.00375 GHz-0.0500.05x (m)-0.0500.05z (m)Port 2 - f = 25.00375 GHzFig. 8. Autocorrelation of the processed near field scan obtained on the x-polarization when feeding the port 1 of the cavity. The pre-processed fields, initially back-propagated to a 18.65 × 18.65 cm2 area and sampled using 63 points over the x and z axes, are downsampled to 16 × 16 points per scans for obtaining a spatial grid of 12 mm, close to the central wavelength of the operating bandwidth λmean = 13.6 mm. The two-dimensional coordinate rt is vectorized on each axis of this figure. The depicted result, corresponding mainly to a diagonal line linking the components sharing the same index, denotes the low level of spatial correlation in this scan obtained by exploiting the frequency diversity in the same way as in a computational imaging experiment. Two low level parallel line are obtained on both sides of the central diagonal, corresponding to the spatial correlation of adjacent points. This phenomenon has already been partially mitigated by resampling the field on a grid closer to the operating wavelength. This study is then extended to both ports and both polarizations, leading to 16 combinations of correlations given in Fig. 9. This global representation of the spatial correlation of the near fields for each couple of ports and polarizations presents the properties expected for interrogating the susceptibility tensor in the target space: the four autocorrelations are close to diagonal matrices, while all the other cross correlations are low. In this way, the polarization information is encoded by the radiation of the metasurface in transmission and reception, and can be retrieved from the inverse formulation by exploiting the frequency diversity. Fig. 9. Spatial correlation of the near field scans represented for each couple of ports and polarizations. The axis have been removed to save space and are equivalent for each sub-figure to these of Fig. 8 4. Practical implementation We investigate the polarimetric imaging capabilities of the system using a set of 2 mm thick copper wires as 8 × 12 cm2 targets, bent to form the four letters of "DUKE" shown in Fig. 10. A target made of thin conductive elements is expected to have an anisotropic scattering according to the orientation of the wires, revealed by the proposed computational imaging system. The letters are positioned at a distance of 30 cm in front of the radiating aperture and sequential measurements are performed using a vector network analyzer connected to ports 1 and 2 of the cavity. For each measured frequency signal, the inverse problem stated in Eq. 15 is computed using an iterative generalized minimal residual method. Fig. 10. Copper wire letters used as targets. Each letter is 15 × 9 cm2. Port 1 - Pol. xPort 1 - Pol. xPort 1 - Pol. zPort 2 - Pol. xPort 2 - Pol. zPort 1 - Pol. zPort 2 - Pol. xPort 2 - Pol. z00.10.20.30.40.50.60.70.80.91The estimation of the first susceptibility tensor ¯χ is shown in Fig. 11 where, first, only the magnitude is displayed. The computation is performed for the 3 × 3 interactions between all the polarization components in a target space discretized in 51200 voxels. The transverse axes x and z are both sampled on a 2.5 mm grid of 80 elements and the optical axis y is sampled on a 5 mm grid of 8 elements. The susceptibility tensor is thus a complex matrix of dimensions ¯χ ∈ C3×3×80×8×80. The cross-polarized symmetric elements of the tensor are averaged to facilitate numerical processing and to reduce the speckle level. Fig. 11. Three-dimensional estimation of the full susceptibility tensor ¯χ of the first target. The opacity of each voxel is coded on a log scale with a -15 dB minimum threshold. The two transverse co-polarized terms "xx" and "zz" have higher amplitudes and allow for the reconstruction of a continuous target. Interestingly, the contribution of the co-polarized "yy" term is quite significant despite a higher level of speckle. To the best of our knowledge, this feature has never been exploited in conventional polarimetric microwave imaging where most applications are restricted to far-field synthetic aperture radars. In a short-range scenario, the paraxial approximation is no longer valid -- leading to an increase amount of measurable information projected along the optical axis. The remaining elements of the susceptibility tensor have a smaller amplitude since they correspond to polarization conversions. This first representation only allows for a study of the magnitude of the tensor in the 3D target space. The resolution of the reconstructed images is directly defined, as in conventional radar imaging, by the size of the radiating aperture and the operating bandwidth [56].This analysis is thus continued by extracting the elements in the target plane to consider the impact of both the magnitude and the phase distributions (Fig. 12). Fig. 12. Susceptibility tensor extracted from the target plane. The opacity of the pixels corresponds to the linear magnitude of the tensor. The phase is color-coded. 4.1. Correlation of the co-polarized transversed terms The results of the two transverse co-polarized images "xx" and "zz" seem similar, but noticeable phase differences are obtained on the horizontal and vertical parts of the target due to the anisotropic nature of the scattering on these thin copper wires. A simple visualization of the anisotropy is thus proposed by computing the correlation χcorr between the two co-polarized transverse terms χx, x and χz,z : (21) χcorr(r) = χx, x χ ∗ z,z The result is presented in Fig. 13, controlling the opacity and the color of each pixel respectively with the magnitude and the phase of χcorr. Similar methods are applied in the field of synthetic aperture radar polarimetry by computing covariance matrices for reducing the speckle noise and extracting relevant soil parameters [35,57]. With this simple processing, the orientation of the wires can easily be retrieved according to the phase value, i.e. close to −π/2 for the vertical direction and π/2 for the horizontal ones. The elements oriented diagonally are represented by the white color standing for a 0 radian phase. However, this approach does not allow for the discrimination of the direction of the diagonal orientations of the wires. A comparable analysis can be performed for a three-dimensional visualization of χcorr, displaying a -20 dB isosurface of the reconstructed data and using a color coded representation of the phase (Fig. 14). Once again, the interference between the two transverse co-polarized terms allows for the Fig. 13. Correlation of two co-polarized transverse components of χcorr. The opacity of the figure corresponds to the magnitude of χcorr and the color coding of the phase of χcorr. Fig. 14. Three-dimensional estimation of χcorr. A -20 dB isosurface is represented, color- coded according to the phase of χcorr. reduction of the speckle noise and the visualization of the anisotropic scattering. However, such an approach does not make it possible to take into account all the information contained in the reconstructed susceptibility tensor. A more complex processing is studied based on the eigendecomposition of the susceptibility tensor to directly identify the main axis of each voxel. 4.2. Eigendecomposition of the susceptibility tensor The ratio between the small distance of the targets (30 cm) and the size of the radiating aperture (15 cm) makes it possible to obtain polarimetric components oriented along the optical axis. An eigendecomposition is performed on the real and imaginary parts of the retrieved susceptibility tensor. Each voxel of the target space is represented by a 3D ellipsoid oriented and scaled according to the associated eigenvalues and eigenvectors. The ellipsoids are color-coded according to the orientation of the main axis. The results extracted from the target plane are depicted in Fig. 15. In contrast to the previous section where the phase information was used to represent the anisotropic nature of the scattering, this additional analysis relies on the amplitude of the set of components composing the susceptibility tensor, where the real and imaginary parts are processed independently. Some ellipsoids reconstructed on the targets clearly correspond to anisotropic signatures, with an elongated shape corresponding to a eigenvalue larger than the other two, while some voxels have a signature closer to an isotropic response, represented by more spherical ellipsoids. Although the principal axes of the ellipsoids are not necessarily all oriented in the directions of the copper wires, it is possible to discern a direct correlation between the shape of the targets and the colors of the ellipsoids. In particular, the red elements oriented along the optical axis mainly occur for diffraction on the edges of the targets. Fig. 15. Set of ellipsoids obtained by computing eigendecompositions of the real (left) and imaginary (right) parts of the retrieved susceptibility tensor. 5. Conclusion A polarimetric microwave imaging technique has been presented in this paper by extending the computational principle of the scalar approaches that have been previously developed in the literature. The theoretical principle proposed in the introduction suggests a key simplification of the architectures for RF systems that seek to achieve high-resolution images from a single or few source system. The leaky multi-modal cavity used to produce pseudo-orthogonal field patterns in frequency and polarization can be generalized to many metasurface aperture paradigms -- all of which are capable of generating field patterns with low correlation as a function of frequency or other parameters.The multiplexing of an increased quantity of information compared to the scalar case requires the development of radiating structures with high quality factors. It is possible in this context to imagine the application of such a technique using a set of metasurfaces operating in a cooperative manner in order to reduce the radiation efficiency and modal diversity constraints imposed on the unique element presented in this proof of concept. In the present example, measurements were taken between two ports of a multi-modal cavity over the band 17.5-26.5 GHz. From the frequency-index measurements, it was possible to reconstruct targets made of copper wires forming the word "DUKE", demonstrating the different modes of operation permitted by the proposed polarimetric approach. Through the two last sections, the anisotropic behavior of the scattering has been highlighted by computing the correlation between the tranverse co-polarized components of the susceptibility tensor, which also allows the reduction of speckle noise, and by computing independent eigendecompositions of the real and imaginary parts of the full susceptibility tensor.. Funding This work was supported by the Air Force Office of Scientific Research (AFOSR, Grant No. FA9550-12-1-0491).
1801.02397	1	1801	2018-01-08T11:59:15	Engineering on-surface spin crossover: spin-state switching in a self-assembled film of vacuum sublimable functional molecule	[ "physics.app-ph", "cond-mat.mes-hall" ]	Realization of spin crossover (SCO) based applications requires studying of spin state switching characteristics of SCO complex molecules at nanostructured environments especially on-surface. Except for a very few cases, the SCO of a surface bound thin molecular film is either quenched or heavily altered due to (i) strong molecule-surface interactions and (ii) differing intermolecular interactions in films relative to the bulk. By fabricating SCO complexes on a weakly interacting surface such as highly oriented pyrolytic graphite (HOPG) and copper nitride (CuN), the interfacial quenching problem has been tackled. However, engineering intermolecular interactions in thin SCO active films is rather difficult. This work proposes a molecular self-assembly strategy to fabricate thin spin switchable surface bound films with programmable intermolecular interactions. Molecular engineering of the parent complex system [Fe(H$_{2}$B(pz)$_{2}$)$_{2}$(bpy)] (pz = pyrazole, bpy = 2,2'-bipyridine) with a dodecyl (C$_{12}$) alkyl chain yielded a classical amphiphile-like functional and vacuum sublimable charge neutral Fe$^{\rm II}$ complex, [Fe(H$_{2}$B(pz)$_{2}$)$_{2}$(C$_{12}$-bpy)] (C$_{12}$-bpy = dodecyl[2,2'-bipyridine]-5-carboxylate). The bulk powder and 10 nm thin film, on quartz glass/SiO$_{\rm x}$ surface, of the complex showed comparable spin state switching characteristics mediated by similar lamellar bilayer like self-assembly/molecular interactions in both bulk and thin film states. This unprecedented observation augurs well for the development of SCO based applications, especially in molecular spintronics.	physics.app-ph	physics	Engineering on-surface spin crossover: spin-state switching in a self- assembled film of vacuum sublimable functional molecule Kuppusamy Senthil Kumar,1* Michał Studniarek,1,2 Benoît Heinrich,1 Jacek Arabski,1 Guy Schmerber,1 Martin Bowen,1 Samy Boukari,1 Eric Beaurepaire,1 Jan Dreiser,2 Mario Ruben3* 1 Institut de Physique et Chimie des Matériaux (IPCMS), Université de Strasbourg, F-67034 Strasbourg, France, * e-mail : [email protected] 2 Swiss Light Source, Paul Scherrer Institut (PSI), CH-5232 Villigen, Switzerland, 3 Institute of Nanotechnology (INT), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, D-76344, Eggenstein-Leopoldshafen, Germany. * email: [email protected] Abstract. Realization of spin crossover (SCO) based applications requires studying of spin state switching characteristics of SCO complex molecules at nanostructured environments especially on-surface. Except for a very few cases, the SCO of a surface bound thin molecular film is either quenched or heavily altered due to (i) strong molecule-surface interactions and (ii) differing intermolecular interactions in films relative to the bulk. By fabricating SCO complexes on a weakly interacting surface such as highly oriented pyrolytic graphite (HOPG) and copper nitride (CuN), the interfacial quenching problem has been tackled. However, engineering intermolecular interactions in thin SCO active films is rather difficult. This work proposes a molecular self-assembly strategy to fabricate thin spin switchable surface bound films with programmable intermolecular interactions. Molecular engineering of the parent complex system [Fe(H2B(pz)2)2(bpy)] (pz = pyrazole, bpy = 2,2'-bipyridine) with a dodecyl (C12) alkyl chain yielded a classical amphiphile- functional and vacuum sublimable charge neutral FeII complex, like [Fe(H2B(pz)2)2(C12-bpy)] (C12-bpy = dodecyl[2,2'-bipyridine]-5-carboxylate). The bulk powder and 10 nm thin film, on quartz glass/SiOx surface, of the complex showed comparable spin state switching characteristics mediated by similar lamellar bilayer like self-assembly/molecular interactions in both bulk and thin film states. This unprecedented observation augurs well for the development of SCO based applications, especially in molecular spintronics. 1 Metal-organic spin crossover (SCO) complexes[1–4] capable of undergoing switching between low-spin (LS) and high-spin (HS) states as a function of temperature, light, pressure, and electric field are suitable candidates to fabricate room temperature operable molecular electronic/spintronics architectures.[5–10] Relentless efforts have been made to study spin state switching characteristics of SCO complexes on different surfaces[11–19] to harness the device utility of SCO entities and several interesting results, e.g., the spin-state dependence of electrical conductance,[6,20] memristance behaviour,[21] electric field or electron induced SCO,[22,23] and locking and unlocking of SCO around room temperature[24] have been reported. Vacuum sublimation of the complexes on to a suitable surface is the preferred method to obtain high-quality surface layers of SCO complexes.[25–30] Despite continuing efforts only a few families of sublimable SCO complexes have been realized, these include charge neutral [Fe(phen)2)(NCS)2] (phen = 1,10-phenanthroline),[31] [Fe(H2B(pz)2)2(L)] (pz = pyrazole, L = 2, 2'–bipyridine (bpy) or 1,10-phenanthroline (phen)),[32,33] and [Fe(HB(pz)3)2][34] systems. Among them, the [Fe(H2B(pz)2)2(L)] complexes have been studied in detail on metallic Au(111),[18,35–37] non-metallic Bi(111),[13] and highly oriented pyrolytic graphite (HOPG) surfaces.[17] The switching is reported to be coverage dependent on metallic Au(111) substrate on which the first layer of the complex undergoes fragmentation and loss of SCO,[38] whereas switching is retained on Bi(111) and HOPG surfaces even at sub-monolayer level due to the weak molecule-surface interfacial interactions.[13,17] While the role of the surface in affecting spin state switching of SCO thin film architectures is well established, control of the intermolecular interactions, a governing factor in dictating SCO in the bulk phase, is a rather difficult problem to tackle at nanostructured environments especially on surface bound thin films due to lack of control associated with the film fabrication processes in engineering intermolecular interactions warranting systematic experimentation in this direction. To this end, a self-organization strategy using tailored intermolecular interactions could be employed. This could be realized by the synthesis of a functional-SCO complex with a propensity to self- assemble via intermolecular interactions, which could be copied "as such" when translating from bulk to thin film state enabling the realization of similar if not identical SCO characteristics between bulk and thin film phases. To start anew in this direction, a functional [Fe(H2B(pz)2)2(C12-bpy)] (C12-bpy = dodecyl [2,2'-bipyridine]-5-carboxylate) SCO complex (cf. Figure 1a), whose structure is reminiscent of classical amphiphile, featuring SCO active pseudo-hydrophilic head and hydrophobic dodecyl (C12) tail groups, capable of self-assembly via intermolecular hydrophobic interactions has been designed and synthesized. Spin state switching characteristics of the [Fe(H2B(pz)2)2(C12-bpy)] complex in bulk and thin film state has been probed to elucidate the role of intermolecular interactions in mediating SCO characteristics upon moving from bulk to thin film environment. Magnetic susceptibility measurement of [Fe(H2B(pz)2)2(C12-bpy)] complex powder showed gradual SCO with T1/2 = 210 K (160 K for the parent complex) without any noticeable hysteresis upon repeated cycling (cf. Figure 1b). The obtained χmT product of 3.55 cm3 K mol- 1 and 0.05 cm3 K mol-1 at 300 K and 100 K, respectively, indicate the presence of pure high and low spin species at those temperatures and genuine temperature-induced SCO behavior. To get insights into the SCO characteristics and self-assembly of the [Fe(H2B(pz)2)2(C12-bpy)] in the thin film state, the complex was sublimed on quartz glass substrate and SiOx surfaces from a 2 crucible held at a temperature of 150 °C, far below the onset of degradation inferred from thermogravimetric analysis (TGA) (cf. Figure S1). Electronic spectral analysis of the thin film sample sublimed on quartz glass showed the ligand centred and metal to ligand charge transfer 1(MLCT) bands analogues to the chloroform solution of the complex (cf. Figure S2a and b) confirming structural integrity of [Fe(H2B(pz)2)2(C12-bpy)] on quartz substrate after sublimation. 4eg 6eg The SCO characteristics of a 10 nm thick film of [Fe(H2B(pz)2)2(C12-bpy)] on a SiOx surface has been probed using X-ray absorption spectroscopy (XAS) at the Fe L2,3 edges. Comparable XAS spectra recorded at room temperature for the 10 nm thick film on SiOx to the powder sample of the complex further confirms the conserved electronic structure of the complex in the film (cf. Figure 2). Upon cooling the bulk powder and the thin film of [Fe(H2B(pz)2)2(C12-bpy)] from 300 K to 100 K, a spectral intensity ratio change of Fe L3 edge 2 configuration (S = 2) and multiplet features at ~ 706.8 eV and ~ 708.1 eV typical for HS t2g 0 (S = 0), respectively, have been observed in line with the previous studies.[18] A LS t2g decrease in the branching ratio,[18,39] defined as b = I3/(I3+I2), Ix = integrated intensity at the corresponding Lx edge, from b = 0.7 at 300 K to b = 0.61 at 100 K, indicative of reduced spin- orbit coupling due to the diamagnetic nature of LS state, further confirms the occurrence of SCO in the thin film sample. Variable temperature XAS measurements have been performed to get insights into the thermal SCO characteristics of bulk powder and 10 nm film of [Fe(H2B(pz)2)2(C12-bpy)]. By using the peak intensity ratios from the HS (LS) reference curve recorded at 300 K (100 K) the HS proportion of the complexes within the probed region at a given temperature has been assessed (cf. Figure S3c). The analysis established T1/2 = 197 K for the bulk sample, which is ca. 13 K lower than the one recorded for the powder sample using SQUID magnetometry and is most probably due to temperature calibration issues. The transition curve of the 10 nm film is clearly more gradual in character with transition temperature shifted by ~20 K towards lower values with respect to the bulk sample, this could be attributed to a weak effect of the interface or to the slightly different intermolecular interactions among the molecules in sublimed film in comparison with the powder form (vide infra).[40] X-ray diffraction (XRD) measurements have been performed to investigate the nature of the self-organization of [Fe(H2B(pz)2)2(C12-bpy)] in the bulk and the thin film states. The powder pattern of the bulk sample is characteristic of a lamellar crystal phase, as it is composed of a series of strong and sharp fundamental and higher-order reflections of the layer sequence (reflections (00l), periodicity d = 25.4 Å), and by some weak and unresolved crossed reflections of the three-dimensional crystalline cell (cf. Figure 3a). This arrangement directly follows the "amphiphile-reminiscent" design of the molecules, as they bind covalently two antagonistic segments, namely the alkyl chains and the metal complex counterpart. These incompatible moieties spontaneously phase separate in juxtaposed domains and most often form multilayered lamellae as in this case.[41–43] A side effect of the intercalation of different nature of layers is the loss of positional correlation between molecular segments of subsequent lamellae. The consequence is some 3 extent of blurring out of the three-dimensional superstructure, as revealed by the lowly resolved crossed reflections. The pattern of the thin film sublimed on quartz glass only displays the (00l) reflections, presumably because the film grew oriented, with the lamellar system parallel to substrate. In this case, all the reflections other than (00l) lie out of the specular plane and thus beyond the measured reciprocal space zone. Remarkably, film and powder patterns exhibit identical lamellar periodicity and same intensity ratios in the reflection series. The lamellar structure of the bulk sample was therefore maintained in the film, which fulfills the experimental realization of an SCO complex maintaining the same phase in bulk powder and thin film states. Some differences might nevertheless exist in the in-plane arrangements and in the related three-dimensional structure, which could explain the gradual and lower T1/2 temperature observed for the thin film of [Fe(H2B(pz)2)2(C12-bpy)] complex. Beyond the information directly extracted from patterns, the common lamellar structure of bulk sample and film can be scrutinized with a geometrical analysis. The basis of this approach is the empirical rule that in nano-segregated systems, the geometrical features of the individual segments are widely independent of the molecular architectures associating them. These features can then be extracted from density and crystallographic data on reference molecules and compared with the molecular area Amol of the layer sequence, defined as the ratio of molecular volume Vmol and lamellar thickness d.[44,45] In this way, the volume contribution of the SCO head VSCO was found to range between 620-660 Å3, on the basis of the single crystal structure of the complex without the dodecyl carboxylate substituent[32] and of reference partial volume data. The volume contribution of the alkyl tail is either of 280-300 Å3 for crystallized chains or ca. 320 Å3 for molten chains. Hence, for the entire molecule, Vmol ranges between 900-1000 Å3, and Amol, between 37-41 Å2. This is about half of the transverse section of the SCO head (σSCO  80 Å2, also determined from the single-crystal data) and about twice the section of the stretched tail (σch = 18.5-20 Å2 if crystallized, and σch  21 Å2, if molten). The compacity of the whole multilayered structure is imposed by the requirement of the bulkiest segments, thus here the SCO heads arranged in double-layers to minimize Amol (cf. Figure 3b). The smaller tails need to spread for filling up the intervals between successive double-layers, which occurs through various easy mechanisms: in the crystallized state, tails generally arrange in strongly tilted double-layers, whereas molten tails rather fold or eventually interdigitate.[46] In any case, the amphiphile-like molecular organization of [Fe(H2B(pz)2)2(C12-bpy)] into double-layers spaced by tails is clearly established from XRD, as well as the identity of the lamellar structure between powder and thin film, and the spontaneous surface alignment of the lamellar system in the film. The morphology of the thin film was further investigated by tapping mode atomic force microscopy (AFM) experiments (cf. Figure 4) performed on 10 nm thick film of [Fe(H2B(pz)2)2(C12-bpy)] on SiOx. Large-scale AFM image on Fig. 4a clearly reveals the formation of flat terraces by the molecules. A relative height of the steps either in range d = 1.1 – 1.3 nm or d = 2.2 – 2.6 nm was found in good agreement with the lamellar spacing deduced from XRD and corresponds to half or entire double-layer structures as deduced from the structural model in Fig. 3b. The above results clearly indicate the role of intermolecular interactions in successfully preserving SCO in nanostructured thin films and signify the role of chemical structure 4 alterations leading to desired material properties. Importantly, functionalization of the parent system with the dodecyl chain also resulted in a 50 K increase in T1/2 value relative to its parent complex clearly elucidating the role of the chemical substituent in tuning SCO in line with the literature reports detailing alkyl chain tuning of SCO characteristics.[47–49] The parent [Fe(H2B(pz)2)2(bpy)] complex was reported to grow on amorphous surfaces in a Volmer-Weber mode by crystallites formation, whereas functionalization of the [Fe(H2B(pz)2)2(bpy)] head with the C12 chain imposed a preferential stacking direction and facilitated vertically ordered growth on the surface beneficial in terms of fabricating multilayer vertical SCO junctions.[50] The alignment of [Fe(H2B(pz)2)2(C12-bpy)] perpendicular to the plane of the surface in thin films clearly indicates the strong intermolecular interactions overcoming molecule-surface interfacial interactions. This enables systematic distance scaling between the SCO head and surface via synthetic modification of alkyl chain length and could be considered as a novel strategy to decouple SCO heads from the surface, via insulating alkyl chains, like adding an insulating layer between the strongly interacting metallic surface and the SCO complex. Note also a recent report by Zhang et al.,[24] detailing quenching of SCO in 5 nm film of [Fe(H2B(pz)2)2(bpy)] on SiO2 and Al2O3; the preservation of SCO in 10 nm films observed in the present work is also important in this context and in terms of realizing large area spintronics junctions based on SCO complexes. To summarize, the present study demonstrated a functionalization strategy of the parent [Fe(H2B(pz)2)2(bpy)] complex system which alters the material properties by modifying the molecular stacking arrangement but preserved its spin transition and sublimation propensity. The functional [Fe(H2B(pz)2)2(C12-bpy)] complex could be sublimed at relatively low temperatures in comparison with its parent counterpart and was found to self-assemble in the same amphiphile-like lamellar structure in the bulk and the thin film state, while effecting near- identical switching characteristics of the complex in both states. To the best of our knowledge, this is the first successful attempt of sublimation of a functional SCO complex with the observation of SCO in a thin film environment. The results presented in this study are expected to open a way for a more molecular-engineering oriented approach in the SCO materials research direction and studying of on-surface switching characteristics of vacuum sublimable functional SCO complexes with strict control of the nanostructure morphology which has strong implications in determining charge carrier mobilities as reported for a poly(3-hexylthiophene) based lamellar self-assembled domains on SiO2/Si substrate.[51] Further, several functional addends featuring advantageous physical properties such as luminescence and chirality could be coupled with the [Fe(H2B(pz)2)2(bpy)] skeleton and the resultant functional-SCO complexes may be sublimed on surfaces which may result in upheaving of SCO based research direction towards realistic SCO based applications. 5 Experimental Section Synthesis and characterization of the ligand and complex were detailed in the supporting information. The 10 nm thick film was sublimed from a molybdenum boat heated up by electric current to 120 oC under pressure of 10-8 mbar using a Plassys deposition machine. The thickness of the film was established based on the quartz balance used during the process and calibrated with x-ray reflectivity performed on a test sample. The XAS measurements were performed at the X-Treme beam line at the Swiss Light Source, Paul Scherrer Institute, Switzerland.[52] The powder sample was pressed into an indium foil. The XAS spectra were recorded at normal X- ray incidence in total electron yield mode and normalized to the integral over the sum of L3 and L2 edges of Fe. Ultraviolet-visible absorption measurements were performed with a Varian Cary 5000 double-beam UV-vis-NIR spectrometer, with referential complex dissolved in chloroform (CHCl3). Magnetic susceptibility measurements were performed on an MPMS-XL5 SQUID magnetometers (Quantum Design) at BDC = 0.1 T field and sweep rate 2 K min-1. Gelatine capsules were used as sample holders in the temperature range 100 ↔ 300 K. The diamagnetic corrections of the molar magnetic susceptibilities were applied using Pascal's constants. The crystalline phase of the powder and the films were determined using a Bruker D8 Advance diffractometer equipped with a LynxEye™ detector in the θ - 2θ mode with a monochromatic wavelength λCuKα1 = 1.54056 Å), 0.2°/min over a 2θ range from 2 to 30° and at room temperature. Atomic force microscopy measurements of the complex sublimed onto SiOx surface were performed with Bruker Ikon microscope in tapping mode at ambient pressure and room temperature. The error bars of the height along the profile were established as a standard deviation of the points in flat regions above and below the step. Acknowledgements M. R. would like to thank the financial support of the Agence Nationale de la Recherche-Labex NIE 11-LABX-0058_NIE within the "investissement d'Avenir" program ANR-10-IDEX-0002- 02 and Deutsche Forschungsgemeinschaft (DFG) through TRR 88 "3MET" (A1, A4, C4, C5, and C6). M. S. and J. D. gratefully acknowledge funding by the Swiss National Science Foundation (Grant no. 200021_165774/1). This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 701647. The authors also thank ANR-11-LABX-0058_NIE within the Investissement d'Avenir program ANR-10-IDEX-0002-02 for the AFM facility at IPCMS. References [1] M. A. Halcrow, Ed. , Spin-Crossover Materials: Properties and Applications, John Wiley & Sons Ltd, Oxford, UK, 2013. [2] K. Senthil, M. Ruben, Coord. Chem. Rev. 2017, DOI 10.1016/j.ccr.2017.03.024. [3] B. Weber, W. Bauer, J. Obel, Angew. Chem. Int. Ed. 2008, 47, 10098. 6 [4] K. Senthil Kumar, I. Šalitroš, B. Heinrich, O. Fuhr, M. Ruben, J Mater Chem C 2015, 3, 11635. [5] E. Coronado, J. R. Galán-Mascarós, M. Monrabal-Capilla, J. García-Martínez, P. Pardo- Ibáñez, Adv. Mater. 2007, 19, 1359. [6] C. Lefter, S. Rat, J. S. Costa, M. D. Manrique-Juárez, C. M. Quintero, L. Salmon, I. Séguy, T. Leichle, L. Nicu, P. Demont, A. Rotaru, G. Molnár, A. Bousseksou, Adv. Mater. 2016, 7508. [7] B. Rösner, M. Milek, A. Witt, B. Gobaut, P. Torelli, R. H. Fink, M. M. Khusniyarov, Angew. Chem. Int. Ed. 2015, 54, 12976. [8] I. Šalitroš, N. T. Madhu, R. Boča, J. Pavlik, M. Ruben, Monatshefte Für Chem. - Chem. Mon. 2009, 140, 695. [9] B. Schäfer, C. Rajnák, I. Šalitroš, O. Fuhr, D. Klar, C. Schmitz-Antoniak, E. Weschke, H. Wende, M. Ruben, Chem. Commun. 2013, 49, 10986. [10] P. Gamez, J. S. Costa, M. Quesada, G. Aromí, Dalton Trans. 2009, 7845. [11] L. Kipgen, M. Bernien, F. Nickel, H. Naggert, A. J. Britton, L. M. Arruda, E. Schierle, E. Weschke, F. Tuczek, W. Kuch, J. Phys. Condens. Matter 2017, DOI 10.1088/1361- 648X/aa7e52. [12] G. Kuang, Q. Zhang, T. Lin, R. Pang, X. Shi, H. Xu, N. Lin, ACS Nano 2017, 11, 6295. [13] S. Ossinger, H. Naggert, L. Kipgen, T. Jasper-Toennies, A. Rai, J. Rudnik, F. Nickel, L. M. Arruda, M. Bernien, W. Kuch, R. Berndt, F. Tuczek, J. Phys. Chem. C 2017, 121, 1210. [14] T. Jasper-Tönnies, M. Gruber, S. Karan, H. Jacob, F. Tuczek, R. Berndt, J. Phys. Chem. Lett. 2017, 1569. [15] T. Groizard, N. Papior, B. Le Guennic, V. Robert, M. Kepenekian, J. Phys. Chem. Lett. 2017, 8, 3415. [16] V. Shalabaeva, S. Rat, M. D. Manrique-Juarez, A.-C. Bas, L. Vendier, L. Salmon, G. Molnár, A. Bousseksou, J Mater Chem C 2017, 5, 4419. [17] M. Bernien, H. Naggert, L. M. Arruda, L. Kipgen, F. Nickel, J. Miguel, C. F. Hermanns, A. Krüger, D. Krüger, E. Schierle, E. Weschke, F. Tuczek, W. Kuch, ACS Nano 2015, 9, 8960. [18] B. Warner, J. C. Oberg, T. G. Gill, F. El Hallak, C. F. Hirjibehedin, M. Serri, S. Heutz, M.-A. A. Arrio, P. Sainctavit, M. Mannini, G. Poneti, R. Sessoli, P. Rosa, J. Phys. Chem. Lett. 2013, 4, 1546. [19] E. Nuin, W. Bauer, A. Hirsch, Eur. J. Org. Chem. 2017, 2017, 790. [20] S. Shi, G. Schmerber, J. Arabski, J.-B. Beaufrand, D. J. Kim, S. Boukari, M. Bowen, N. T. Kemp, N. Viart, G. Rogez, E. Beaurepaire, H. Aubriet, J. Petersen, C. Becker, D. Ruch, Appl. Phys. Lett. 2009, 95, 043303. [21] T. Miyamachi, M. Gruber, V. Davesne, M. Bowen, S. Boukari, L. Joly, F. Scheurer, G. Rogez, T. K. Yamada, P. Ohresser, E. Beaurepaire, W. Wulfhekel, Nat. Commun. 2012, 3, 938. [22] T. G. Gopakumar, F. Matino, H. Naggert, A. Bannwarth, F. Tuczek, R. Berndt, Angew. Chem. Int. Ed. 2012, 51, 6262. 7 [23] V. Meded, A. Bagrets, K. Fink, R. Chandrasekar, M. Ruben, F. Evers, A. Bernand- Mantel, J. S. Seldenthuis, A. Beukman, H. S. J. van der Zant, Phys. Rev. B 2011, 83, 245415. [24] X. Zhang, P. S. Costa, J. Hooper, D. P. Miller, A. T. N'Diaye, S. Beniwal, X. Jiang, Y. Yin, P. Rosa, L. Routaboul, M. Gonidec, L. Poggini, P. Braunstein, B. Doudin, X. Xu, A. Enders, E. Zurek, P. A. Dowben, Adv. Mater. 2017, 1702257. [25] T. Palamarciuc, J. C. Oberg, F. El Hallak, C. F. Hirjibehedin, M. Serri, S. Heutz, J.-F. Létard, P. Rosa, J. Mater. Chem. 2012, 22, 9690. [26] H. J. Shepherd, G. Molnár, W. Nicolazzi, L. Salmon, A. Bousseksou, Eur. J. Inorg. Chem. 2013, 2013, 653. [27] A. Pronschinske, R. C. Bruce, G. Lewis, Y. Chen, A. Calzolari, M. Buongiorno-Nardelli, D. A. Shultz, W. You, D. B. Dougherty, Chem. Commun. 2013, 49, 10446. [28] M. Gruber, V. Davesne, M. Bowen, S. Boukari, E. Beaurepaire, W. Wulfhekel, T. Miyamachi, Phys. Rev. B 2014, 89, 195415. [29] M. Gruber, T. Miyamachi, V. Davesne, M. Bowen, S. Boukari, W. Wulfhekel, M. Alouani, E. Beaurepaire, J. Chem. Phys. 2017, 146, 092312. [30] T. Mahfoud, G. Molnár, S. Cobo, L. Salmon, C. Thibault, C. Vieu, P. Demont, A. Bousseksou, Appl. Phys. Lett. 2011, 99, 053307. [31] E. Koenig, K. Madeja, Inorg. Chem. 1967, 6, 48. [32] J. A. Real, M. C. Muñoz, J. Faus, X. Solans, Inorg. Chem. 1997, 36, 3008. [33] N. Moliner, L. Salmon, L. Capes, M. C. Muñoz, J.-F. Létard, A. Bousseksou, J.-P. Tuchagues, J. J. McGarvey, A. C. Dennis, M. Castro, R. Burriel, J. A. Real, J. Phys. Chem. B 2002, 106, 4276. [34] L. Salmon, G. Molnár, S. Cobo, P. Oulié, M. Etienne, T. Mahfoud, P. Demont, A. Eguchi, H. Watanabe, K. Tanaka, A. Bousseksou, New J. Chem. 2009, 33, 1283. [35] X. Zhang, T. Palamarciuc, P. Rosa, J.-F. Létard, B. Doudin, Z. Zhang, J. Wang, P. A. Dowben, J. Phys. Chem. C 2012, 116, 23291. [36] A. Pronschinske, Y. Chen, G. F. Lewis, D. A. Shultz, A. Calzolari, M. Buongiorno Nardelli, D. B. Dougherty, Nano Lett. 2013, 13, 130327091617008. [37] S. Beniwal, X. Zhang, S. Mu, A. Naim, P. Rosa, G. Chastanet, J.-F. Létard, J. Liu, G. E. Sterbinsky, D. A. Arena, P. A. Dowben, A. Enders, J. Phys. Condens. Matter 2016, 28, 206002. [38] T. G. Gopakumar, M. Bernien, H. Naggert, F. Matino, C. F. Hermanns, A. Bannwarth, S. Mühlenberend, A. Krüger, D. Krüger, F. Nickel, W. Walter, R. Berndt, W. Kuch, F. Tuczek, Chem. - Eur. J. 2013, 19, 15702. [39] B. T. Thole, G. van der Laan, Phys. Rev. B 1988, 38, 3158. [40] V. Davesne, M. Gruber, M. Studniarek, W. H. Doh, S. Zafeiratos, L. Joly, F. Sirotti, M. G. Silly, a. B. Gaspar, J. a. Real, G. Schmerber, M. Bowen, W. Weber, S. Boukari, V. Da Costa, J. Arabski, W. Wulfhekel, E. Beaurepaire, J. Chem. Phys. 2015, 142, 194702. [41] J.-M. Rueff, N. Masciocchi, P. Rabu, A. Sironi, A. Skoulios, Chem. - Eur. J. 2002, 8, 1813. [42] V. M. Coiro, F. Mazza, G. Pochetti, Acta Crystallogr. C 1986, 42, 991. [43] T. R. Lomer, K. Perera, Acta Crystallogr. B 1974, 30, 2912. [44] F. Mathevet, P. Masson, J.-F. Nicoud, A. Skoulios, J. Am. Chem. Soc. 2005, 127, 9053. 8 [45] S. Marzouk, A. Khalfallah, B. Heinrich, J. E. Khiari, A. Kriaa, S. Méry, J. Fluor. Chem. 2017, 197, 15. [46] J. Fouchet, L. Douce, B. Heinrich, R. Welter, A. Louati, Beilstein J. Org. Chem. 2009, 5, DOI 10.3762/bjoc.5.51. [47] H. L. C. Feltham, C. Johnson, A. B. S. Elliott, K. C. Gordon, M. Albrecht, S. Brooker, Inorg. Chem. 2015, 54, 2902. [48] C. J. Johnson, G. G. Morgan, M. Albrecht, J Mater Chem C 2015, 3, 7883. [49] W. Zhang, F. Zhao, T. Liu, M. Yuan, Z.-M. Wang, S. Gao, Inorg. Chem. 2007, 46, 2541. [50] S. Cobo, G. Molnár, J. A. Real, A. Bousseksou, Angew. Chem. 2006, 118, 5918. [51] H. Sirringhaus, P. J. Brown, R. H. Friend, M. M. Nielsen, K. Bechgaard, B. M. W. Langeveld-Voss, A. J. H. Spiering, R. A. J. Janssen, E. W. Meijer, P. Herwig, D. M. de Leeuw, Nature 1999, 401, 685. [52] C. Piamonteze, U. Flechsig, S. Rusponi, J. Dreiser, J. Heidler, M. Schmidt, R. Wetter, M. Calvi, T. Schmidt, H. Pruchova, J. Krempasky, C. Quitmann, H. Brune, F. Nolting, J. Synchrotron Radiat. 2012, 19, 661. 9 the pseudo-amphiphile Figure 1. (a) Molecular structure of like SCO complex [Fe(H2B(pz)2)2(C12-bpy)] and (b) χmT vs. T plot of bulk powder form of [Fe(H2B(pz)2)2(C12- bpy)], inset shows d(χmT)/dT vs. T curve indicating T1/2 = 210 K. 10 Figure 2. XAS at the Fe L2,3 edges of [Fe(H2B(pz)2)2(C12-bpy)]. XAS spectra acquired for powder (top) and sublimed 10 nm thick film on SiOx (bottom) at 300 K and 100 K. The vertical dotted lines are guides for eyes indicating multiplet features at ~ 706.8 eV and ~ 708.1 eV characteristic for HS and LS state molecules, respectively, of bulk and thin film samples. The curves were recorded in total electron yield mode at normal X-ray incidence and normalized to the sum of integrals over the Fe L3 and L2 edges. 11 70070571071572072573010 nm thick film on SiOx HS (300 K) LS (100 K)TEY (a.u.)Fe L2,3 edgesH = 0.1 TpowderTEY (a.u.)Photon energy (eV) Figure 3. Self-organization of a [Fe(H2B(pz)2)2(C12-bpy)] thin film. (a) X-ray diffraction pattern of a 10 nm thick sublimed [Fe(H2B(pz)2)2(C12-bpy)] film on a quartz substrate compared to the powder reference. The equidistant reflections (001) – (004) reveal the lamellar structure of 2.54 nm periodicity in powder and film. The broad feature at 17 – 25° is the scattering signal from the quartz glass substrate. (b) Schematic view of the internal structure of lamellae of [Fe(H2B(pz)2)2(C12-bpy)] consisting of double-layers of the complex heads (blue spheres) alternating with aliphatic tails arranged in double layers, according to the deduced bilayer crystalline arrangement. 12 51015202530(004)(003)(002)Fe(bpz)2C12-bpy: film on quartz powder Intensity (a. u.)2 (deg.)(001)a) Figure 4. Surface morphology of a sublimed [Fe(H2B(pz)2)2(C12-bpy)] film on SiOx substrate. (a) Surface topography measurement performed by room temperature ambient pressure AFM in tapping mode of Si/SiOx(400 nm)/[Fe(H2B(pz)2)2(C12-bpy)](10 nm) highlighting bilayers of the lamellar molecular stacking. (b) Height profile of the step along the profile 1 shown in (a) of height d = 2.47 ± 0.07 nm consistent with lamellar spacing extracted from XRD. 13 Supporting Information Synthesis of [Fe(H2B(pz)2)2(C12-bpy)] Scheme S1. Synthesis of [Fe(H2B(pz)2)2(C12-bpy)]. Key: (a) DCC/DMAP, DCM, RT and (b) Fe(ClO4)2.6H2O, dihydro(bispyrazolyl)borate and C12-bpy, MeOH/DCM. Compound 1 was synthesized according to the literature procedure.[1] Synthesis of C12-bpy An oven dried 50 mL Schlenk flask with a stir bar was charged with [2, 2'] Bipyridinyl- 5-carboxylic acid (1) (0.29 g, 1.34 mmol) and 1-dodecanol (0.26 g, 1.34 mmol), and the solids were placed under argon atmosphere. To this, anhydrous CH2Cl2 (20 mL) was added and the mixture cooled to 0 °C in an Ice-Water bath. DCC (0.276 g, 1.34 mmol) and DMAP (0.163 g, 1.34 mmol) were then added to the stirring mixture as solids; the reaction mixture was slowly allowed to attain RT and stirred for 48 h followed by filtration to remove dicyclohexyl urea. Purification was accomplished by silica gel column chromatography using CH2Cl2 as an eluent. Yield: 0.185 g (37%). 1H NMR (300 MHz, CDCl3, 300 K, TMS): δ = 9.1 (s, 1H), 8.5 (d, 8.1 Hz, 2H), 8.4 (dd, 8.4 Hz, 1H), 8.3 (d, 9.3 Hz, 1H), 8.2 (m, 4H), 8 (m, 3H), 7.9 (m, 2H), 7.8 (m, 1H), 4.5 (t, 6.6 Hz, 2H), 3.4 (t, 7.2 Hz, 2H) and 2.1 (m, 4H). 13C NMR (75 MHz, CDCl3, 300 K, TMS): 28, 28.6, 33, 65, 120, 122, 123, 123.3, 124.3, 124.6, 124.8, 124.9, 125, 125.1, 126, 127.2, 127.3, 127.5, 128.6, 130, 131, 131.4, 136, 137, 134, 149, 150, 155, 159 and 165.3. Elemental Analysis: Calc. For: (C23H32N2O2): C, 74.96; H, 8.75; N, 7.6; Found: C, 74.87; H, 8.72; N, 7.6. Synthesis of [Fe(H2B(pz)2)2(C12-bpy)] To a solution of iron (II) perchlorate hydrate (0.073 g, 0.2 mmol) in 5 ml of methanol, potassium dihydro(bispyrazolyl)borate (0.075 g, 0.4 mmol) was added and the mixture was stirred for 15 mins followed by filtration to remove the precipitated KClO4. A solution of C12-bpy (0.074 g, 0.2 mmol) dissolved in 1 ml of dichloromethane solvent was added dropwise to the filtrate leading to the formation of greenish precipitate upon continuous stirring for 4 h under argon 14 protection. The precipitate was filtered, washed with methanol and dried in a vacuum oven for 6 h at 50 °C to yield the title complex as a pale green powder. Yield: 56 mg (38%). Elemental Analysis of the pale green powder: Calc. for [Fe(H2B(pz)2)2(C12-bpy)]: (C36H48B2FeN10O2.1H2O) C, 57.78; H, 6.73; N, 18.72; Found: C, 57.37; H, 6.57; N, 18.9. Sublimation of [Fe(H2B(pz)2)2(C12-bpy)] To verify the sublimability of the [Fe(H2B(pz)2)2(C12-bpy)] thermogravimetric analysis (TGA) was performed at an ambient pressure and argon atmosphere using an SDT Q600 apparatus from TA Instruments at a scanning rate of 5 °C min−1. The result presented in Fig. S1 implies the onset of the sublimation at temperature Tsublim ~ 150 oC, what was later used for fabrication of a 10 nm thick sample discussed in the main text. Figure S1. TGA of [Fe(H2B(pz)2)2(C12-bpy)]. The measurement was performed at an ambient pressure and argon atmosphere. Ultraviolet-visible spectroscopy To confirm the presence of [Fe(H2B(pz)2)2(C12-bpy)] molecules and its structural integrity on the quartz substrate after sublimation, ultraviolet-visible (UV-Vis) spectroscopy measurements were performed employing Varian Cary 5000 double-beam UV-vis-NIR spectrometer. A comparison between absorption spectrum of the [Fe(H2B(pz)2)2(C12-bpy)] complex in solution (cf. Fig. S2a) and thin complex film on quartz substrate revealed similar spectral features (cf. Fig. S2b) indicating the structural integrity of the complex in the sublimed film. 15 501001502002503003504004500.20.40.60.81.01.21.41.61.8Tsublim ~ 150 oCTemperature (°C) Weight (mg)-100-80-60-40-20020Heat Flow (V) Figure S2. UV-Vis absorption and transmission spectra of [Fe(H2B(pz)2)2(C12-bpy)] (a) in chloroform solution and (b) as a film (specify the thickness) on quartz (Qz) glass substrate. The peaks present at ca. 632 nm and 651 nm for the complex in solution and thin film state clearly indicate the HS state of the complex at ambient temperature in line with the experimental data presented in this manuscript. The bathochromic shift in the absorption maxima upon moving from solution to thin film state is tentatively attributed to strong intermolecular interactions between the complex entities. Variable temperature X-ray absorption measurements Variable temperature x-ray absorption measurements at Fe L2,3 edges (Fig. S3) were performed to get insights into the SCO characteristics of the complex in powder and thin film state. The microcrystalline complex powder which also served as a reference, enabling comparison between bulk and thin film state, was pressed into indium foil. The spectra were recorded in total electron yield mode in an applied magnetic field of 0.1 T to facilitate the extraction of the electrons, at normal incidence and with linear x-ray polarization in the sample plane. The XAS spectra at Fe L3 edge of the thin film sample showed a single intersection point for all the curves at ~ 707.6 eV resulting from good uniformity of the film (Fig. S3b). On the other hand, deviation of the curves at this photon energy observed for the powder sample (Fig. S3a) could be attributed to the changes in the amount of probed matter due to the slight displacement of the x-ray beam position on the sample while varying the temperature. 16 Figure S3. Variable temperature XAS at the Fe L2,3 edges of (a) polycrystalline powder of [Fe(H2B(pz)2)2(C12-bpy)] complex, (b) 10 nm [Fe(H2B(pz)2)2(C12-bpy)] film on SiOx, and (c) plot of the high spin fraction (γHS) vs. temperature for the powder and thin film samples of [Fe(H2B(pz)2)2(C12-bpy)] indicating nearly complete spin state switching of both powder and thin film samples with approximately 20 K reduction of the T1/2 value for the thin film sample. From the pre-edge normalized spectra we extracted high spin proportion for a given temperature based on the spectral intensity ratio of the features at ~ 706.8 eV and ~ 708 eV typical for HS t2g 0 (S = 0), respectively, according to the formula: 4eg 2 configuration (S = 2) and LS t2g 6eg where γHS is the high spin proportion, f(L3a) and f(L3b) are spectral intensities of the Fe L3 edge feature at ~ 706.8 eV and ~ 708.1 eV (see Fig. 2 dashed vertical lines), respectively, and the indices x, HS and LS correspond to a mixed state spectrum, pure high spin (300 K) and pure low spin (100 K) references. The resulting curves of γHS as a function of temperature for the film and powder sample are presented in Fig. S3. 17 LS3bHSHS3bHSLS3aHSHS3HS3b3a)()1()()()1()()()(LfLfLfLfLfLfaxx References [1] C. A. Panetta, H. J. Kumpaty, N. E. Heimer, M. C. Leavy, C. L. Hussey, J. Org. Chem., 1999, 64 (3), 1015–1021. 18
1902.05973	1	1902	2019-02-15T19:32:19	Triplet-sensitization by lead halide perovskite thin films for near-infrared-to-visible upconversion	[ "physics.app-ph" ]	Lead halide-based perovskite thin films have attracted great attention due to the explosive increase in perovskite solar cell efficiencies. The same optoelectronic properties that make perovskites ideal absorber materials in solar cells are also beneficial in other light-harvesting applications and make them prime candidates as triplet sensitizers in upconversion via triplet-triplet annihilation in rubrene. In this contribution, we take advantage of long carrier lifetimes and carrier diffusion lengths in perovskite thin films, their high absorption cross sections throughout the visible spectrum, as well as the strong spin-orbit coupling owing to the abundance of heavy atoms to sensitize the upconverter rubrene. Employing bulk perovskite thin films as the absorber layer and spin-mixer in inorganic/organic heterojunction upconversion devices allows us to forego the additional tunneling barrier owing from the passivating ligands required for colloidal sensitizers. Our bilayer device exhibits an upconversion efficiency in excess of 3% under 785 nm illumination.	physics.app-ph	physics	Triplet-sensitization by lead halide perovskite thin films for near-infrared-to-visible upconversion Lea Nienhaus,‡,&,* Juan-Pablo Correa-Baena,† Sarah Wieghold,†,& Markus Einzinger,¶ Ting-An Lin,¶ Katherine E. Shulenberger,‡ Nathan D. Klein,‡ Mengfei Wu,¶ Vladimir Bulović,¶ Tonio Buonassisi,† Marc A. Baldo,¶,* and Moungi G. Bawendi‡,* ‡Department of Chemistry, ¶Department of Electrical Engineering and Computer Science, †Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139 corresponding authors: [email protected], [email protected] and [email protected] ABSTRACT Lead halide-based perovskite thin films have attracted great attention due to the explosive increase in perovskite solar cell efficiencies. The same optoelectronic properties that make perovskites ideal absorber materials in solar cells are also beneficial in other light-harvesting applications and make them prime candidates as triplet sensitizers in upconversion via triplet- triplet annihilation in rubrene. In this contribution, we take advantage of long carrier lifetimes and carrier diffusion lengths in perovskite thin films, their high absorption cross sections throughout the visible spectrum, as well as the strong spin-orbit coupling owing to the abundance of heavy atoms to sensitize the upconverter rubrene. Employing bulk perovskite thin films as the absorber layer and spin-mixer in inorganic/organic heterojunction upconversion devices allows us to forego the additional tunneling barrier owing from the passivating ligands required for colloidal sensitizers. Our bilayer device exhibits an upconversion efficiency in excess of 3% under 785 nm illumination. Keywords: lead halide perovskites, upconversion, free carriers, triplet excitons, charge-separated state, triplet-charge annihilation, triplet-triplet annihilation Photon upconversion describes the generation of high-energy photons by combining two or more low-energy photons. This conversion of low-energy photons to high-energy photons provides a means to overcome the Shockley-Queisser limit in single-junction solar cells1 by sub-bandgap sensitization of the absorber layer,2 -- 4 and shows promise in near-infrared (NIR) sensitization of low-cost silicon-based devices for applications such as biological imaging,5,6 3-dimensional displays7 and photocatalysis.6 In this work, we focus on upconversion via diffusion-mediated triplet-triplet annihilation (TTA) in organic semiconductors.8,9 Unlike other upconversion processes such as such as second harmonic generation in nonlinear crystals or upconversion using lanthanides ions,6 TTA can be effective even at sub-solar photon fluxes since the energy is stored in long-lived triplet states.10,11 In the TTA process investigated here, two triplet states on neighboring annihilator molecules (rubrene) generate a higher-lying emissive singlet state by electronic coupling. Direct optical excitation into a spin-triplet state would violate the selection rules, therefore these are "spin- forbidden" and unable to radiatively couple to the ground state. We require sensitizers to convert optically excited spin-singlet states to optically inactive triplet states in rubrene, which are then transferred to the triplet state of rubrene via a spin-allowed exchange-mediated (Dexter)12 energy transfer process. Historically, sensitization has been achieved by using metal-organic complexes containing heavy metal atoms, which facilitate intersystem crossing (ISC) due to strong spin-orbit coupling. However, this approach has the disadvantage of large exchange energies, which can result in energy losses up to 300 meV.10,13,14 More recently, we showed lead sulfide (PbS) nanocrystals (NCs) to be efficient triplet sensitizers.11,15 -- 19 This system has the major advantage of exchange energies on the order of kT (~25meV), resulting in negligible energy loss during the singlet-triplet conversion. Strong spin-orbit coupling in PbS NCs renders the spin quantum number meaningless, and only the total angular momentum quantum number accurately describes the states. As a result, the spin of the optically excited exciton rapidly dephases and its wave function shows both singlet and triplet character.20,21 However, current PbS NC based upconversion devices exhibit poor exciton diffusion lengths which leads to inefficient exciton transport within the PbS NC films. As a result, the PbS NC layer thickness is limited to one or two monolayers, causing these devices to suffer from low NIR absorption.17,19 To circumvent the issue of poor exciton transport within large-scale NC arrays, we seek to replace the "bulk" NC array with a true bulk semiconductor material which shows long carrier diffusion lengths and also can undergo efficient energy transfer to rubrene. Lead halide perovskite thin films have shown great promise as absorber layers in photovoltaics. Commonly used in perovskite solar cells (PSCs), this technology has made impressive progress in just a few years with efficiencies climbing from 3.8%22 in 2009 to a certified 22.7%23 in 2017. The general formula of the commonly used perovskite is ABX3, containing an organic cation A, such as methylammonium (MA) or formamidinium (FA),22,24 a divalent metal B, such as lead or tin,25 and a halogen X, such as bromine or iodine. The impressive performances achieved to date have been attributed to exceptional material properties such as low non-radiative recombination,26 low exciton binding energies,27 -- 29 high light absorption over the visible spectrum, and charge carrier diffusion lengths in the micrometer range due to long carrier lifetimes.30 -- 32 All of these characteristics allow for efficient extraction of charges at the interface of the electron- and hole-transport layers in photovoltaic devices. Unlike the PbS NCs investigated previously,15,16,19 which exhibit long-lived excitonic (bound electron-hole pairs) behavior, the initially-excited bound excitons in MA lead triiodide (MAPbI3) perovskite films commonly dissociate within picoseconds into unbound and highly mobile charge carriers.33 Due to the abundance of heavy atoms (Pb, I) in lead halide perovskites, strong spin- orbit coupling is expected in the perovskite film and the electron and hole can both rapidly spin- mix.34,35 Hence, injection of free electrons and holes into the upconverting organic semiconductor can provide a new avenue for sensitization of rubrene, and may allow us to move away from the necessity of efficient singlet-to-triplet exciton converters.36 Here, we investigate the upconversion in rubrene sensitized by MA0.15FA0.85PbI3 thin films. The rubrene layer is doped with 1% dibenzotetraphenylperiflanthene (DBP). This host- guest/annihilator-emitter approach is adopted from organic light-emitting diodes and has been shown to enhance the quantum yield (QY) of the rubrene film, as singlets generated by the upconversion process in rubrene are harvested via Förster resonance energy transfer (FRET) to the emitter DBP, outcompeting singlet fission.15 Figure 1: AFM height profile of the MA0.15FA0.85PbI3 thin film. The inset shows the AFM image of the perovskite film, with a grain size distribution of 50 ± 20 nm. Atomic force microscopy (AFM) is used to determine the morphology and thickness of the perovskite film. Figure 1 shows a line profile across a boundary of the perovskite film and the underlying substrate yielding a perovskite film thickness of ~14 nm. The inset shows the morphology of the perovskite film, highlighting the small grain size of ~50 nm (compare also SI Figure S1 & S2). This grain size has previously been shown to be large enough to have a true "bulk" behavior of the excited charge carriers and avoid any quantum confinement effects which may change the underlying carrier dynamics.37 Quantum confinement effects are known to increase the bandgap of semiconductors. We further confirm the lack of energetic confinement and thus, the bulk-like behavior of our MA0.15FA0.85PbI3 thin films by steady-state photoluminescence (PL) spectroscopy (Figure 2a, brown), where we see the peak emission at the expected value of ~1.55 eV (800 nm) for bulk MA0.15FA0.85PbI3.38 The absorbance of the neat MA0.15FA0.85PbI3 thin film (black line) and of the bilayer upconversion device (gray), as well as the emission at 610 nm of the upconverted light (red) are shown in Figure 2a. The inset shows the upconverted light created in the upconversion device under 785 nm illumination, observed as the orange line running horizontally through the device. Figure 2b depicts the upconversion device structure investigated here consisting of the 15 nm thin film of MA0.15FA0.85PbI3, with a subsequent layer of rubrene/1% DBP spin coated from a 10 mg/ml stock solution. In Figure 2c the band diagrams of the MA0.15FA0.85PbI3 perovskite and rubrene are shown.16,38 Two triplet sensitization mechanisms are possible: i) direct triplet exciton transfer (TET) of a bound triplet exciton as reported previously for quantum confined perovskite nanocrystals39 or as in the case of PbS sensitized upconversion15,16 and ii) sequential electron and hole transfer resulting in a bound triplet on the rubrene molecule. For sensitization via direct TET from the initially created band-edge excitons to occur at the interface, the combined rate of exciton diffusion to the interface and TET must outcompete the exciton dissociation rate. However, as discussed previously, lead halide perovskites exhibit binding energies below kT and the initially optically excited band-edge excitons generally rapidly dissociate within picoseconds. Hence, TET from these band-edge excitons is unlikely to be the dominant mechanism of triplet sensitization at room temperature by the bulk perovskite investigated here. However, shallow traps may result in localized "long-lived" sub-bandgap excitons which could to transfer to the rubrene layer within their lifetime.40 Figure 2: (a) Normalized absorbance spectra of a neat MA0.15FA0.85PbI3 perovskite film (black), MA0.15FA0.85PbI3/rubrene/1%DBP bilayer device (gray), the upconverted emission stemming from rubrene/DBP (red) and the NIR emission of the MA0.15FA0.85PbI3 perovskite (brown). The inset shows the device under 785 nm illumination, and the upconverted light is seen as the orange line. (b) Schematic of the bilayer upconversion device, which emits 610 nm light upon 785 nm excitation. (c) Band diagram of MA0.15FA0.85PbI3 and rubrene, showing the energetically favored hole transport from the perovskite VB to the rubrene HOMO. The electron cannot directly transfer from the perovskite CB to the rubrene LUMO due to an energy barrier of ~1eV. The energetic alignment indicates that hole transfer from the perovskite valence band (VB) to the rubrene highest occupied molecular orbital (HOMO) can be very efficient due to the driving energy of ~0.4 eV. An unbound electron on the other hand, cannot transfer directly into the lowest unoccupied molecular orbital (LUMO) of rubrene due to the ~1 eV energy gap between the perovskite conduction band (CB) and the rubrene LUMO, despite significant band bending reported by Ji et al.41 However, with a hole residing on the rubrene molecule, it may be possible to transfer the electron directly into the bound triplet state (T1) through a mediating charge- transfer state (CT) at the interface. To further investigate the upconversion process, as well as determine the timescale of triplet sensitization, we turn to time-resolved transient PL spectroscopy. Figure 3a shows the time- resolved NIR PL of a neat MA0.15FA0.85PbI3 film (black trace), and of the bilayer upconversion device (red trace). Both decay traces show highly multiexponential dynamics, indicating a broad distribution of emission rates. To avoid overfitting, we define the PL lifetime of the neat MA0.15FA0.85PbI3 as the time in which the population decays to 1/e (t1/e = 50 ns). The very long tail (t1 = 8 µs) is attributed to delayed fluorescence stemming from trapped carriers.42 Interestingly, we observe that the overall emission intensity of the MA0.15FA0.85PbI3 perovskite film is not notably quenched as previously seen in PbS NC-based devices,15,16 despite seeing a decrease in the PL lifetime. This residual NIR PL intensity can likely be attributed to carriers that radiatively recombine in the "bulk" of the perovskite film prior to reaching the extraction point at the MA0.15FA0.85PbI3/rubrene interface. In addition, the organic molecules can passivate dangling bonds at the surface of the perovskite lattice. This reduces non-radiative recombination occurring at surface defects in the MA0.15FA0.85PbI3 perovskite film, thus increasing the overall PLQY.43,44 To extract the dynamics of energy transfer process at the MA0.15FA0.85PbI3/rubrene interface, we apply our previously introduced extraction method,15,16 where we subtract the residual NIR PL dynamics of the neat MA0.15FA0.85PbI3 film from the dynamics of the bilayer device. This method relies on the basis of an active and an inactive population. The active population couples strongly to the rubrene and is quenched by the added upconverting layer, while the inactive population does not interact with the rubrene. This can be due to the morphology of the rubrene layer (e.g. pinholes in the rubrene film), or be a result of carriers which radiatively recombine within the thin perovskite film prior to reaching the inorganic/organic interface. The transfer kinetics obtained by this method are seen in Figure 3a as green dots. We find a characteristic decay time of t2 = 15 ± 5 ns (dark blue), highlighted in the inset in Figure 3a, which shows a magnification of the early time dynamics. The observed subsequent rise and slow decay in the extracted transfer dynamics (green dots) at later times can be attributed to back transfer of singlet excitons created after diffusion-mediated TTA.16 The extracted transfer time is slower than reported previously for quenching by hole- or electron- acceptors30,45 and may indicate the presence of a long-lived state which mediates the energy transfer process as observed previously in triplet sensitization by PbS NCs46,47 or could stem from an energy barrier for the electron transfer process due to the reported band bending at the perovskite/rubrene interface.41 To further investigate the upconversion mechanism, we investigate the dynamics of the upconverted emission, which correlate to a convolution of the rate of energy transfer, diffusion- mediated TTA in rubrene and the emission from DBP following FRET. Figure 3b highlights the transient PL spectroscopy performed on a bilayer device when monitoring the DBP emission. The inset shows an enlargement of the PL dynamics at a short time scale. We observe that the long-lived triplets (ttriplet > 12 µs) have not fully decayed at 25 µs, which results in a buildup of triplet excitons in the rubrene film that do not decay between pulses. An increase in the triplet population has been shown to increase the rate of TTA, which results in an apparent reduction in the extracted characteristic time for TTA, as well as the extracted triplet lifetime.15,16 The extracted rise time trise = 24 ± 3 ns of the upconverted emission therefore portrays only a lower bound for the rate of the total upconversion process. Interestingly, the peak emission does not occur until t = 3 µs, and a second slower rise is observed leading up to this time. We attribute this to additional delayed TTA caused by slow triplet diffusion and by delayed triplet generation by thermally detrapped carriers, which can occur in addition to the prompt transfer of free carriers. Figure 3: (a) Time-resolved PL dynamics of the NIR emission from neat MA0.15FA0.85PbI3 (black), the quenched NIR dynamics of MA0.15FA0.85PbI3 in the bilayer device (red) under 785 nm excitation. The trapped carrier lifetime (t1 = 8 µs) is extracted from the long-time dynamics (light blue). The extracted dynamics (green) yield an extracted quenching rate of t2 = 15 ± 5 ns. (b) Dynamics of the upconverted light emitted from the bilayer device. The rise time gives an upper bound for the triplet sensitization rate (trise = 24 ± 3 ns). The insets show an enlargement of the early-time dynamics. However, as both bound triplet excitons and free carriers (holes) can be present in the rubrene layer, triplets may be annihilated not only by TTA, but also through triplet-charge annihilation (TCA). In the TCA process, a bound triplet exciton collides with an adjacent single charge and either scatters it, or the collision dissociates the bound triplet exciton into free charges. Therefore, this latter mechanism creates more freely diffusing charges in the film which can undergo TCA, effectively reducing the achievable upconverted PL.35,48 To further elucidate roles of TCA and TTA we investigate the magnetic field effect (MFE) of the upconverted light, as well as the power dependence of the emission, both of which exhibit features unique to TTA. The theory of MFEs in molecular crystals was first developed by Merrifield in 1968.49,50 When two triplet excitons collide, one of nine possible triplet-triplet pair states is formed. Only those with singlet character can effectively couple to the singlet state. The external magnetic field can modify the number of states with singlet character. At zero field only three of the nine states have singlet character, and as the field is increased, additional states acquire singlet character, thus increasing the rate of TTA. At high fields, the singlet character is localized on only two pair states, and therefore the yield of TTA is reduced, resulting in the negative MFE typical of TTA. Figure 4a shows the MFEUC of the upconverted emission from a MA0.15FA0.85PbI3/rubrene/1%DBP bilayer device, which shows the typical signature of TTA (red hexagons). The MFEPL of a neat MA0.15FA0.85PbI3 film under 532 nm excitation is shown as black squares, and despite reports of negative MFEs (up to 2.5% on photocurrent at 300mT) in similar perovskite thin films35,51 we are not able to unambiguously detect an effect by the applied magnetic field. Furthermore, the power dependency of the MA0.15FA0.85PbI3 PL (~800 nm) and of the upconverted emission of a MA0.15FA0.85PbI3/rubrene/1%DBP bilayer device, under 785 nm pulsed excitation at 10 MHz is highlighted in the log-log plot in Figure 4b (compare also to SI Figure S3). The MA0.15FA0.85PbI3 PL (black squares) shows a slightly super-linear power dependency, with a slope of 1.1 (gray line) indicating the emission primarily stems from monomolecular recombination in the perovskite thin film, consistent with previous reports of shallow-trap-assisted recombination.29,52,53 The upconverted emission (red hexagons) shows the transition of the slope from quadratic (slope = 2) to linear (slope = 1), which is typical of bimolecular TTA.54 In the weak annihilation regime (lower powers) triplets decay primarily via quasi first-order kinetics and the upconverted PL intensity is quadratic with excitation power (white background). At higher powers (strong annihilation regime), the triplets decay primarily through bimolecular TTA, resulting in a linear dependence of the upconversion efficiency and the excitation power (green background). As a result, the upconversion efficiency is dependent on the excitation power and the maximal upconversion efficiency is obtained in the linear regime. In the MA0.15FA0.85PbI3/rubrene/1%DBP bilayer devices this quadratic-to-linear transition occurs at 500 mW/cm2 (dashed green line) average incident power under pulsed excitation, nearly an order of magnitude lower than the incident power required in previous PbS NC-based upconversion devices,15 and two-fold less than in interference enhanced upconversion devices.15 Finally, we investigate the upconversion efficiency (𝜂"#) of the bilayer device. 𝜂"# is defined as (𝜂$%) and the TTA efficiency of the annihilator (𝜂%%&). On the other hand, the external quantum the fraction of excited carriers in the sensitizer that are converted to singlet excitons in the annihilator,6 which is the product of the energy transfer efficiency from sensitizer to annihilator efficiency (EQE) is the ratio of number of upconverted visible photons to number of incident infrared photons. In our bilayer device, this is further equivalent to the product of absorption of MA0.15FA0.85PbI3 at 785nm excitation (𝐴𝑏𝑠+,-.), the upconversion efficiency (𝜂"#/2), and the PLQY of rubrene:1 wt.% DBP film (𝜂12:456) at visible excitation, as shown in equation (1): Note that 𝜂"# is divided by 2 to normalize to a maximum efficiency of 100%. Both the EQE of the bilayer device under 785nm excitation and 𝜂12:456 at 450nm excitation are measured with an integrating sphere,55 giving an EQE of (1.5 ± 0.3) ×10-4 % and 𝜂12:456 of 9.3 ± 0.8 %. 𝐸𝑄𝐸=𝐴𝑏𝑠+,-.∙(<=>?)∙𝜂12:456 (1) However, due to the very low absorption of MA0.15FA0.85PbI3 thin films at the excitation wavelength of 785 nm, and parasitic interference effects, it is difficult to accurately measure the percentage of absorbed photons (compare also SI Figure S4). We estimate the upper bound of the absorption to be 0.1%, which yields a conservative estimate for the internal upconversion efficiency of 𝜂"# = 3.1%. In conclusion, we have been able to show that MA0.15FA0.85PbI3 thin films can function as efficient triplet sensitizers of the annihilator rubrene. The presented results show that TTA is the prevalent process observed in the MA0.15FA0.85PbI3/rubrene/1%DBP bilayer device, with no detectable detrimental effect caused by TCA. The extracted triplet sensitization rate of t = 15 ns is unusually slow for an electron transfer process and may point to a long-lived intermediate localized bound excitonic state mediating the energy transfer process, or an energy barrier for the electron transfer at the interface stemming from the reported band bending and will be subject to further investigation. Optimization of the energy transfer process, the energetic band alignment in the upconversion device, as well as the device structure provide a means to further improve the device performance. Figure 4: (a) Change in the upconverted emission (red hexagons) and the NIR MA0.15FA0.85PbI3 emission (black squares) under an applied magnetic field. (b) Power dependency of the NIR emission of the MA0.15FA0.85PbI3 perovskite in a bilayer device, showing a slightly super-linear trend (black squares). The upconverted emission (red hexagons) shows the quadratic-to-linear slope change typical of TTA, with an average incident power density threshold of 500 mW/cm2. ASSOCIATED CONTENT Supporting Information The Supporting Information is available free of charge at DOI: Experimental section, additional AFM data (Figure S1 & S2), upconverted PL dynamics during the power dependency (Figure S3), and the absorbance spectra of the bilayer device showing the parasitic absorption effects (Figure S4). AUTHOR INFORMATION Corresponding Author Lea Nienhaus: [email protected] Moungi G. Bawendi: [email protected] Marc A. Baldo: [email protected] Present address &Department of Chemistry and Biochemistry, Florida State University, Tallahassee, FL 32306 ORCID Lea Nienhaus: 0000-0003-1412-412X Juan-Pablo Correa-Baena: 0000-0002-3860-1149 Sarah Wieghold: 0000-0001-6169-3961 Marc A. Baldo: 0000-0003-2201-5257 Moungi G. Bawendi: 0000-0003-2220-4365 Notes The authors declare no competing financial interests. ACKNOWLEDGEMENT This work was supported as part of the Center for Excitonics, an Energy Frontier Research Center funded by the US Department of Energy, Office of Science, Office of Basic Energy Sciences under Award Number DE-SC0001088 (MIT). J.P.C.B. is supported by a postdoctoral fellowship awarded by the U.S. Department of Energy, Office of Science, Office of Energy Efficiency and Renewable Energy. S.W. was supported by the National Science Foundation (NSF) under Grant NSF CA No. EEC-1041895. N.D.K. was supported by the National Science Foundation Graduate Research Fellowship under Grant No. 1122374 and funded by the US Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under Award No. DE-FG02-07ER46454. This work was supported in part at the Center for Nanoscale Systems (CNS), a member of the National Nanotechnology Coordinated Infrastructure Network (NNCI), which is supported by the National Science Foundation under NSF Award No. 1541959. CNS is part of Harvard University. REFERENCES Shockley, W.; Queisser, H. J. Detailed Balance Limit of Efficiency of P-n Junction Solar Cells. J. (1) Appl. Phys. 1961, 32 (3), 510 -- 519. (2) Trupke, T.; Green, M. A.; Würfel, P. Improving Solar Cell Efficiencies by Up-Conversion of Sub- Band-Gap Light. J. Appl. Phys. 2002, 92 (7), 4117 -- 4122. (3) Meng, F.-L.; Wu, J.-J.; Zhao, E.-F.; Zheng, Y.-Z.; Huang, M.-L.; Dai, L.-M.; Tao, X.; Chen, J.-F. High-Efficiency near-Infrared Enabled Planar Perovskite Solar Cells by Embedding Upconversion Nanocrystals. Nanoscale 2017, 9 (46), 18535 -- 18545. (4) He, M.; Pang, X.; Liu, X.; Jiang, B.; He, Y.; Snaith, H.; Lin, Z. Monodisperse Dual-Functional Upconversion Nanoparticles Enabled Near-Infrared Organolead Halide Perovskite Solar Cells. Angew. Chem. 2016, 128 (13), 4352 -- 4356. (5) Wang, F.; Banerjee, D.; Liu, Y.; Chen, X.; Liu, X. Upconversion Nanoparticles in Biological Labeling, Imaging, and Therapy. Analyst 2010, 135 (8), 1839 -- 1854. (6) Zhou, J.; Liu, Q.; Feng, W.; Sun, Y.; Li, F. Upconversion Luminescent Materials: Advances and Applications. Chem. Rev. 2015, 115 (1), 395 -- 465. (7) Deng, R.; Qin, F.; Chen, R.; Huang, W.; Hong, M.; Liu, X. Temporal Full-Colour Tuning through Non-Steady-State Upconversion. Nat. Nanotechnol. 2015, 10 (3), 237. Schmidt, T. W.; Castellano, F. N. Photochemical Upconversion: The Primacy of Kinetics. J. Phys. Chem. Lett. 2014, 5 (22), 4062 -- 4072. Schulze, T. F.; Schmidt, T. W. Photochemical Upconversion: Present Status and Prospects for Its Application to Solar Energy Conversion. Energy Environ. Sci. 2014, 8 (1), 103 -- 125. (8) (9) (10) Singh-Rachford, T. N.; Castellano, F. N. Photon Upconversion Based on Sensitized Triplet -- Triplet Annihilation. Coord. Chem. Rev. 2010, 254 (21), 2560 -- 2573. (11) Mahboub, M.; Huang, Z.; Tang, M. L. Efficient Infrared-to-Visible Upconversion with Subsolar Irradiance. Nano Lett. 2016, 16 (11), 7169 -- 7175. (12) Dexter, D. L. A Theory of Sensitized Luminescence in Solids. J. Chem. Phys. 1953, 21, 836 -- 850. (13) Monguzzi, A.; Tubino, R.; Meinardi, F. Multicomponent Polymeric Film for Red to Green Low Power Sensitized Up-Conversion. J. Phys. Chem. A 2009, 113 (7), 1171 -- 1174. (14) Amemori, S.; Sasaki, Y.; Yanai, N.; Kimizuka, N. Near-Infrared-to-Visible Photon Upconversion Sensitized by a Metal Complex with Spin-Forbidden yet Strong S 0 -- T 1 Absorption. J. Am. Chem. Soc. 2016, 138 (28), 8702 -- 8705. (15) Wu, M.; Congreve, D. N.; Wilson, M. W. B.; Jean, J.; Geva, N.; Welborn, M.; Van Voorhis, T.; Bulović, V.; Bawendi, M. G.; Baldo, M. A. Solid-State Infrared-to-Visible Upconversion Sensitized by Colloidal Nanocrystals. Nat. Photonics 2016, 10 (1), 31 -- 34. (16) Nienhaus, L.; Wu, M.; Geva, N.; Shepherd, J. J.; Wilson, M. W. B.; Bulović, V.; Van Voorhis, T.; Baldo, M. A.; Bawendi, M. G. Speed Limit for Triplet-Exciton Transfer in Solid-State PbS Nanocrystal-Sensitized Photon Upconversion. ACS Nano 2017, 11 (8), 7848 -- 7857. (17) Nienhaus, L.; Wu, M.; Bulovic, V.; Baldo, M. A.; Bawendi, M. G. Using Lead Chalcogenide Nanocrystals as Spin Mixers: A Perspective on near-Infrared-to-Visible Upconversion. Dalton Trans 2018, 47, 8509 -- 8516. (18) Huang, Z.; Li, X.; Mahboub, M.; Hanson, K. M.; Nichols, V. M.; Le, H.; Tang, M. L.; Bardeen, C. J. Hybrid Molecule -- Nanocrystal Photon Upconversion Across the Visible and Near-Infrared. Nano Lett. 2015, 15 (8), 5552 -- 5557. (19) Wu, M.; Jean, J.; Bulović, V.; Baldo, M. A. Interference-Enhanced Infrared-to-Visible Upconversion in Solid-State Thin Films Sensitized by Colloidal Nanocrystals. Appl. Phys. Lett. 2017, 110 (21), 211101. (20) Scholes, G. D. LONG-RANGE RESONANCE ENERGY TRANSFER IN MOLECULAR SYSTEMS. Ann. Rev. Phys. Chem. 2003, 54, 57 -- 87. (21) Kim, J.; Wong, C. Y.; Scholes, G. D. Exciton Fine Structure and Spin Relaxation in Semiconductor Colloidal Quantum Dots. Acc. Chem. Res. 2009, 42 (8), 1037 -- 1046. (22) Kojima, A.; Teshima, K.; Shirai, Y.; Miyasaka, T. Organometal Halide Perovskites as Visible- Light Sensitizers for Photovoltaic Cells. J. Am. Chem. Soc. 2009, 131 (17), 6050 -- 6051. (23) Green, M. A.; Hishikawa, Y.; Dunlop, E. D.; Levi, D. H.; Hohl-Ebinger, J.; Ho-Baillie, A. W. Y. Solar Cell Efficiency Tables (Version 51). Prog. Photovolt. Res. Appl. 2017, 26 (1), 3 -- 12. (24) Pellet, N.; Gao, P.; Gregori, G.; Yang, T.-Y.; Nazeeruddin, M. K.; Maier, J.; Grätzel, M. Mixed- Organic-Cation Perovskite Photovoltaics for Enhanced Solar-Light Harvesting. Angew. Chem. Int. Ed. 2014, 53 (12), 3151 -- 3157. (25) Stoumpos, C. C.; Malliakas, C. D.; Kanatzidis, M. G. Semiconducting Tin and Lead Iodide Perovskites with Organic Cations: Phase Transitions, High Mobilities, and Near-Infrared Photoluminescent Properties. Inorg. Chem. 2013, 52 (15), 9019 -- 9038. (26) Tress, W. Perovskite Solar Cells on the Way to Their Radiative Efficiency Limit -- Insights Into a Success Story of High Open-Circuit Voltage and Low Recombination. Adv. Energy Mater. 2017, 7 (14), 1602358. (27) Galkowski, K.; Mitioglu, A.; Miyata, A.; Plochocka, P.; Portugall, O.; Eperon, G. E.; Wang, J. T.- W.; Stergiopoulos, T.; Stranks, S. D.; Snaith, H. J.; et al. Determination of the Exciton Binding Energy and Effective Masses for Methylammonium and Formamidinium Lead Tri-Halide Perovskite Semiconductors. Energy Environ. Sci. 2016, 9 (3), 962 -- 970. (28) Yang, Y.; Yang, M.; Li, Z.; Crisp, R.; Zhu, K.; Beard, M. C. Comparison of Recombination Dynamics in CH 3 NH 3 PbBr 3 and CH 3 NH 3 PbI 3 Perovskite Films: Influence of Exciton Binding Energy. J. Phys. Chem. Lett. 2015, 6 (23), 4688 -- 4692. (29) Yamada, Y.; Nakamura, T.; Endo, M.; Wakamiya, A.; Kanemitsu, Y. Photocarrier Recombination Dynamics in Perovskite CH3NH3PbI3 for Solar Cell Applications J. Am. Chem. Soc. 2014, 136, 11610-11613. (30) Stranks, S. D.; Eperon, G. E.; Grancini, G.; Menelaou, C.; Alcocer, M. J. P.; Leijtens, T.; Herz, L. M.; Petrozza, A.; Snaith, H. J. Electron-Hole Diffusion Lengths Exceeding 1 Micrometer in an Organometal Trihalide Perovskite Absorber. Science 2013, 342 (6156), 341 -- 344. (31) Xing, G.; Mathews, N.; Sun, S.; Lim, S. S.; Lam, Y. M.; Grätzel, M.; Mhaisalkar, S.; Sum, T. C. Long-Range Balanced Electron- and Hole-Transport Lengths in Organic-Inorganic CH3NH3PbI3. Science 2013, 342 (6156), 344. (32) Tress, W.; Marinova, N.; Inganäs, O.; Nazeeruddin, M. K.; Zakeeruddin, S. M.; Graetzel, M. Predicting the Open-Circuit Voltage of CH3NH3PbI3 Perovskite Solar Cells Using Electroluminescence and Photovoltaic Quantum Efficiency Spectra: The Role of Radiative and Non-Radiative Recombination. Adv. Energy Mater. 2014, 5 (3), 1400812. (33) Ponseca, C. S.; Savenije, T. J.; Abdellah, M.; Zheng, K.; Yartsev, A.; Pascher, T.; Harlang, T.; Chabera, P.; Pullerits, T.; Stepanov, A.; et al. Organometal Halide Perovskite Solar Cell Materials Rationalized: Ultrafast Charge Generation, High and Microsecond-Long Balanced Mobilities, and Slow Recombination. J. Am. Chem. Soc. 2014, 136 (14), 5189 -- 5192. (34) Even, J.; Pedesseau, L.; Jancu, J.-M.; Katan, C. Importance of Spin -- Orbit Coupling in Hybrid Organic/Inorganic Perovskites for Photovoltaic Applications. J. Phys. Chem. Lett. 2013, 4 (17), 2999 -- 3005. (35) Lu, K.; Zhao, C.; Luan, L.; Duan, J.; Xie, Y.; Shao, M.; Hu, B. Exploring the Role of Spin-Triplets and Trap States in Photovoltaic Processes of Perovskite Solar Cells. J. Mater. Chem. C 2018, 6 (18), 5055 -- 5062. (36) Kafle, T. R.; Kattel, B.; Lane, S. D.; Wang, T.; Zhao, H.; Chan, W.-L. Charge Transfer Exciton and Spin Flipping at Organic -- Transition-Metal Dichalcogenide Interfaces. ACS Nano 2017, 11 (10), 10184 -- 10192. (37) Anaya, M.; Rubino, A.; Rojas, T. C.; Galisteo-López, J. F.; Calvo, M. E.; Míguez, H. Strong Quantum Confinement and Fast Photoemission Activation in CH3NH3PbI3 Perovskite Nanocrystals Grown within Periodically Mesostructured Films. Adv. Opt. Mater. 2017, 5 (8), 1601087. (38) Correa-Baena, J. P.; Steier, L.; Tress, W.; Saliba, M.; Neutzner, S.; Matsui, T.; Giordano, F.; Jacobsson, T. J.; Srimath Kandada, A. R.; Zakeeruddin, S. M.; et al. Highly Efficient Planar Perovskite Solar Cells through Band Alignment Engineering. Energy Environ. Sci. 2015, 8 (10), 2928 -- 2934. (39) Mase, K.; Okumura, K.; Yanai, N.; Kimizuka, N. Triplet Sensitization by Perovskite Nanocrystals J. Chem. Phys. 1968, 48 (9), 4318 -- 4319. for Photon Upconversion. Chem. Commun. 2017, 53 (59), 8261 -- 8264. (40) He, H.; Yu, Q.; Li, H.; Li, J.; Si, J.; Jin, Y.; Wang, N.; Wang, J.; He, J.; Wang, X.; et al. Exciton Localization in Solution-Processed Organolead Trihalide Perovskites. Nat. Commun. 2016, 7, 10896. (41) Ji, G.; Zheng, G.; Zhao, B.; Song, F.; Zhang, X.; Shen, K.; Yang, Y.; Xiong, Y.; Gao, X.; Cao, L.; et al. Interfacial Electronic Structures Revealed at the Rubrene/CH3NH3PbI3 Interface. Phys. Chem. Chem. Phys. 2017, 19 (9), 6546 -- 6553. (42) Chirvony, V. S.; González-Carrero, S.; Suárez, I.; Galian, R. E.; Sessolo, M.; Bolink, H. J.; Martínez-Pastor, J. P.; Pérez-Prieto, J. Delayed Luminescence in Lead Halide Perovskite Nanocrystals. J. Phys. Chem. C 2017, 121 (24), 13381 -- 13390. (43) Luo, D.; Yang, W.; Wang, Z.; Sadhanala, A.; Hu, Q.; Su, R.; Shivanna, R.; Trindade, G. F.; Watts, J. F.; Xu, Z.; et al. Enhanced Photovoltage for Inverted Planar Heterojunction Perovskite Solar Cells. Science 2018, 360 (6396), 1442. (44) Yavari, M.; Mazloum-Ardakani, M.; Gholipour, S.; Tavakoli, M. M.; Taghavinia, N.; Hagfeldt, A.; Tress, W. Reducing Surface Recombination by a Poly(4-Vinylpyridine) Interlayer in Perovskite Solar Cells with High Open-Circuit Voltage and Efficiency. ACS Omega 2018, 3 (5), 5038 -- 5043. (45) Westbrook, R. J. E.; Sanchez-Molina, D. I.; Manuel Marin-Beloqui, D. J.; Bronstein, D. H.; Haque, D. S. A. Effect of Interfacial Energetics on Charge Transfer from Lead Halide Perovskite to Organic Hole Conductors. J. Phys. Chem. C 2018, 122 (2), 1326 -- 1332. (46) Bender, J. A.; Raulerson, E. K.; Li, X.; Goldzak, T.; Xia, P.; Van Voorhis, T.; Tang, M. L.; Roberts, S. T. Surface States Mediate Triplet Energy Transfer in Nanocrystal -- Acene Composite Systems. J. Am. Chem. Soc. 2018, 140 (24), 7543 -- 7553. (47) Garakyaraghi, S.; Mongin, C.; Granger, D. B.; Anthony, J. E.; Castellano, F. N. Delayed Molecular Triplet Generation from Energized Lead Sulfide Quantum Dots. J. Phys. Chem. Lett. 2017, 8 (7), 1458 -- 1463. (48) Shao, M.; Yan, L.; Li, M.; Ilia, I.; Hu, B. Triplet -- Charge Annihilation versus Triplet -- Triplet Annihilation in Organic Semiconductors. J. Mater. Chem. C 2013, 1 (7), 1330 -- 1336. (49) Merrifield, R. E. Theory of Magnetic Field Effects on the Mutual Annihilation of Triplet Excitons. (50) Merrifield, R. E.; Avakian, P.; Groff, R. P. Fission of Singlet Excitons into Pairs of Triplet Excitons in Tetracene Crystals. Chem. Phys. Lett. 1969, 3 (3), 155 -- 157. (51) Zhang, C.; Sun, D.; Sheng, C.-X.; Zhai, Y. X.; Mielczarek, K.; Zakhidov, A.; Vardeny, Z. V. Magnetic Field Effects in Hybrid Perovskite Devices. Nat. Phys. 2015, 11, 427. (52) Shi, Z.-F.; Sun, X.-G.; Wu, D.; Xu, T.-T.; Tian, Y.-T.; Zhang, Y.-T.; Li, X.-J.; Du, G.-T. Near- Infrared Random Lasing Realized in a Perovskite CH3NH3PbI3 Thin Film. J. Mater. Chem. C 2016, 4 (36), 8373 -- 8379. Large-Grain Organic -- Inorganic Perovskite Photovoltaic Efficiencies. Adv. Funct. Mater. 2016, 26 (53) Blancon, J.-C.; Nie, W.; Neukirch, A. J.; Gupta, G.; Tretiak, S.; Cognet, L.; Mohite, A. D.; Crochet, J. J. The Effects of Electronic Impurities and Electron -- Hole Recombination Dynamics on TOC Graphic (24), 4283 -- 4292. (54) Cheng, Y. Y.; Khoury, T.; Clady, R. G. C. R.; Tayebjee, M. J. Y.; Ekins-Daukes, N. J.; Crossley, M. J.; Schmidt, T. W. On the Efficiency Limit of Triplet -- Triplet Annihilation for Photochemical Upconversion. Phys. Chem. Chem. Phys. 2010, 12 (1), 66 -- 71. (55) Mello, J. C. de; Wittmann, H. F.; Friend, R. H. An Improved Experimental Determination of External Photoluminescence Quantum Efficiency. Adv. Mater. 1997, 9.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.